Home Creators Posts Import Register Favorites Logout

Downloads

Content

SUPIR upscales and restores images better than Magnific, for example a 3x upscale from 1024 to 3072 px with face restore, and now supports up to 4 LoRAs.

1-click installers for Windows, RunPod, Massed Compute and Linux plus a free Kaggle notebook, with Torch 2.8, CUDA 12.9 and libraries compiled for RTX 1000 to 5000 series and A100, H100 and B200 GPUs. Includes batch processing, custom model paths and a resumable, hash-checked model downloader.

1 Click Windows, Kaggle, Massed Compute and RunPod / Linux Installer: Image Upscaling and Enhancing Open Source SUPIR. Better than Magnific

Patreon exclusive posts index to find our scripts easily, Patreon scripts updates history to see which updates arrived to which scripts and amazing Patreon special generative scripts list that you can use in any of your task.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI  GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

======

Latest Zip File : SUPIR_v101.zip

  • SUPIR: Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild 1 click installer scripts.

  • SUPIR Sampler and Text CFG Comparison : https://imgsli.com/MjU2ODQz/2/1

  • Gemini 2.5 Pro prompt to get image description for free :

  • Use Default preset for highest loyalty and Replicate preset for adding more details

  • Human upscale from 1024x1024 to 3072x3072 (3x upscale and total 9x resolution) with face restore comparison

  • Owl upscale from 1024x1024 to 3072x3072 (3x upscale and total 9x resolution)

9 November 2025 Update V101

  • Once the model downloads starts, you can close cmd and use Windows_Model_Download_Min.bat to avoid downloading LLaVA model and unnecassary other base SDXL models

  • LLaVA model is used for auto captioning

21 September 2025 Update V100

  • Now you can give custom models path and loras path

  • For custom models path (SDXL models) add this args --ckpt_dir

    • e.g. : --ckpt_dir "C:\temp\b"

  • For custom loras (SDXL Loras) path give this args --loras_dir

    • e.g. : --lora_dir "C:\temp\c"

  • To add this args edit Windows_Start_SUPIR_app.bat and add there

    • You will also see an example case that has REM at the beginning

    • REM means it is commented - not executed

  • You can also set outputs folder ass --outputs_folder

    • e.g. : --outputs_folder "C:\temp\gg"

  • Just execute Windows_Install_or_Update.bat file to update

  • LoRA support implemented

  • You can use up to 4 LoRAs with individual LoRA strength / scale

  • Put LoRAs into Lora folder inside models folder

  • Just execute Windows_Install_or_Update.bat file to update

11 September 2025 Update V97

  • Kaggle notebook updated and now working

  • Batch processing will show which file is being processed on info panel now

  • New prefix and suffix feature implemented

    • Now you can set custom prefix or suffix for your saved images

    • It will work both for single image processing and also batch image processing

  • Libraries upgraded to Torch 2.8, CUDA 12.9, xFormers 0.0.33, Flash Attention 2.8.3

  • You don't need to have CUDA or anything else installed and it should work with Python 3.10.11 and Git installed

  • When compiling libraries, I added support for all GPUs starting from RTX 1000 to 5000 series including other GPUs like A100, H100, B200, L40, etc

  • Downloader script is 100% improved and now it is 100x more robust even for slow connection internets

    • It has 100% continue feature + hash check feature now

  • To upgrade libraries, extract latest zip file and overwrite files

  • Enter inside SUPIR folder and delete venv folder

  • Then run Windows_Install_or_Update.bat file

  • Installation process is also now way faster for Windows, RunPod and Massed Compute

  • To fresh install extract zip file into drive like E:\SUPIR_v91

  • Just execute Windows_Install_or_Update.bat file

  • Then once installation finished use Windows_Start_SUPIR_app.bat file

  • If you have RTX 2000 or 1000 series still should work but if you get error please report me and try If_Regular_Install_Fails_On_RTX_2000_or_Below_GPUs.bat

  • If your download of models fails for any reason, you can always run Windows_Fix_Model_Download.bat

    • It will resume and also hash check downloaded files

Windows Requirements

2 September 2025 Update V87

  • With V87 full screen image preview bug fixed

  • For update run Windows_Update_Version.bat

30 August 2025 Update V86

  • We have added the new requested features

  • Feature 1 : Skip already upscaled images in target folder option when doing batch processing

    • This is simply resume

  • Feature 2 : In Face options Only Face Restoration

    • This will apply SUPIR upscale to only face and rest of the image will be deterministic Lanczos upscale (simply exactly same when you downscale back to original resolution)

  • Feature 3 : Now when doing both batch and single processing, it will show extra screen

    • Actually this was added for batch processing but due to Gradio limitations, single processing will display too

  • This screen will be displayed at left top

  • Fix : Cancel button should work perfectly fine for Single and Batch processing now

  • With V86 from V85, messages displayed at the top panel fixed + improved

  • For update, use latest zip file, extract and overwrite, and run Windows_Update_Version.bat

27 August 2025 Update V80

  • New Feature : First Downscale to Match Target Upscale Size

  • When enabled, the image is first downscaled using high-quality Lanczos algorithm to match the target resolution (after max megapixel/resolution constraints), then upscaled by the specified factor. This can help achieve more consistent results when working with size-constrained outputs.

  • So lets say your input image is 2000x2000 = 4 mega pixels

    • You want to sharpen and increase quality with 4x upscale but that would make it 16 mega pixel which is quite not possible

    • So set Max Megapixels = 4

    • Set Upscale Size = 4

    • And enable First Downscale to Match Target Upscale Size checkbox

    • So it will first downscale image into 500x500 and then it will upscale into 2000x2000

  • To update just use latest zip file and run Windows_Update_Version.bat

1 June 2025 Update V74

  • Zip file updated - now installs latest official xFormers

  • Windows_Resume_Fix_Model_Downloads.bat updated to be more robust

  • Delete venv folder and run installer again if you get any errors

23 May 2025 Update V73

  • New option Skip LLaVA if .txt caption exists added to LLaVA section

  • If you uncheck this, it will re-generate a caption for the image each time when you run it

  • Just run Windows_Update_Version.bat file it should work

  • RunPod and Massed Compute installers upgraded to Torch 2.7 and CUDA 12.8 so now supports RTX 5000 series as well including older GPUs like 4000 series, 3000 series, RTX A6000, H100, A100 etc

    • Linux users can use RunPod or Massed Compute installers

17 April 2025 Update V69

  • Image MetaData saving and loading back improved and some bugs fixed

  • Presets / Configs saving and loading back improved and some bugs fixed

  • 2 Extremely important new features added

    • Max Megapixels (0 = no limit) : Lets say you set 10x upscale and maximum 3 MegaPixels, then it will upscale it until reaches exactly 3 Megapixels resolution

    • Max Resolution (0 = no limit) : Lets say you set 10x upscale and maximum Resolution as 1536, then it will upscale it until the bigger dimension of the image reaches 1536px

  • Previously Estimated Output Resolution was displaying inaccurate resolutions and now it should display 100%

    • Estimated Output Resolution should also accurately display what you will get when using Max Megapixels and Max Resolution

  • Cancel button will now cancel the generation immediately - this bug fixed

  • There were some errors with loading presets fixed

  • Now when you save a preset, it will be auto selected and loaded back

    • Your last saved and loaded presets will be remembered now when you restart the application

    • Selecting a preset from dropdown list will auto load

  • In some cases face restore and then combining restored face was not working

    • Now it should work in all cases, please report me if fails in any case and this really improves quality of faces significantly

  • The slider and where the image cut off was mis-aligned this issue fixed

  • Now supports avif image format as well

  • Run Windows_Update_Requirements.bat to update

  • A new feature Compare Images tab implemented

    • It will list last 20 generations where you can pick and compare or you can upload images and compare

  • Kaggle notebook update and fixed

  • Now it will download models like 8 times faster also

  • New older GPUs install added

  • If you have RTX 2000 series and below GPUs please use Windows_Install_Older_GPUs_RTX_2000_And_Below.bat file

  • Now supports RTX 5000 series as well (includes RTX 3000, 4000 too)

Older Update Logs

https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/App_Updates_Change_Logs/SUPIR_App.md

How To Install

  • Download latest zip file from the attachments and extract files wherever you want to install

  • Double click windows_install .bat and let it install and download necessary models

  • Requires 12 GB VRAM without LLaVA. Tested on RTX 3060

  • Make sure that your VRAM usage is not bigger than 500 MB before starting the SUPIR

  • If you want to use LLaVA you can use our better installer : https://www.patreon.com/posts/90744385

  • LLaVA is not mandatory as shown in full tutorial video : https://youtu.be/PqREA6-bC3w

Massed Compute (Recommend Cloud) :

RunPod (Cloud):

Extra Features

  • 1 click install including automatically downloading necessary models

  • Generate any number of images each time with different seed 

  • So you can generate 200 different images with different seeds at 1 click

  • Every generated image saved in outputs folder automatically

  • Display progress via progress info of Gradio including each image generation speed

  • This info displayed at the CMD / terminal as well

  • Improved more useful Gradio interface

  • Randomize seed feature

  • Includes pip freeze info for future library fixes

  • Auto installers for both Windows & RunPod (Any Linux)

  • Supports VRAM optimizations and now works at 12 GB GPUs

  • Tested on RTX 3060

  • Make sure that your VRAM usage is not bigger than 500 MB before starting the SUPIR

How To Use

  • The biggest difference making thing is under SUPIR options Text Guidance Scale

    • This will impact how much add new details or stay loyal to the original image

  • Next big thing is Sampler

    • EDM is more loyal to the original image but DPMPP2M can add more details

  • S-Churn and S-Noise doesn't make much valuable difference but you can test

  • In the Advanced Options

    • Linear CFG and Linear Stage2 Guidance can make difference so you can try them

  • All generated images will be automatically saved inside outputs folder

  • If you also want to use LLaVA we have better LLaVA scripts which supports 4-bit, 8-bit and 16-bit loading 

  • But even with simple captions it works great you can just type

  • Image comparison sliders to Test : https://imgsli.com/ , https://web-toolbox.dev/en/tools/image-compare-slider , https://www.diffchecker.com/image-compare/

  • Here example. All default except Linear Stage2 Guidance is enabled and set 0.5

  • input image : input_1.png

Basic Caption : a trex dinosaur in jurassic park

Output : basic_caption.png

Better caption : A gigantic dinosaur with sharp teeth is standing in a lush green landscape with mountains in the background. The sky is partly cloudy, and the dinosaur appears to be in motion, possibly running or lunging forward. The setting seems to be reminiscent of prehistoric or Jurassic environments, likely intended to represent the natural habitat of such a creature.

Output : better_caption.png

Camel

A family photo

Related posts

All posts in this topic: Image and Video Upscaling and Restoration · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

Comments

AI Squad

Any plans of having a Colab or Kaggle version of it?

SECourses

it is impossible. it requires 30 GB VRAM at the moment :)

SECourses

Thank you. I am working on improvements right now.

Sid

GeekZolda

Man that sounds and looks amazing. Shame it's essentially unusable in it's current state due to the VRAM requirement.

puk

I became an error on start the test.py file under windows: 国 C:\Windows\py.exe X + Traceback (most recent call last): File "C: Users \AI\SUPIR \SUPIR\test.py", Line 3, in ‹module> from SUPIR.util import create_SUPIR_model, PIL2Tensor, Tensor2PIL, convert File "C: \Users\AI\SUPIR\SUPIR\SUPIR\util.py", line 7, in ‹module> from omegaconf import OmegaConf ModuleNotFoundError: No module named 'omegaconf'

SECourses

hello welcome. this installer developed for gradio py. I just added Gradio starter file for windows. please download V6 and run start_SUPIR_Windows.bat

Neil Rhodes

installing now! WIll let you know how it goes!

puk

New install with Version 7 flowoing errors: Enter your choice (1-2): 1 stabl usage: gradio_demo.py [-h] [--ip IP] [--share SHARE] [--port PORT] [--no_llava] [--use_image_slider] [--log_history] gradio_demo.py: error: unrecognized arguments: --loading_half_params --use_tile_vae Windows 11 RTX4090

Neil Rhodes

first test Error A:\SUPIR_v7\SUPIR\venv\lib\site-packages\torch\nn\functional.py:5476: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at ..\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:263.) attn_output = scaled_dot_product_attention(q, k, v, attn_mask, dropout_p, is_causal)

SECourses

hello. please do a reinstall. delete older folder and install again. those arguments added with newest update. if you know how git works, you can do git pull, activate venv and install diffusers. after it should work

Samuel

Can you control level of hallucinations?

SECourses

This model designed to prevent hallucinations. I don't know for sure yet though but it almost has 0 hallucinating

Neil Rhodes

how long should it take as this is still running at 308 seconds with zero per cent progress

SECourses

Follow cmd. If your input image is high resolution it could be using shared vram. That reduces speed a lot. Also did you check vram usage before starting the app?

puk

I did a new Install, now i have the following error: │ │ 84 │ │ 85 │ │ 86 def load_state_dict(checkpoint_path: str, map_location='cpu'): │ │ ❱ 87 │ checkpoint = torch.load(checkpoint_path, map_location=map_location) │ │ 88 │ if isinstance(checkpoint, dict) and 'state_dict' in checkpoint: │ │ 89 │ │ state_dict = checkpoint['state_dict'] │ │ 90 │ else: │ │ │ │ C:\Users\AI-PUK\SUPIR\SUPIR\venv\lib\site-packages\torch\serialization.py:1005 in load │ │ │ │ 1002 │ │ │ # reset back to the original position. │ │ 1003 │ │ │ orig_position = opened_file.tell() │ │ 1004 │ │ │ overall_storage = None │ │ ❱ 1005 │ │ │ with _open_zipfile_reader(opened_file) as opened_zipfile: │ │ 1006 │ │ │ │ if _is_torchscript_zip(opened_zipfile): │ │ 1007 │ │ │ │ │ warnings.warn("'torch.load' received a zip file that looks like a To │ │ 1008 │ │ │ │ │ │ │ │ " dispatching to 'torch.jit.load' (call 'torch.jit.loa │ │ │ │ C:\Users\AI-PUK\SUPIR\SUPIR\venv\lib\site-packages\torch\serialization.py:457 in __init__ │ │ │ │ 454 │ │ 455 class _open_zipfile_reader(_opener): │ │ 456 │ def __init__(self, name_or_buffer) -> None: │ │ ❱ 457 │ │ super().__init__(torch._C.PyTorchFileReader(name_or_buffer)) │ │ 458 │ │ 459 │ │ 460 class _open_zipfile_writer_file(_opener): │ ╰──────────────────────────────────────────────────────────────────────────────────────────────────╯ RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

AI Squad

WTF?! haha... Is there any demo available online? Do you know?

Neil Rhodes

1024 x 1024 image trying to change to 2048x2048 fresh reboot nothing running except v7 cpu5950x ram:64GB VRam 12GB 3080

daniel mendoza

excellent, thanks for your work. Hey, I had the idea of ​​using another model other than the base and it works even better, for the test I used the "juggernaut v9 sdxl" model

SECourses

Hello. We just updated to V7. now works even with 12 GB. so you can locally install and use enjoy :) I plan to make kaggle once they fix FP16 bug.

SECourses

Hello. With V7 now works even at 12 GB GPUs. you can use on windows with your GPU now. enjoy

SECourses

Hello. With V7 now works even at 12 GB GPUs. you can use on windows with your GPU now. enjoy

SECourses

I think download of some files were failed for some reason. can you try this? make a folder inside your C:/test_supir and make a new install there. and let me know

SECourses

Ok I think your VRAM not being sufficient for that resolution. Can you try 768x768. So probably it used shared VRAM which caused huge slow down. 768x768 still should yield a good upscale with 1x upscale. try and let me know please. I am also looking for even further improvements of VRAM

SECourses

wow i haven't tested this. nice

George Gostyshev

First of all - nice release. Second - is there a way to update without redownloading all models? Also have a strange behaviour - 1x upscale works fine but 2x longs for ages without any ooms - it's just computing and computing and never ending. Using 4090 card

SECourses

hello. for v7 to v8, just replace old yaml file with new one and download RunDiffusion/Juggernaut-XL-v9 into models folder you can also edit downloader.py file and remove the models you already have that taking forever means that system started using shared VRAM. therefore slightly reduce input image resolution with resizing. like from 1024x1024 into 768x768 and watch shared VRAM

George Gostyshev

If it's possible - I'm asking for a future adding such as Batch Upscale - it'll be perfect strike for such technology)

puk

🙄 Next try with version 8: C: \Users\AI\SUPIR2\SUPIR\venv\lib\site-packages\safetensors\torch.py:308 in load_file 305 306 307 ) 308 309 310 311 result= 青 with safe_open(filename, framework="pt", device=device) as f: for k in f. keys): result[k] = f.get_tensor(k) return result SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

SECourses

hello. what is your python version? looks like download fail. if you are from china, your internet is being blocked and failing some downloads

Elevo

I just checked, for the photo model Juggernaut-XL-v9 works worse than xl-base-1.0

SECourses

interesting. in my test was better. but you can just change config file and use sdxl 1.0 base

Jonathan Streeter

I get this error on runing: RuntimeError: Pretrained weights (models/open_clip_pytorch_model.bin) not found for model ViT-bigG-14.Available pretrained tags (['laion2b_s39b_b160k'].

SECourses

hello. this happens when you failed to download accurately. please check that downloader.py is run accurately during installation. if you show me entire install cmd logs i can show you where error is

puk

Git pull fix the problem, thanks.

John Dopamine

Works well, thanks! If you do any other updates it might be nice if a double click on the "Upscaled Images Output" opened the image full screen (a la A1111) instead of just cycling to the other image. Lava doesn't work for me at the moment ("LLaVA is not available. Please add text manually.") but that could be my firewall or something I need to fix on my end.

PM

I'm using the same GPU you used on your demo, but do you think using a newer one would make it any faster? or stop it from crashing on larger upscales? Or does it not so much depended on hardware

Khoa Vo

I get "LLaVA is not available. Please add text manually." when clicking the LLaVA button. Is there anything else I have to do after using the 1-click installer?

John Dopamine

One other suggestion if you do an update: Please make an option to carry the caption file (if there is one) when processing the batch and/or an option to use the same filename for the output. I have folders w/ 0001.jpg, 0002.jpg, etc w/ same named .txt files with their captions. When I batch through your app the output filenames are dates(etc) and the caption is not carried over to output folder w/ a similar name. If the names are left the same it'd be easy to use the old .txt caption files, or if the captions are carried over and renamed that would work also. Thanks if you consider should there be a next version.

Erik

Also, can you add face restoration to SUPIR? It's the only thing missing I think. If you add it, then it's perfect.

Kallamamran

We need to be able to change what checkpoint is used. Preferably via drop down and a configuration setting pointing to a checkpoint folder of choice.

Algi

It doesn't work.. I'm on a 3060 and I've let an update run for 30 minutes when trying to upscale 512 to 1024 without changeing any of the settings.. I installed v10. Did a git pull but says everything is up to date..

JackOppss

Hi, I keep getting errors when I try to run the Windows version, I got first a gradio and gradio_imagslider error but I installed them with pip, now I am getting this: Traceback (most recent call last): File "D:\AI\SUPIR\SUPIR\gradio_demo.py", line 7, in from SUPIR.util import HWC3, upscale_image, fix_resize, convert_dtype, Tensor2PIL File "D:\AI\SUPIR\SUPIR\SUPIR\util.py", line 4, in import cv2 ModuleNotFoundError: No module named 'cv2' Press any key to continue . . .

SECourses

Hello. Can you check your task manager before starting the app and tell me how much VRAM your GPU using? If it starts using shared VRAM it will get 20 times slower

SECourses

Hello. This happens when there were errors in windows installer. are you using Python 3.10? can you reinstall and show me entire output of the installer CMD?

SECourses

well they made it loading from a yaml file all. so i didn't spend time to it. but you are right. I will make suggestion of it.

JackOppss

I am reinstalling it now, but I use python 3.11, is it this issue?

daniel mendoza

I just compared juggernaut with base, and juggernaut produces better results.

SECourses

Hello. Today hopefully I am going to add that. Original file name + 001.png

SECourses

It depends on vram. So if you need higher res you need to get more vram. Also start with optimizations and get higher Ram machine. When optimizations are enabled Ram is also important. You can get much bigger resolution

SECourses

So first face restore and then upscale or first upscale and then face restore?

Erik

Upscale and then face restore. Since this model warps the faces a little after upscaling.

SECourses

OK which face restore model you prefer? I think I can add

Erik

CodeFormer or GFPGAN could work.

SECourses

Hello not stuck working. If taking too long that means you did big upscale thus using shared vram. Please check from task manager shared vram usage. I contacted original authors to add cpu off loading to reduce vram further

John Dopamine

Thanks! It'd be perfection if it could include the caption file w/ the same name in .txt format.

Zoltán István Bíró

What can I do wrong? I have a 12 GB RTX 3060, I installed SUPIR without any error message, but above upscale "1" the computer just works indefinitely (64 GB RAM) and the image does not get rendered. Currently I do not set any value, I just paste the original image file, click on "Stage 1 Run", then "Stage 2 Run". Do I need any more settings? Should I install LLaVA manually separately? What should I do to make it work?

SECourses

ok you need to do these 2. restart computer. open task manager and see how much VRAM is being used. then while generating the image check if it is using shared vram or not and let me know please.

SECourses

that caption carry option is also good suggestion. we can add optionally it

SECourses

ok i will try to add either of them. which one you prefer more?

Erik

They are pretty similar. Maybe both? You do get better results if you blend the two. But if you can only do one, then I would go with GFPGAN.

SECourses

ok will try to add GFPGAN hopefully

AI Squad

It's incredible how good it is for an open-source model. In my opinion it was better than Magnific.

Steve

When I try to do a git pull it asks me to identify myself? Is there a good strategy for using your other SOTA with this for training? I was thinking maybe trying to use your auto crop and resize pictures to 512x512 and then running batch supir 1x and again batch at 1.5x and ending up at 768x768?

Zoltán István Bíró

Thank you! After restarting the computer, Utilisation: 98%, Dedicated GPU Memory: 11.5/12.0 GB, GPU Memory: 11.6/43.9 GB and requested Shared GPU memory: 0.0/31.9 GB

Elevo

Check with a picture of the text and you'll see that Juggernaut made the wrong text.

Vlad Selotkin

What could be an options to work with text. Is there any way to preserve text on the image?

SECourses

on which images you plan to apply this strategy? yes it can work

SECourses

yes way better. hopefully i am going to add more features today

SECourses

this is while running right? it should work fast this way because not using shared gpu

Steve

I want to training models of images of people scraped from Reddit. I've been collecting them for a long time now and I keep pruning the bad ones a lot of off content stuff gets posted in the wrong subs. I noticed your gender classifier has a sort sharpness? does that basically sort images by quality? because that would really save me a ton of time

SECourses

you can try sorting. yes in that case upscale may work better

Cyril F

Unfortunately you cannot upscale at very high resolution - With a 4090, starting with an image at 3000px - it takes 300s for an upscale at 1.2 & 50 steps - and do not work with an upscale of x1.5 or x2. The Quality is pretty good and probably one of the best OpenSrouce upscalers, but do not compare (yet) with Magnific.

SECourses

i literally compared with magnific :) in every case SUPIR is better. if you need higher resolution you can rent a big GPU from any cloud such as RunPod and do big upscale there. by the way the tile size may also allow higher resolution. i will research this

SECourses

I see. So if there is a text on images SDXL base is better

Adan Aguilar Cisneros

I get this error after trying to start stage one or stage 2. Could not locate cudnn_ops_infer64_8.dll. Please make sure it is in your library path! then closes.

Adan Aguilar Cisneros

Hi, it happens after i add an image to SUPIR and click Stage 1 Run. I have Kohya, Stable Diffusion, InstaID, OneTrainer and i dont get this error with any of them. Do i have to install it manually within the SUPIR_v12 folder?

SECourses

ok in that case my guess is that you installed during hugging face was down. can you try to install into a fresh folder and send me entire install cmd logs? monstermmorpg@gmail.com and you are using python 3.10.x right?

Diggy Dre

I am getting this when trying to run a stage 2: py:5476: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at ..\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:263.) attn_output = scaled_dot_product_attention(q, k, v, attn_mask, dropout_p, is_causal)

Raf Stahelin

It is absolutely essential to allow the script to BATCH SAVE image filenames SAME AS SOURCE. I have 500 images in my dataset that I need to keep the filneame intact in my pipeline. Please please dont change the filename!!!

Zoltán István Bíró

Yes, I think so too, everything is set up correctly, and yet it can only do 1 upscle, the rest fail. It would help me - and maybe others - if you could make a much more detailed, step-by-step guide to USING it. Click here, set it with parameters from here, etc.

Adan Aguilar Cisneros

Thank you, i tried reinstalling in a different folder and it works fine now.

sadelcri

I can't find the SUPIR_v0.yaml file, where is it supposed to be?

Ant-2014

Thanks for your Herculean labor and regular updates of your projects. I want it to give you more strength and motivation for all your future endeavors! I would like to ask you - if it is possible to insert in SUPIR 1 Click Windows in the program interface Menu a button or window with a drop-down LIST of my different models from PC folder for testing and comparing: Juggernaut, Realvision, SD 1.5 and others from the list (as it is in Stable automatic1111.) :)

SECourses

yes this is on my todo list. while running changing model. i will try to do hopefully

SECourses

it is inside options folder. options folder is inside SUPIR folder

SECourses

yes working on it today. sorry for delay

Erik

Also, it would be great if you could add an option for fixed seeds between generations. Thanks!

Cyril F

There's a difference between upscaling little images to have fun with and very hi-res images for billboard and prints (8K res and up) - You should do test with an image starting at 4000px and upscale it x2. Yes we can rent a big GPU, set it up, download all the 30+GB of files... And do it there, for a monthly cost that would be similar than Maginifc, where you slide an image, wait for a minute and have your beautifully upscaled ready to be retouched. If it is for fun, yes why not, but professionally with big clients expectations and short deadlines, you just pay for the convenience...

SECourses

fixed seed option is there. just uncheck the randomize seed checkbox.

SECourses

hello it is warning but still working. if taking too much time that means you are using shared vram. therefore you need to reduce upscale resolution

SECourses

I see. Well I guess what you are saying true. If you need such big resolution you are right.

Erik

My bad, missed it.

SECourses

no problem. can happen. thanks for support.

daniel mendoza

ON v14 stage 1 don't work. AttributeError: 'str' object has no attribute 'dtype'

SECourses

please do another git pull fixed

SECourses

I plan to make another tutorial after I added some more features. currently adding new features requested by supporters.

Dazzastrous

Still no image slider ,😉

daniel mendoza

I installed it from scratch deleting everything, it still gives an error. I went to the version that previously had V12. Now work perfectly

SECourses

just added that. please test and let me know

John Dopamine

Batch isn't working for me after the update due to picking up the filenames(?). Error processing 2024-02-29_01-59-11-training-sample-0-0-0.png: [WinError 3] The system cannot find the path specified: '' Error processing 2024-02-29_02-19-11-training-sample-757-0-757.png: [WinError 3] The system cannot find the path specified: '' Error processing 2024-02-29_02-39-12-training-sample-1458-0-1458.png: [WinError 3] The system cannot find the path specified: '' Error processing 2024-02-29_02-59-11-training-sample-2204-0-2204.png: [WinError 3] The system cannot find the path specified: '' Error processing 2024-02-29_03-19-12-training-sample-2896-0-2896.png: [WinError 3] The system cannot find the path specified: '' Error processing 2024-02-29_03-39-13-training-sample-3585-0-3585.png: [WinError 3] The system cannot find the path specified: '' Error processing 2024-02-29_03-59-13-training-sample-4334-1-583.png: [WinError 3] The system cannot find the path specified: '' Error processing 2024-02-29_04-19-11-training-sample-5076-1-1325.png: [WinError 3] The system cannot find the path specified: '' etc

John Dopamine

Looks like the issue is that I didn't specify an output path. I don't think leaving it blank so it outputs to the default output folder works anymore. If I have any further issues I'll update.

SECourses

thanks for this info. i will try to fix this too

SECourses

ye sorry for delay fixing right now. working on it

SECourses

just fixed with v15. please close app, do a git pull or run update, start again and it should work now. ty

SECourses

just fixed with v15. please close app, do a git pull or run update, start again and it should work now. ty

Onkel Fraggel

RuntimeError: Current CUDA Device does not support bfloat16. Please switch dtype to float16. So I had to change Auto-Encoder Data Type to fp32. Is this the right way to do it? Im running on a 2080ti 11gb vram card. How to optimize the settings to get upscale faster?

Pedro Serapio

No luck on Window 11, the installation run without any errors. When starting it always complains that gradio is not present.

SECourses

that means during installation you had an error. what is your python version? can you reinstall into a fresh folder and show me entire cmd output? you can email logs to monstermmorpg@gmail.com

Betinho Formado

this new version just broked my working version sadly not running anymore I will try to re-install again would be good for this install recognise the models so doesn't download when is existing on the folder I think would save time.

Neven Krcmarek Astrophotography

Thank you for this easy installment. I have 3090 and when I do Stage 2 for x2 upscale of the image 2048x2560px. It eats up whole 24 vram and it takes very long to upscale. Is this something that will get faster? I would very much like to make 8k res upscaling but this is taking really long just for 4k per image. Thank you.

SECourses

Thank you so much. Yes the code is very VRAM demanding and they didn't implement shard in into multiple GPUs. So in your case it starts using shared VRAM instead of double gpu. This is also NVIDIA guilt. NVIDIA want you to buy pro cards. But I am talking with authors to get implemented cpu off loading to further reduce vram usage.

SECourses

You can do this. Delete downloader py so it won't download models. But make sure to copy paste models later. By the can you give me more info how broke? I should fix if there is error

Betinho Formado

oh awesome I will give a go on the next update thanks :)

Betinho Formado

completely broke reinstall on my computer v12 and v13 don't run anymore try installing both versions from scratch and again like before stop when reads the "SUPIR_v0.yaml" Then no extra message just saying press any key and close the cmd and doesn't give me any more information about the URL I'm back to square one haha I need to avoid those updates.

Andre Bopp

A suggestion: I found out that my graphics card can generate an output of at most 1344 x 1344 pixels without generating an OOM. To use this information, I had to calculate the upscale factor for each input image depending on the resolution. Would it be possible to add a field with "upscale to ... max pixel/side" so that the upscale factor for each image is automatically calculated in the background? So the longer side of the image is taken and scaled up to a length of 1344 pixels, in my case.

Attila Karácsonyi

Just a tip: user Everything from void tools, is an instant search tool and You will find "everything"!

Attila Karácsonyi

I just joined and would like to express my gratitude for the extreme effort You are putting into educating all of us!! ❤️

Attila Karácsonyi

I had the same issue then checked my SysEnv and fixed the path for python, and voilà :) it loaded the UI. I am about to try it for the first time NOW! :)

SECourses

the thing is i am not sure exactly how output is calculated. for example if i give 1024x1024 and upscale 1x i get 1024x1024 output then i give 512x512 and do 1x upscale and i get output 1024x1024 then i give 256x256 and do 1x upscale and i get output 1024x1024

SECourses

dmn :D you can rent a big GPU on RunPod or anywhere and get done quickly. by the way upscale is mind blowing

Hung Do

a little confused about Param Setting and Model Selection. If I choose Fidelity in Param Setting shouldn't Model Selection auto switch to v0F.ckpt? Or is it 4 possible combination

SECourses

I also don't know for sure. but we use best models. i didn't change their implementation at this feature.

SECourses

during install what do you seen? you can email me entire log so i can see : monstermmorpg@gmail.com

SECourses

awesome thank you so much and welcome. i am working on adding more features right now

AI Squad

Still dreaming with an Update where you will say that a Kaggle notebook is available \o/ 😁

SECourses

I am waiting authors to fix FP16 bug :/

marly.ai

I'm using it on Windows 11 and it doesn't work. It loads phase 1 and when it goes to phase 2, it keeps loading the preloading animation and doesn't finish. :/

SECourses

how much vram you have? what is your input resolution image and how much upscale you are trying?

marly.ai

My computer has 128gb ram with a 12gb video card. I was trying to upscale to 4x but was unsuccessful. I formatted the computer and will try again. It would be interesting if you published a new version or made available an updated version of 1click install

marly.ai

https://i.imgur.com/DMrYipI.jpeg As you can see, I formatted the computer, updated the drivers and it still didn't work for me again. I tried decreasing the upscale to 2x and it didn't work either (I followed exactly the recommendations in your video, clicking on 1click install) I tried a new test, putting it on the windows c drive (Because before I tried on an external ssd and it didn't work either)

marly.ai

in stage 2 it loads infinitely, I check the command prompt and it loads the processing of the queue several times and does not give me the result. Nor does the progress bar appear as it does in the video tutorial. can you help me?

SECourses

hello. i can connect your computer via any desk and check out if you wish. message me from Discord

marly.ai

Perfect! I just sent you a message on discord

Dmitry

Not a man, you're a machine) I can't imagine when you manage to do so much in so many different directions. Just thank you for your hard work

SECourses

thank you so much. yes taking huge huge time :/

GET GOING FAST | @TheAIHobbyGuy

Good stuff. thanks! and I paid ;D question: running a 4090 .. i uploaded a photo of a cowboy wearing a detailed shirt... the SUPIR image comes out with his face looking better but his clothes all "smoothed" out. Suggestions? Thanks

Meito

hi anyway to roll back to supir v16, im getting worse results with the new one and its taking 3x longer for me

daniel mendoza

It happened to me with ocean waves, the texture was incorrect until I specified it in the prompt. Put the type of texture of the clothing fabric in the prompt and it will come out as it should.

SECourses

interesting. are you using the model we defined or another model? did you define a good prompt? can you share image with me? monstermmorpg@gmail.com

SECourses

Hi yes you can but can you message me from discord? if there is an issue I would like to fix because we are adding all new features. I need some details. For example we added Linear CFG by default did you turned it off and tested? to return back v16, open a cmd inside supir and do this git checkout 7702b200a082f3173ccaf6d31d457a3b3ccad704

jamzsam

i have python 3:10:11 installed but it won't install.

jamzsam

I get this error : ERROR: Could not install packages due to an OSError: HTTPSConnectionPool(host='cdn-lfs.huggingface.co', port=443): Max retries exceeded with url: /repos/1b/95/1b9587c6131e50dc30fec9d8d829ec2de37d44cf22183191ea1235a5089d904f/223b123e44b95e148904ab2475e6a6230a013813cb25e80d11f9d6d062b768b8?response-content-disposition=attachment%3B+filename*%3DUTF-8%27%27triton-2.1.0-cp310-cp310-win_amd64.whl%3B+filename%3D%22triton-2.1.0-cp310-cp310-win_amd64.whl%22%3B&Expires=1709861169&Policy=eyJTdGF0ZW1lbnQiOlt7IkNvbmRpdGlvbiI6eyJEYXRlTGVzc1RoYW4iOnsiQVdTOkVwb2NoVGltZSI6MTcwOTg2MTE2OX19LCJSZXNvdXJjZSI6Imh0dHBzOi8vY2RuLWxmcy5odWdnaW5nZmFjZS5jby9yZXBvcy8xYi85NS8xYjk1ODdjNjEzMWU1MGRjMzBmZWM5ZDhkODI5ZWMyZGUzN2Q0NGNmMjIxODMxOTFlYTEyMzVhNTA4OWQ5MDRmLzIyM2IxMjNlNDRiOTVlMTQ4OTA0YWIyNDc1ZTZhNjIzMGEwMTM4MTNjYjI1ZTgwZDExZjlkNmQwNjJiNzY4Yjg~cmVzcG9uc2UtY29udGVudC1kaXNwb3NpdGlvbj0qIn1dfQ__&Signature=gQNTYXxgejsDNSw7iEWSDNVPesa0orlgv6EAbzuS2yC0aYYC0MmF5VclER4369Egx-3EvSMzOMR-mq0yblToEeLuzldrDLvQBpKBbSNhUm9cbVpY1g2hYdF~YJMgCwDBfTnRbdBEqNzSNbjqtVUxAerQfNb0KR1S1jUajuOsOkCUxP9WfiD5v4diJaTetpk9z9sdWOKMQUlb-NQ5KL3jHg2TpUPf9Wm1ak~950aB0lH1G~IJFpVyO9rL8PHAyfbqHLhWI6sVhS2Wx1FElDpfKhJAGGWqzmXsg4F94whXNe3NqoXv0YaaCsbQi-gtFa8976-X3ToqA6QNNgj1AEGoRQ__&Key-Pair-Id=KVTP0A1DKRTAX (Caused by ProtocolError('Connection aborted.', ConnectionResetError(10054, 'An existing connection was forcibly closed by the remote host', None, 10054, None)))

SECourses

hello. this is related to your internet connection. it can be due to your internet connection provider or your anti virus. i would restart computer and try again. reset your modem too

SECourses

sorry for delay. I just added target resolution display. can you let me know if that solves your issue?

jamzsam

Ok. Ill give it a try. Thanks for the reply

Andre Bopp

Thanks for the added feature. 😊

GET GOING FAST | @TheAIHobbyGuy

Suggestion: Would be great to have a option to create a ffmpeg video of the slider.. smooth movement from original to SUPIR'd

SECourses

no idea how can we make it haha :D

SECourses

have you seen anything that does it anywhere else? i can look code if exists

SECourses

I am looking a way to make right now. I think I can implement

SECourses

I just added that feature check it out :)

GET GOING FAST | @TheAIHobbyGuy

that solved it for me as well. Lol. I keep getting tempted to forget how AI works.. its funny/weird remembering to prompt

GET GOING FAST | @TheAIHobbyGuy

hahah... love that you did that. hahaha! Awesxome. Sidenote: Listen.. im not trying to blow smoke up anything, but you did a great job with this. I sent some photos to my family and they were blown away. I hope you make a lot of $$$ cause it was way worth the $5

GET GOING FAST | @TheAIHobbyGuy

Save to video feedback: It's hit or miss.. i've had a perfectly square photo work and then I've had a height taller than width fail. Here's the latest error (I run a 4090): RuntimeError: CUDA error: an illegal memory access was encountered CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

Dead mau5

Hi could you please help me resolve this issue, it happens after clicking on stage 2 button: File "X:\Upscaling\SUPIR\venv\lib\site-packages\gradio\queueing.py", line 495, in call_prediction output = await route_utils.call_process_api( File "X:\Upscaling\SUPIR\venv\lib\site-packages\gradio\route_utils.py", line 235, in call_process_api output = await app.get_blocks().process_api( File "X:\Upscaling\SUPIR\venv\lib\site-packages\gradio\blocks.py", line 1627, in process_api result = await self.call_function( File "X:\Upscaling\SUPIR\venv\lib\site-packages\gradio\blocks.py", line 1173, in call_function prediction = await anyio.to_thread.run_sync( File "X:\Upscaling\SUPIR\venv\lib\site-packages\anyio\to_thread.py", line 56, in run_sync return await get_async_backend().run_sync_in_worker_thread( File "X:\Upscaling\SUPIR\venv\lib\site-packages\anyio\_backends\_asyncio.py", line 2144, in run_sync_in_worker_thread return await future File "X:\Upscaling\SUPIR\venv\lib\site-packages\anyio\_backends\_asyncio.py", line 851, in run result = context.run(func, *args) File "X:\Upscaling\SUPIR\venv\lib\site-packages\gradio\utils.py", line 690, in wrapper response = f(*args, **kwargs) File "X:\Upscaling\SUPIR\gradio_demo.py", line 313, in stage2_process load_qf() File "X:\Upscaling\SUPIR\gradio_demo.py", line 112, in load_qf ckpt_Q, ckpt_F = load_QF_ckpt('options/SUPIR_v0.yaml') File "X:\Upscaling\SUPIR\SUPIR\util.py", line 75, in load_QF_ckpt ckpt_F = torch.load(config.SUPIR_CKPT_F, map_location=device) File "X:\Upscaling\SUPIR\venv\lib\site-packages\torch\serialization.py", line 1005, in load with _open_zipfile_reader(opened_file) as opened_zipfile: File "X:\Upscaling\SUPIR\venv\lib\site-packages\torch\serialization.py", line 457, in __init__ super().__init__(torch._C.PyTorchFileReader(name_or_buffer)) RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

SECourses

hi. looks like your downloads were not complete. also what is your python version?

SECourses

i think it is related to your computer. try restart the computer. i plan to add batch extract frames and also convert frames back into video feature as well

Tiger

thank you for the great works. Much appreciated.

Cyril F

it seems there's a bug with FaceRestore. It happened a few times where it is restoring the face properly, but not upscaling the rest of the image, or sometimes it is upscaling everything, but leaving a grid (or artifacts) on the face. See examples here >> https://i.gyazo.com/45a67262f1dea39b9a074d7fcb2a9ab2.jpg

he da

ModuleNotFoundError: No module named 'facexlib'

Ivan Stoyneshki

On V22 I get no error but in shell I see - Press any key to continue.... and thats all it shuts off... this happens every time just after one upscale, the second image crashes every time. Tried numerous times. And the process of start i really slow. The interface shows up faster but after that its very slow. I run ot 4060 TI 16GB with 32 gb ram system memory.

SECourses

it fast load because it doesn't load models at beginning. then it starts loading after you click. can you message me from discord i can connect via any desk and check out

SECourses

yes either you need to activate venv and install it or do a fresh install with latest version.

masharegister

Hello/ How can i fix this mistake Traceback (most recent call last): File "C:\Supir0603\SUPIR\venv\lib\site-packages\gradio\queueing.py", line 495, in call_prediction output = await route_utils.call_process_api( File "C:\Supir0603\SUPIR\venv\lib\site-packages\gradio\route_utils.py", line 235, in call_process_api output = await app.get_blocks().process_api( File "C:\Supir0603\SUPIR\venv\lib\site-packages\gradio\blocks.py", line 1627, in process_api result = await self.call_function( File "C:\Supir0603\SUPIR\venv\lib\site-packages\gradio\blocks.py", line 1173, in call_function prediction = await anyio.to_thread.run_sync( File "C:\Supir0603\SUPIR\venv\lib\site-packages\anyio\to_thread.py", line 56, in run_sync return await get_async_backend().run_sync_in_worker_thread( File "C:\Supir0603\SUPIR\venv\lib\site-packages\anyio\_backends\_asyncio.py", line 2144, in run_sync_in_worker_thread return await future File "C:\Supir0603\SUPIR\venv\lib\site-packages\anyio\_backends\_asyncio.py", line 851, in run result = context.run(func, *args) File "C:\Supir0603\SUPIR\venv\lib\site-packages\gradio\utils.py", line 690, in wrapper response = f(*args, **kwargs) File "C:\Supir0603\SUPIR\gradio_demo.py", line 446, in stage2_process results = [face_helper.paste_faces_to_input_image(upsample_img=_bg[0])] File "C:\Supir0603\SUPIR\SUPIR\utils\face_restoration_helper.py", line 374, in paste_faces_to_input_image assert len(self.restored_faces) == len( AssertionError: length of restored_faces and affine_matrices are different.

masharegister

Thanks for the advice. The problem was solved by completely reinstalling the application. There is no more of this error. I've updated the video card driver too! çok teşekkür ederim

Mikael Svenson

A minor thing you should rename the option "Start With White Theme" to be "Start with Light Theme". Matches "Dark" better and is the correct term :)

VRMOTION

I have installed V22 but I don't have pressing stage 2 only stage1. rather, the stage 2 button is missing

SECourses

we changed the interface. we are still working. just use upscale image button.

VRMOTION

After I press the button, the result ends with an error. assert len(self.restored_faces) == len( AssertionError: length of restored_faces and affine_matrices are different.

SECourses

hello hopefully it will get fixed. can you test with writing a face prompt ?

Davit Sharian

can I change the checkpoint for example to realviis lightning model, and make it 4-5 steps and lower cfg scale ?

SECourses

some people did it and reported it is working. please let me know if you try

SECourses

you can try. some said it works. if you try let me know please.

hacbachvotinh

In version 26, if we can restore the video RAM (VRAM) back to its original state after using LLAVA to process information, then running Supir won't occupy more than 12GB of video memory and the processing speed won't significantly slow down.

masharegister

How can I install and not update anymore version 14 or 16 of the program? It worked so well for me.

Flambo

Yes this is possible, but you must modify SUPIR\gradio_demo.py and change minimum=20 to minimum=4 to allow for lower than 20 steps to be picked.

daniel mendoza

Hello, I have a suggestion. You can set the upscale number to be a target resolution instead of a number. And use this formula, for example, the input image is 1035x1035 and my target resolution is 2560x2560. Divide the target resolution with the input image resolution and this is my scale number. 2560/1035=2.47 unique value. This is so that the scale always has a target output resolution and does not have to accommodate it depending on the pixel size of the input image. I'm doing it this way manually but I would like the app to have that feature and I would appreciate it if you included it. I understand that if the image is not a 1:1 aspect ratio it becomes complicated. But it would be applying the formula only on one of the dimensions (width or height), for example: 1024x768 is the input resolution, I want to expand it to 1560, then: 1560/1024=1.52 upscale value.

hacbachvotinh

I tested it and found that the realvisxlV40_XL_V40_Bakedvae.safetensors model produces good quality images similar to Juggernaut-XL_v9_RunDiffusionPhoto_v2.

Нона Ангелова

I just subscribed to your Patreon and downloaded the latest version (I'm on Windows 11 and have all the requirements) However, when I try to start it, I bet this error "H:\A.I. Software\Supir\SUPIR\gradio_demo.py:1046: SyntaxWarning: invalid escape sequence '\S' placeholder="R:\SUPIR video\comparison_images") H:\A.I. Software\Supir\SUPIR\gradio_demo.py:1049: SyntaxWarning: invalid escape sequence '\S' placeholder="R:\SUPIR video\comparison_images\outputs") Traceback (most recent call last): File "H:\A.I. Software\Supir\SUPIR\gradio_demo.py", line 11, in import einops ModuleNotFoundError: No module named 'einops'" In the same time, I have no issue running Supir from Pinokio

Meito

Hi Dr, can you please share the git checkout for v19 v20 These seem to be the best versions for me. Also doing backtesting. Latest one doesn't work for me. Trying to figure out the issue as well. Thanks

puk

Hy Furkan, i tested he version 27 when i apply face restoration the restored face is not fused with the scaled image. also the output image is blurry and look like the output of stage 1

SECourses

hello just fixed faces issue with V35 can you test?

SECourses

you can do. open SUPIR folder and for V19 do this : git checkout d97388a067da322dab97cbe7469fdb2aeffd661b

SECourses

hello. if your python version is 3.10.11 then your error is using same folder name. please install into h:\supir_new and it will work hopefully

SECourses

nice. now you can hot reload models it is added

SECourses

when you set 2.47 does it work exactly? i think we can add target resolution option but what if aspect ratio of user input not matches? how should it behave?

SECourses

open SUPIR folder and do this for v16: git checkout 7702b200a082f3173ccaf6d31d457a3b3ccad704

SECourses

i will try to add an option for this. so deload LLaVA after process

SECourses

sure i just did it now please update to latest

100Sachen80

Hi Mate, when using LLAVA, the VRAM does not clean afterwards and before Stage 2? Without LLAVA it uses aroung 10GB, using LLAVA it takes > 14GB, see: [Tiled VAE]: Done in 51.395s, max VRAM alloc 14153.721 MB I have a 12GB 3060, so Stage 2 then takes forever when LLAVA is activated. Can you implement a VRAM wipe before Stage 2, or write out the LLAVA result in the output folder before stage 2? So we could use it manually as a prompt, if we have to restart SUPIR (then without LLAVA) because of VRAM

Нона Ангелова

I tried it but without any luck. The compile is wrong, because it tries to find something that's structured like on your PC, for example "R:\SUPIR video\comparison_images\outputs" - I don't have "R" directory on my PC, I guess this is yours.

VRMOTION

when you click on LLaVA, nothing happens, no hint is generated. I just did everything as in the lesson. Should LLaVA work?

SECourses

in code we don't have anything like that except as a placeholder to show you how to set your paths accurately. can you please show me screenshots from discord? error screenshot

hacbachvotinh

A pretty good experience: I exited Photoshop before running supir. Although it only saved 0.8GB of video ram, the image speed of Supir is much faster because it uses less than 12GB of video ram of the 3080ti.

kenishii

Hey there, I don't know what I am doing wrong, but it alwas crashes: Processing images (Stage 2) Building a Downsample layer with 2 dims. --> settings are: in-chn: 320, out-chn: 320, kernel-size: 3, stride: 2, padding: 1 Building a Downsample layer with 2 dims. --> settings are: in-chn: 640, out-chn: 640, kernel-size: 3, stride: 2, padding: 1 Drücken Sie eine beliebige Taste . . . (press any key...)

Diego Sienra

Hi, great update!! I have a problem, I have a rtx 3090 and I was using no tiled mode and I seems to have better results in some cases, for example more natural hears running the whole image in one run vs tiled, the problem with the new version is that i can run it just one time, in the second generation I get out of memory, there is way to free the vram after each generation? may be this can fix the issue. thanks

Diego Sienra

another question, because the models are sdxl models with a native resolution of 1024x1024 tiling at 1024x1024 rather than 512x512 is posible to get better results? additional unrelated question what is BG restoration?

Davit Sharian

V22 works for me, but when starting the V37 getting this error message, my gpu is RTX3090 WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for: PyTorch 2.1.0+cu121 with CUDA 1201 (you have 2.1.0+cpu) Python 3.11.6 (you have 3.11.5) Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers) Memory-efficient attention, SwiGLU, sparse and more won't be available. Set XFORMERS_MORE_DETAILS=1 for more details Traceback (most recent call last): File "C:\Users\davsh\Documents\SUPIR_v37\SUPIR\gradio_demo.py", line 69, in raise ValueError('Currently support CUDA only.') ValueError: Currently support CUDA only.

SECourses

BG restore means restoring entire image except face. I didnt try tiling but i plan to test later

SECourses

i was never able to run without tiles on RTX 3090 :D i dont know if we can reduce VRAM. are you using LLaVA?

SECourses

helo. this doesnt give the reason. what is your python version when you open cmd and type python?

SECourses

hello we fixed that. please use latest version. now it will return llava prompt into prompt box

SECourses

the reason is Python 3.11. because on Python 3.10 we don't get that error. i will test to verify this now on a fresh install

Diego Sienra

Yes Im using LLaVA, It works for one generation starting with "Start As Half Params" and "Start Without Tiled VAE". without LLaVA it works fine, I tested it in previews versions but if I set larger resolutions starts to use gpu shared memory and it gets very slow, I have 48 RAM and 24gb VRAM

SECourses

true 24 gb is not being sufficient without optimizations

Diego Sienra

ok thanks, I understand, but why it works in the first generation?

Diego Sienra

Use both VRAM optimizations and it uses around 12 GB VRAM GPU If you have over 30 GB VRAM, you can start both full Params and no Tiled-VAE modify this file and add --share if you want Gradio share Please select an option: 1. Start As Half Params - Uses Lesser VRAM Preferred 2. Start As Full Params Enter your choice (1-2): 1 Please select an option: 1. Start Using Tiled VAE - Uses Lesser VRAM Preferred 2. Start Without Tiled VAE Enter your choice (1-2): 2 Please select an option: 1. Start With Light Theme 2. Start With Dark Theme Enter your choice (1-2): 2 Running on local URL: http://127.0.0.1:7860 To create a public link, set `share=True` in `launch()`. Processing LLaVA Processing images (Stage 1) Building a Downsample layer with 2 dims. --> settings are: in-chn: 320, out-chn: 320, kernel-size: 3, stride: 2, padding: 1 Building a Downsample layer with 2 dims. --> settings are: in-chn: 640, out-chn: 640, kernel-size: 3, stride: 2, padding: 1 making attention of type 'vanilla-xformers' with 512 in_channels building MemoryEfficientAttnBlock with 512 in_channels... Working with z of shape (1, 4, 32, 32) = 4096 dimensions. making attention of type 'vanilla-xformers' with 512 in_channels building MemoryEfficientAttnBlock with 512 in_channels... Building a Downsample layer with 2 dims. --> settings are: in-chn: 320, out-chn: 320, kernel-size: 3, stride: 2, padding: 1 Building a Downsample layer with 2 dims. --> settings are: in-chn: 640, out-chn: 640, kernel-size: 3, stride: 2, padding: 1 Loaded model config from [options/SUPIR_v0.yaml] and moved to cpu Loaded state_dict from [C:\Users\Diego\Downloads\SUPIR_v36\SUPIR\models/checkpoints\Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors] Loaded state_dict from [models/v0Q.ckpt] You are using a model of type llava to instantiate a model of type llava_supir. This is not supported for all configurations of models and can yield errors. Loading vision tower: openai/clip-vit-large-patch14-336 Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████| 2/2 [00:17<00:00, 8.85s/it] LLaVA loaded. LLaVA moved to GPU. C:\Users\Diego\Downloads\SUPIR_v36\SUPIR\venv\lib\site-packages\transformers\models\llama\modeling_llama.py:671: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at ..\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:263.) attn_output = torch.nn.functional.scaled_dot_product_attention( Processing images (Stage 2) Seed set to 475310794 Image 1/1 upscale completed. Last upscale completed in 26.83 seconds Label changed: Image Upscaling Completed: processed 1 images at in 30.23 seconds #2 Image Upscaling Completed: processed 1 images at in 30.23 seconds #2 Updating Single Output Image

Diego Sienra

This is my log my gpu usage is 21.3GB but it works including LLaVA, I tried again it works up to 1024x1024 multiple times, fails at higher resolutions or aspect ratios

SECourses

i see. we are trying to improve memory usage

kenishii

Hi, thank you very much for your fast reply. It is 3.10.11. but today in the morning I tried it again and now it works. I have no clue what the reason was. sometimes just going to sleep helps. :-)

Casper

Could you please me a short walkthrough? I can't keep up with the updates

SECourses

you are welcome. i will make a new tutorial for all changes

So Sha

Unfortunately after first run my system crashed! ( RTX 4090 & 72 GB RAM ) ( Enabled "Apply LLaVa" & "Apply Stage 1" & "Apply Stage 2" ) I restarted again and run again without activating llava and it worked. One more thing, how we can add more details to the images and lets AI do some imagination like magnific?

SECourses

if computer got blue screen or restart, it is likely due to RAM CPU voltage and speeds. happened me as well. to add more details you can try to increase Text Guidance Scale

Ec Jep

I also have 4090 with 24gb and using s1, s2 and llava so far no issues. Thx for releasing this image processor! It is really fixing blurry pics with excellent quality

masharegister

Good day. Why the application Supir installed to the C:\Windows\System32 , but not in the same folder where I run the installation.

SECourses

no dont install there. install into C:\SUPIR_AI

SECourses

this happens when donwload models were not completed. please complete them. you can do either fresh install or execute download_models.py

Siva

Is it possible to make it work on 3080 10GB?

SECourses

it can work but would be super slow due to using shared VRAM. but on 12 GB works great

hacbachvotinh

You are a person who is extremely dedicated to the product, constantly updating and very responsive to user needs. I will follow and support you for a long time.

100Sachen80

Hi Furkan, the file "update_windows_requirements.bat" is new, do we need to run it regularly?

SECourses

no only when you get error related libraries while running

100Sachen80

v41 gives some weird ghost effects on the created images. looks like the face can be seen twice but very faint and only parts of it, shifted to the left or right of the actual face. default settings on v41 with juggernautxl. switched back to v38 for now.

SECourses

interesting. i didnt notice such effect. are you sure exactly same settings?

Scott

This is such a great tool, though it takes me a very long time to run anything over about 2,000 x 2,000 pixels (and even that takes about 2-5 minutes to enlarge on a rtx 4060 ti 16GB). if I have different size images in the batch folder and some upscale too far it takes a long time. It would be great if the UI could also target upscaling by size or megapixel size. Some graphics programs have something like this like "resize so longest side is x pixels." or even "target x megapixels" and have it choose the right scale while keeping the aspect ratio. That would make it a bit more predictable when running batches. Also, is there a rule about what photos need stage 1 run first? Is it mostly just tiny thumbnails that need to have that run?

SECourses

we removed stage 1. yes i am planning to add maximum resolution as you suggested

VRMOTION

ERROR: Could not install packages due to an OSError: [WinError 5] Отказано в доступе: 'F:\\TUTORIAL_AI\\SUPIR\\SUP\\SUPIR_v41\\SUPIR\\venv\\Lib\\site-packages\\~orch\\lib\\asmjit.dll' Check the permissions.

SECourses

hello. Try to install exactly into this folder C:\\SUPIR_v41 moreover what is your python version 3.10.11?

SECourses

it looks like worked. where did you set your output param? which disk how? remove outputs_folder param and try default folder

Ivan Stoyneshki

--outputs_folder_button "True" --outputs_folder "F:\Supir-Output" - supir is installed o F

Ivan Stoyneshki

never had a problem with previous versions

Ivan Stoyneshki

started with default same the right windows stays empty and no images in output folders

SECourses

are you using latest version? i have no such issues. what do you see on CMD can you email me? monstermmorpg@gmail.com. message me entire CMD window message after upscaling an image

SECourses

No errors. Can you try different images simple ones. No face restore. Just make sure all images are failing. Also message me from discord so only way is connect your computer and debug

Ivan Stoyneshki

Contacted you on Discord the Face Restore is the problem

Andre Bopp

I don't know why, but my results are always more blurry if I'm using face restoration. (other settings are default) Am I the only one with this issue? And does the resolution of the input image have a influence on this process? Nevertheless, thanks for the great work and commitment.

SECourses

thank you so much. sadly face restore is far from perfect. it is a hit or miss. in some images works perfect and in some fails.

crow

ran the installer and everything went good. loaded up the UI and tried to run an upscale but my machine keeps hard crashing on stage2 loading model SUPIR_v0_tiled.yaml. any help would be great

SECourses

can you give me more info regarding your machine? python version? cpu? gpu? thank you

crow

5800x, 3090, 3.10.9

SECourses

ye they look great. so to debug your issues i need these 2 info or connect your pc 1 : fresh install and entire log of your cmd 2 : entire log of your cmd when you do inference / upscale can you email me them : monstermmorpg@gmail.com

HFS123

I have it installed and running but the final generated images are all just textured gray. installed it slightly differently since I already had the models in my machine, i made sure all the paths are correct in the scripts as well as the config files. So I don't know what is missing?

SECourses

sadly no idea. i think try to do a fresh install and try again. can you also post example of textured gray?

HFS123

i've added the screenshot to snipboard. https://snipboard.io/SZMUcL.jpg you can see the grey/brown image on the right side of the image slider ** update :- it's okay, it's working now. just an issue with one of my paths! all fine

HFS123

all working now

HFS123

so i have narrowed down the problem. my SUPIR models were originally named: SUPIR-v0F.ckpt and SUPIR-v0Q.ckpt your package renames them to v0F.ckpt and v0Q.ckpt which is fine. but if I choose to keep the original names (SUPIR-v0F.ckpt and SUPIR-v0Q.ckpt) - and i make sure the paths are correct in the yaml files, that is when i get gray/brown images. I searched all the files for references to these files thinking they are maybe hardcoded somewhere, but I can't find them.. can you explain? is there somewhere where it insists that they must be named v0F.ckpt and v0Q.ckpt?

SECourses

we develop with multiple people so yes probably we have some hard coded of them. you can search entire code to see

HFS123

thanks, I found it.

Additional Contributions

File "D:\SUR\SUPIR\gradio_demo.py", line 3, in import gradio as gr ModuleNotFoundError: No module named 'gradio'

SECourses

you had installation error. reinstall and send me entire cmd window as email : monstermmorpg@gmail.com

Additional Contributions

Dr., I re-installed and installed successfully, China's Tsinghua source is not working well, I canceled it and it works, thanks!

Additional Contributions

Doc, why don't you combine the captioning model with the GPT, Claude, which could make you more profitable.

SECourses

it depends on GPU but I just tested it for you with newest FP8 on RTX 3090 TI and it took 321 seconds

Siva

I think with this new version it should work on mine.

Male AI Art

I did a v44 install on windows 10, GPU is a 3080ti 12gb. The installation was completed successfully, but when I started the program, dragged and dropped the image into the input image, set the upscale size to 2, and clicked on Process Signle, I got the message "An exception occurred. Please try again." The process does not proceed.

SECourses

what do you see on cmd? can you tell me

Marko Radosavljevic

An exception occurred: stat: path should be string, bytes, os.PathLike or integer, not NoneType at Traceback (most recent call last): File "P:\SUPIR_V44\SUPIR\gradio_demo.py", line 744, in start_single_process _, result = batch_process(img_data, **values_dict) File "P:\SUPIR_V44\SUPIR\gradio_demo.py", line 1054, in batch_process ckpt_select = get_ckpt_path(ckpt_select) File "P:\SUPIR_V44\SUPIR\gradio_demo.py", line 249, in get_ckpt_path if os.path.exists(ckpt_path): File "C:\Users\TBR\AppData\Local\Programs\Python\Python310\lib\genericpath.py", line 19, in exists os.stat(path) TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType

Marko Radosavljevic

Tried everything, also clean install... always the same error... BTW, good job man, thanks :)

Christophe Rolland

Same error : An exception occurred: stat: path should be string, bytes, os.PathLike or integer, not NoneType at Traceback (most recent call last): File "C:\Windows\System32\SUPIR\gradio_demo.py", line 744, in start_single_process _, result = batch_process(img_data, **values_dict) File "C:\Windows\System32\SUPIR\gradio_demo.py", line 1054, in batch_process ckpt_select = get_ckpt_path(ckpt_select) File "C:\Windows\System32\SUPIR\gradio_demo.py", line 249, in get_ckpt_path if os.path.exists(ckpt_path): File "C:\Program Files\Python310\lib\genericpath.py", line 19, in exists os.stat(path) TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType. Everything worked fine before v44 installation. Tryed clean install too but no succes :[

SECourses

hello. did you provide --ckpt argument? did you get any error while installing? i think your models were not downloaded. what do you see in models folder and in models/checkpoints folder?

SECourses

hello. did you provide --ckpt argument? did you get any error while installing? i think your models were not downloaded. what do you see in models folder and in models/checkpoints folder?

Marko Radosavljevic

Models are in folder, but invisible in model selector in gradio... No errors during install, no arguments added... I restored previous version, works like charm :) I just wanted to inform you about ver_44 error.

SECourses

I just did a fresh install and literally no issues. if you contact me from discord i can do any desk and check your computer

Christophe Rolland

Hi, after another reinstall, models were downloaded. All is right now, thanks !

Male AI Art

To create a public link, set `share=True` in `launch()`. Exception in callback _ProactorBasePipeTransport._call_connection_lost(None) handle: Traceback (most recent call last): File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\asyncio\events.py", line 80, in _run self._context.run(self._callback, *self._args) File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\asyncio\proactor_events.py", line 165, in _call_connection_lost self._sock.shutdown(socket.SHUT_RDWR) ConnectionResetError: [WinError 10054] 既存の接続はリモート ホストに強制的に切断されました。 Exception in callback _ProactorBasePipeTransport._call_connection_lost(None) handle: Traceback (most recent call last): File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\asyncio\events.py", line 80, in _run self._context.run(self._callback, *self._args) File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\asyncio\proactor_events.py", line 165, in _call_connection_lost self._sock.shutdown(socket.SHUT_RDWR) ConnectionResetError: [WinError 10054] 既存の接続はリモート ホストに強制的に切断されました。 An exception occurred: stat: path should be string, bytes, os.PathLike or integer, not NoneType at Traceback (most recent call last): File "E:\SUPIR_v44\SUPIR\gradio_demo.py", line 744, in start_single_process _, result = batch_process(img_data, **values_dict) File "E:\SUPIR_v44\SUPIR\gradio_demo.py", line 1054, in batch_process ckpt_select = get_ckpt_path(ckpt_select) File "E:\SUPIR_v44\SUPIR\gradio_demo.py", line 249, in get_ckpt_path if os.path.exists(ckpt_path): File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\genericpath.py", line 19, in exists os.stat(path) TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType

SECourses

hello. this means your model downloads were failed. please do a fresh install and pay attention to all messages. you can email me entire log of fresh install : monstermmorpg@gmail.com

Male AI Art

I should have reinstalled first, I reinstalled and it is working fine, thank you very much for your help.

Jian Shen

Hello, can you help me with this error: raise RepositoryNotFoundError(message, response) from e huggingface_hub.utils._errors.RepositoryNotFoundError: 401 Client Error. (Request ID: Root=1-660952ce-60eb16b47db69e39669e7575;714b3109-cbde-4694-8055-0d10bc28f6c7) Repository Not Found for url:https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main. Please make sure you specified the correct 'repo_id' and 'repo_type'. If you are trying to access a private or gated repo, make sure you are authenticated. InValid username or password.

SECourses

Hello. welcome. i just replied you from email. the reason is that during initial install some models were not fully downloaded.

Jian Shen

Thank you! fully reinstalled and solved the problem. And the app works well, thumbs up!

Diggy Dre

For some reason, RunPod refuses to start a Supir job now. It just comes right back with "An exception occurred, please try again"

SECourses

pod can be broken. it happens frequently. try another pod

Abdo Asker

I have a problem .. this appears to me:

Abdo Asker

An exception occurred: stat: path should be string, bytes, os.PathLike or integer, not NoneType at Traceback (most recent call last): File "/workspace/SUPIR/gradio_demo.py", line 744, in start_single_process _, result = batch_process(img_data, **values_dict) File "/workspace/SUPIR/gradio_demo.py", line 1054, in batch_process ckpt_select = get_ckpt_path(ckpt_select) File "/workspace/SUPIR/gradio_demo.py", line 249, in get_ckpt_path if os.path.exists(ckpt_path): File "/usr/lib/python3.10/genericpath.py", line 19, in exists os.stat(path) TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType

Abdo Asker

Loading model from [options/SUPIR_v0_tiled.yaml] ./run_linux.sh: line 67: 4708 Killed python gradio_demo.py $lowvram $tiledVAE $theme $cpuMove --open_browser --share True

Abdo Asker

Okay, this is solved but there's another problem (below this comment)

SECourses

that means out of ram. restart pod try again. also how much ram your pod has? if fails again add this to arguments --fast_load_sd

Abdo Asker

another error: RuntimeError: CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero.

SECourses

i dont have telegram please message from discord

Cemil Hacimahmutoglu

hocam üyeliği aktif ettik ve şu anda windows_install.bat dosyasını tıkladım ve indirmeye başladı sürecek gibi görünüyor. Merak ettiğim bende 3080ti var localde çalıştırabilirim değilmi, ayrıca windows_install.bat osyasıa tıklamam yeterli olcak mı, yapmam gereken başka bir şey olacakmı

SECourses

evet çalıştırabilirsin. install bittikten sonra start_SUPIR_Windows calistir. fp8 seç ve tiling seç

Abdo Asker

what should I do then ? I use RTX A6000 and installed every thing as you have done in the video.

SECourses

start a new pod. runpod is cheap but has a lot of issues. hopefully i will introduce massed compute

Philip Heggie

I installed C++ and python 3.10.11 but running the windows install bat file python shows requirements.txt is missing and downloader.py is missing where are they how do i fix this install aborted

SECourses

hello did you download latest v44? can you message me from discord? show some screenshot?

SECourses

can you DM me on Twitter and show screenshots how you are trying to install. screenshot of your folder and install logs CMD logs

Philip Heggie

Yes but i found the problem git for windows wasn't installed on my system installing properly after that sorry for my ignorance.

SECourses

i am glad you solved. i will mention this more importantly hopefully in next video

Tuncay Akyazıt

11gb vramli 1080ti kartımla sorunsuz çalışıyor teşekkürler.

SECourses

harika teşekkürler geri bildirim için

rey

an option for Creativity Strength like krea ai please

Matt Zamaites

Thanks for this comment, fixed my issue!

SECourses

not very much yet but we may add that

SECourses

i made a new video. hopefully will be published today. please watch it once published

SECourses

i made a new video and mentioned this issue there :d hopefully will be published today

SECourses

this is something i plan to add. hopefully soon. thanks for comment. can you also share your settings me

rey

can we add the new Juggernaut XL v10 model?

SECourses

that means you didnt download models accurately

SECourses

yes i added it already. update version (run Windows_Update_Version) and in checkpoint downloader you will see it

puk

Thank you for this great tool. I've been using this tool since you introduced it and it just keeps getting better. I have experimented myself and used my own trained high quality sdxl loras with 1500 training images as a model. When it came to just the face it worked quite well. My question is, does it make sense to integrate an extra LoRA file in addition to the models as a second option in your workflow to achieve better results? I could of course create my own Dreambooth file. But these are much bigger than a Lora.

SECourses

i plan to add lora option too hopefully. currently you can merge your lora with your model to use. it should work fine

egormly

I am getting: gradio_demo.py", line 13, in import einops ModuleNotFoundError: No module named 'einops' I have confirmed einops is installed and running update or requirements says all is there, any ideas?

Flambo

Has the massive slowdown issues since v37 been fixed?

SECourses

this means your python version is not 3.10 when installing. make sure that your python is 3.10 and reinstall. i also just made update to the installer file to accurately select python 3.10 if you have multiple versions

SECourses

by the way if you dont use tiled vae it works even faster but it requires more vram. we are looking into problem

J

I'm noticing that on some full body photos the face restoration doesn't work correctly and parts of the face are pixelated in the final photo: https://imgsli.com/MjU3NjY0

SECourses

yes you are right. in some cases it sadly fails to detect image

Flambo

Yes but it also overloads a 4090 for anything bigger than a thumbnail due to vram usage, making it practically unusable for local upscaling compared to v37 which just works.

SECourses

i agree. this was done by d8ahazard and i keep asking him to fix. waiting him

Rick B

SUPIR_V47, used the Kaggle notebook, getting the following when running gradio /kaggle/temp/SUPIR * Serving Flask app '__main__' * Debug mode: off Traceback (most recent call last): File "/kaggle/temp/SUPIR/gradio_demo.py", line 2, in from asyncio.windows_events import NULL File "/opt/conda/lib/python3.10/asyncio/windows_events.py", line 6, in raise ImportError('win32 only') ImportError: win32 only Same error with somewhat earlier Kaggle notebook, not certain which zip file. Then the notebook stops running, so doen't run in kaggle suddenly?

Art

yes... same here...

Dmitry

The same mistake)

Hexadecimal22

I'm getting the same error. And I just downloaded the newest SUPIR_47.zip and installed it from the Windows_Install.bat , didn't change or do anything at all during the install. What am I doing wrong?

egormly

Had multiple versions, the updated script and also updating requirements fixed it, thanks.

SECourses

gonna fix it now hopefully thanks for mentioning

SECourses

should be fixed now. testing again.

SECourses

should be fixed now. testing again.

SECourses

should be fixed now. testing again.

SECourses

it is not your mistake. downloading huge files via CMD sometimes failing. please download all models single zip file and extract as shown in this video : https://youtu.be/OYxVEvDf284

Rick B

I'm sorry, I don't understand. What got fixed? Is there a new notebook to download or is there a new kaggle notebook in the same SUPIR_V47 zip file?

SECourses

the kaggle notebook auto downloads the remote script. i fixed the script. so same notebook now works

Chinmoy Basak

Commenting hear after Youtube: Could not create share link. Please check your internet connection or our status page: https://status.gradio.app. ### Facing this issue with runpod I tried with pytorch 2.1 and 2.0.1 Also tried to follow this following..... Please check your internet connection. This can happen if your antivirus software blocks the download of this file. You can install manually by following these steps: 1. Download this file: https://cdn-media.huggingface.co/frpc-gradio-0.2/frpc_linux_amd64 2. Rename the downloaded file to: frpc_linux_amd64_v0.2 3. Move the file to this location: /workspace/SUPIR/venv/lib/python3.10/site-packages/gradio

SECourses

could be temporary problem. where did you install? massed compute or runpod?

Chinmoy Basak

Thank you, runpod, tried with two different versions of pytorch, please suggest

SECourses

it is the error of gradio. gradio share is down at the moment. you can use massed compute . a good tutorial for massed compute here but i will also hopefully make tutorial for supir too : https://youtu.be/0t5l6CP9eBg

Chinmoy Basak

Thank you again , massed compute I found A6000[alt config] has 24GB RAM, will be enough for SUPIR? As you have mentioned we would need 30 GB. Do you think creating Docker would solve these kind of versioning issues

SECourses

no it wont be . there are 2 options on runpod. rent the one with bigger RAM : Specs: RAM: 48 GB, Storage: 256 GB, vCPU: 6

Nate

Getting "ModuleNotFoundError: No module named 'einops'", I think because: C:\SUPIR_v47\SUPIR>pip install -r requirements.txt Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu121 ERROR: triton-2.1.0-cp310-cp310-win_amd64.whl is not a supported wheel on this platform. Any ideas?

Nate

python --version Python 3.11.5 Is this OK? Do I really need 3.10?

SECourses

yes you need 3.10. if you install it our installer will automatically select it if you have python launcher installed

Nate

Dear god this is stupid: >python --version Python 3.10.11 >curl -sS https://bootstrap.pypa.io/get-pip.py | python Collecting pip Using cached pip-24.0-py3-none-any.whl.metadata (3.6 kB) Collecting setuptools Using cached setuptools-69.5.1-py3-none-any.whl.metadata (6.2 kB) Collecting wheel Using cached wheel-0.43.0-py3-none-any.whl.metadata (2.2 kB) Using cached pip-24.0-py3-none-any.whl (2.1 MB) Using cached setuptools-69.5.1-py3-none-any.whl (894 kB) Using cached wheel-0.43.0-py3-none-any.whl (65 kB) Installing collected packages: wheel, setuptools, pip Successfully installed pip-24.0 setuptools-69.5.1 wheel-0.43.0 >pip --version Traceback (most recent call last): File "runpy.py", line 196, in _run_module_as_main File "runpy.py", line 86, in _run_code File "C:\SUPIR_v47\python-3.10.11\Scripts\pip.exe\__main__.py", line 4, in ModuleNotFoundError: No module named 'pip' I hate all of this. I'm using your 1-click installer to avoid this terrible, terrible python hell. Maybe you can just package the right python with the installer and have it use that.

SECourses

it is easy relax. i have shown how to install everything step by step this tutorial. or else i can connect your pc and fix all if you become a gold member : https://youtu.be/-NjNy7afOQ0

Nate

I wanted to use portable Python 3.10 so I don't litter junk all over my OS. I used the Python 3.10 installer, deleted venv, and now I'm downloading everything again. If this doesn't work I swear I'm deleting it all and giving up. During this new install I see: Installing collected packages: torch, torchvision, torchaudio Attempting uninstall: torch Found existing installation: torch 2.1.0+cu121 Uninstalling torch-2.1.0+cu121: Successfully uninstalled torch-2.1.0+cu121 Attempting uninstall: torchvision Found existing installation: torchvision 0.16.0+cu121 Uninstalling torchvision-0.16.0+cu121: Successfully uninstalled torchvision-0.16.0+cu121 Attempting uninstall: torchaudio Found existing installation: torchaudio 2.1.0+cu121 Uninstalling torchaudio-2.1.0+cu121: Successfully uninstalled torchaudio-2.1.0+cu121 ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. xformers 0.0.22.post7 requires torch==2.1.0, but you have torch 2.2.0+cu121 which is incompatible. Successfully installed torch-2.2.0+cu121 torchaudio-2.2.0+cu121 torchvision-0.17.0+cu121 It's frustrating that everything is so fragile with Python. If 3.10 is required, the script should fail if Python is not 3.10. Better would be to include Python so the script can't fail this way. After all, a 1-click install should not have such gotchas. I've been a software engineer for 30 years. This is not a user problem, it's a Python problem. The setup is absolutely terrible. Most AI projects have instructions that are out of date. Most commonly dependencies aren't specified properly, so the instructions worked when they were written, but later newer versions are used which breaks pickle deserialization or otherwise have API breakage. Your 1-click setup is nice, if I can get past this Python 3.10 nonsense, but I imagine takes you a lot of effort to keep it working, since so much needs to be downloaded. I see that not every dependency has a version, which is just a bad idea and will inevitably break in the future. The only project that has worked well is Upscayl. It just worked out of the box. The custom models just work. Topaz Gigapixels also just worked, and is better than Upscayl. I want to try SUPIR to see how it compares, but man the pain is high.

SECourses

well i am also a software engineer since 2008 and you are absolutely right about python :D i hate it. i am a c# developer. giving a version to every dependency is even more painful. but i include pip freeze

Nate

When my firewall blocks communication (I enabled it for 30 minutes, but downloading took longer) the command fails, but the script keeps going, running more commands that try to download and fail. Then at the end it says everything was installed just fine, but it wasn't. Traceback (most recent call last): File "C:\Users\Nate\Desktop\SUPIR_v47\downloader.py", line 48, in download_file(file_url, folder, file_name) File "C:\Users\Nate\Desktop\SUPIR_v47\downloader.py", line 19, in download_file with requests.get(url, stream=True) as r: File "C:\Users\Nate\Desktop\SUPIR_v47\SUPIR\venv\lib\site-packages\requests\api.py", line 73, in get return request("get", url, params=params, **kwargs) File "C:\Users\Nate\Desktop\SUPIR_v47\SUPIR\venv\lib\site-packages\requests\api.py", line 59, in request return session.request(method=method, url=url, **kwargs) File "C:\Users\Nate\Desktop\SUPIR_v47\SUPIR\venv\lib\site-packages\requests\sessions.py", line 589, in request resp = self.send(prep, **send_kwargs) File "C:\Users\Nate\Desktop\SUPIR_v47\SUPIR\venv\lib\site-packages\requests\sessions.py", line 703, in send r = adapter.send(request, **kwargs) File "C:\Users\Nate\Desktop\SUPIR_v47\SUPIR\venv\lib\site-packages\requests\adapters.py", line 519, in send raise ConnectionError(e, request=request) requests.exceptions.ConnectionError: HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /RunDiffusion/Juggernaut-XL-v9/resolve/main/Juggernaut-XL_v9_RunDiffusionPhoto_v2.safetensors (Caused by NewConnectionError(': Failed to establish a new connection: [WinError 10013] An attempt was made to access a socket in a way forbidden by its access permissions')) Virtual environment made and installed properly

Nate

Thanks, I should have tried that to start. It seems better than scripts downloading so many things. My latest reinstall actually worked though, so I'm happy about that! Thanks for your help, I'd never find the patience to get this working otherwise!

Nate

I got SUPIR to work, cheers! I want to go from 4000x2192 to 18600x9900, which is 4.65x. SUPIR crashes trying that. I did get 2x to work and the results scaled up the rest of the way using bicubic are amazing, much better than Topaz Gigapixel. If it could do 4.65x it would be amazing! Any tips for getting 4.65x to work?

SECourses

for such mega pixel rent A6000 on Massed Compute, 48 GB GPU, try tiled and 8 bit. we have auto installers and instructions for massed compute

Nate

Oh, that sounds awesome! Can you point to the instructions I should follow? You've been a great help, I'll happily send a donation your way.

SECourses

the instructions and installers are inside SUPIR_v47.zip file. by also watching this video you can understand how to use Massed compute so easy : https://youtu.be/LeHfgq_lAXU

Nate

Damn this is cool! I get "python3: can't open file .../SUPIR/downloader.py". I can run SUPIR, but no models appear in the selectbox in the web UI. Any ideas?

Nate

Ah I didn't upload "downloader.py"!

Nate

I tried to push the limits using an H100 (128GB VRAM), first with full precision and no tiled, failed needing 80GB+ VRAM. Did again full precision and tiled, seemed to be working! The failed with: [Tiled VAE]: Executing Decoder Task Queue: 56%|... | 47063/84132 [02:13<1:14:35, 8.28it/s] ./Massed_Compute_Start_SUPIR.sh: line 67: 8101 Killed python3 gradio_demo.py $lowvram $tiledVAE $theme $cpuMove --open_browser --share True Is it weird is just stopped with "Killed" like that? Currently I'm running again with BF16/FP16 and tiled.

Nate

Damn, fails in exactly the same place, at 56%. I'm guessing it's the OS's out of memory killer. Trying again with 8bit, as you suggested.

Nate

It succeeded! I had to increase the swap size to 500GB. At least, 200GB was not enough. Maybe this helps someone else: sudo swapoff /swapfile sudo rm /swapfile sudo fallocate -l 500G /swapfile sudo chmod 600 /swapfile sudo mkswap /swapfile sudo swapon /swapfile Unfortunately 4.65x doesn't look as good as 2x. It hallucinated leaves where there should have been wood texture, and other problems. I will try 2x and then 2x again.

Nate

I did a year of gold. Thanks again for your help!

SECourses

wow man you are amazing. feel free to message me from discord any time for help. sorry that i just saw

SECourses

you can reduce hallucination effect by reducing text cfg

jonathan sanchez

for a 14900k and a rtx4090, what starting configuration I choose, it is not clear in the video.

SECourses

select 1-1-1-1 it will allow you to upscale into very high res if you need even bigger resolution select 2-1-1-1

Matt

Thanks for announcing supir v48, but where can I find SUPIR_v48.zip?

Matt

Got it - saw the change on github in gradio_demo.py - Thanks again!

SECourses

yes. zip file may remain same but we also do remote updates :)

SECourses

your best option to fix all frist 1-4 issues is finding the best model that can produce best results like your taste and doing dreambooth / fine tuning over it. then if you need LoRA you can extract it Reactor of Facefusion all uses a single same low faceswap quality so won't work. but fine tuned / dreambooth model can be used with SUPIR as a base model to restore face when upscaling

ivan

Hello. Thank you for the excellent automation of the installation process. You are stating that there is an update: "7 May 2024 Update Updated to V48", however in the attached files there is only V47. Can you please tell me where I can download V48.zip?

SECourses

The zip file same. It downloads from remote repo which is v48 👍

Z

Linux?

SECourses

Massed Compute is Ubuntu Linux should work

Andy Limited

how to fix this: ImportError: cannot import name 'packaging' from 'pkg_resources' (E:\Download\Compressed\SUPIR_v49\SUPIR\venv\lib\site-packages\pkg_resources\__init__.py)

SECourses

that means install error. please reinstall and send me all the output of the installation CMD. you can select all copy save into text file and email me : monstermmorpg@gmail.com

Christian G.

get the error: "C:\Users\gecko\AppData\Local\Programs\Python\Python310\python.exe: can't open file 'H:\\sd\\supir\\SUPIR\\gradio_demo.py': [Errno 2] No such file or directory" followed your instructions, what went wrong?

SECourses

As I replied in discord. You can't copy venv or rename parent folder after installed. You can only copy models

SECourses

ignore all messages it works. i just tested it like few days ago and updated

SECourses

i think it wouldnt work but you can try. give your folder path with --ckpt_dir

SECourses

show where did you download the folder path? also restart supir to see if works

SECourses

Well sadly without seeing what you do can't know. Please record screen and send me

Lou

I just wanted to say thanks for your 1 click Windows installers. In the past I've been more or less able to get most projects installed on my own but SUPIR was really difficult. I gave up after a few days and found your YouTube channel and subscribed to your Patreon. SUPIR was the first 1 Click Windows I tried and it worked flawlessly. I was so impressed that I'm now in the process of replacing my other AI programs with your 1 Click Installers, if available. Thanks for all your hard work and I look forward to any new Windows 1 Click Installers you make.

SECourses

I will try another model from hugging face and let you know. Currently not at home.

SECourses

updated zip file fixed issue. you have to move into folder before using wget - %cd /kaggle/temp/SUPIR/models/checkpoints

reaper557

Thank you very much for the all-in-one zip that has all the models. I was running into the exact problem of getting model download errors, and this is just what I need.

Ay**e Animation

EDIT: Nvm, I watched your video and it showed how to do it! Thank you very much! Hi! Thank you so much for this! Even on my 8GB VRAM 2070, it works very well! I thank you so much for your hard work of reducing the 48GB VRAM usage down to being within even an 8GB VRAM environment! By the way, I want to know how to get it to change the directory for Stable Diffusion checkpoints so I can save on space (since I already have SD models installed in another location)? Thank you!

SECourses

thank you so much it was a huge work

druhl

Makes it super easy to install and use! Kudos to you, Dr. Furkan.

Walt

thanks in Advance

SECourses

hello that means you had error during install. so i need you to verify install your cuda and python exactly as shown in this video and reinstall supir and send me your installation logs. if you also upgrade to gold member i can connect your pc and install for you : https://youtu.be/-NjNy7afOQ0

hacbachvotinh

6x256 [SAR 1:1 DAR 1:1], 907 kb/s, 8 fps, 8 tbr, 160k tbn (default) Metadata: creation_time : 2024-07-13T15:07:11.000000Z handler_name : Mainconcept MP4 Video Media Handler vendor_id : [0][0][0][0] encoder : AVC Coding Stream map '1:a:0' matches no streams. To ignore this, add a trailing '?' to the map. Failed to set value '1:a:0' for option 'map': Invalid argument Error parsing options for output file outputs_audio_restored.mp4. Error opening output files: Invalid argument Error restoring audio: Command '['ffmpeg', '-hwaccel', 'auto', '-i', 'outputs.mp4', '-i', 'D:\\temp\\gradio\\c283e9cdcad2354e07b4b1ffe87c0025122da96a\\CountrySide-256.mp4', '-ss', '0.0', '-to', '2.0', '-map', '0:v:0', '-map', '1:a:0', '-c:v', 'copy', '-shortest', 'outputs_audio_restored.mp4']' returned non-zero exit status 4294967274. Audio restoration failed: outputs.mp4 Video compiled successfully. Model moved to CPU All moved to CPU i got this error when upscale video.

SECourses

yes upscale video is still not working perfectly. it has the frames upscaled in the folder so you can merge later. but i would suggest you to extract all frames of the video into a folder and do a batch upscale

hacbachvotinh

okie i see! Tks for reply quicky.

meteor 1942

past couple of days local install (Win11) of SUPIR has gotten excruciatingly slow in loading the SUPIR yaml file the first time (tried different startup options too - didn't help). Note that GUI load works fine, image upscaling after the first time loading yaml file also works as expected. But the first SUPIR model load has gone from something like 200 seconds to 6000 seconds. Also tried brand new install with requirements update in a new env - didn't help either.

SECourses

looks like your harddrive is failing. you can monitor your hdd speed while model is getting loaded from task manager > performance > hdd

meteor 1942

Just checked the hdd - normal performance. Will check by installing on another hdd

SECourses

what reading speed do you see when loading?

Scott

Hello Furkan, thanks for all your hard work on this! I had some questions about the process your 1-click uses for SUPIR. I moved over to a ComfyUI workflow for it since it runs about twice as fast for me (about 4 minutes to double the size of a 1024x1024 image vs around 9 minutes in the 1-click). The results from it are *mostly* OK, but some results don't look quite as good (even tried with exact same seeds, prompts and steps and settings). From the ComfyUI workflow, I understand that SUPIR doesn't really upscale by itself, but a separate upscale is used first before it refines details to make the upscaled image look correct. My questions are: 1) What upscale are you using in the 1-click before sending to supir? I'm wondering if that may be a difference in the output I see. 2) I thought you mentioned somewhere that stage1 isn't necessary for many images. Is that correct or did I imagine it? If so, I can't quite figure out how to create the supir vae, denoised image, denoised latents that get passed from stage 1 into some other nodes like supir encode and super conditioner nodes. I tried a few things, but none worked and I can't seem to bypass stage 1, either. Any of your feedback or thoughts are helpful.

SECourses

Thanks for questions. We used the official implementation of SUPIR. I don't know at the moment which upscaler is used to be fair :D about second question Stage 1 was only implemented to get a better caption from llava. i asked this to the official developers. but we found it is pretty useless so we removed it entirely. about speed difference you are right. when tiled VAE is used currently APP is not working super fast due to some implementation d8ahazard did. sadly i dont know what exactly he did :/ we have 2 samplers and each one have different impacts : https://imgsli.com/MjU2ODQz/2/1

Scott

Thanks for the great information. When I use just a lanczos upscale my generations tend to be a bit softer than then 1click tool. If I add an upscaler like 4xLexicaDAT2_otf (followed by a lanczos, which may scale back down to get to correct canvas size) then it looks sharper, but a few times I've had it do stuff like add hairs in the whites of people's eyes and other weird things. I'm sure I could try different upscalers. I managed to bypass the "SUPIR 1st stage (denoiser)" node, but I clearly didn't pull the right things into the inputs of the SUPIR encode and conditioner nodes for stage 2. It created an image, but it was very dark and noisy without going through the stage 1 denoiser first. Thanks again.

SECourses

you are welcome. sadly i couldnt help you further. but our app is working great when you find accurate settings :D

Carlos L. Romero

Hi, thanks for sharing this, I wanted to try but gimme a gradio demo error and it wont start, should I download the 45gb file?

pis

SECourses

what is the error message? port in use is not important

SECourses

thank you just fixed the issue. refresh page, download SUPIR_v49_v4.zip file and use v4 kaggle notebook. start from fresh should work now. thank you

eyal giron

does it support Flux models?

楠 陈

Hope to support FLUX model

SECourses

i hope SUPIR authors add it so we can have too. that would be game changer

Hrabia M

hi! do U know, what's this issue? "Decoded output.: done in 0.173s, RAM: 9.57GB, VRAM: 10.10GB Wavelet reconstructed.: done in 0.018s, RAM: 9.57GB, VRAM: 10.10GB An exception occurred: 'Textbox' object is not subscriptable at Traceback (most recent call last): File "/home/rafis88/SUPIR/SUPIR/gradio_demo.py", line 783, in start_single_process _, result = batch_process(img_data, **values_dict) File "/home/rafis88/SUPIR/SUPIR/gradio_demo.py", line 1156, in batch_process last_result = supir_process(img_data, a_prompt, n_prompt, num_samples, upscale, edm_steps, s_stage1, s_stage2, File "/home/rafis88/SUPIR/SUPIR/gradio_demo.py", line 1066, in supir_process image_data = save_image(image_data, status_container.is_video) File "/home/rafis88/SUPIR/SUPIR/gradio_demo.py", line 1230, in save_image history_path = f'./history/{evt_id[:5]}/{evt_id[5:]}' TypeError: 'Textbox' object is not subscriptable"

SECourses

if you are using your own linux system you have to install all libraries accurately. python 3.10, cuda 11.8, c++ tools. the app works on massed compute, ubuntu system, you can rent a machine and compare with yours

James

Hi I'm getting this error but I have checked and einhops is installed: C:\AI\SUPIR_install\SUPIR\gradio_demo.py:1594: SyntaxWarning: invalid escape sequence '\S' placeholder="R:\SUPIR video\comparison_images") C:\AI\SUPIR_install\SUPIR\gradio_demo.py:1597: SyntaxWarning: invalid escape sequence '\S' placeholder="R:\SUPIR video\comparison_images\outputs") Traceback (most recent call last): File "C:\AI\SUPIR_install\SUPIR\gradio_demo.py", line 14, in import einops ModuleNotFoundError: No module named 'einops' Press any key to continue . . .

SECourses

what is your python version may i learn that?

James

Thanks for quick response its Python 3.12.3

James

I've just spotted the comments about using Python 3.10 elsewhere I guess I need to install that older version?

SECourses

yes you have to have 3.10 installed as well. install 3.10.11 and reinstall SUPIR

Kacper Gregorowicz

On Runpod can't run supir because "not found gradio module". I did everthing from runpod instructions. I choiced an options as you and not have a gradio link because not found a module. I downloaded SUPIR_v50 from your Patreon. What I'm doing wrong?

SECourses

please send me entire install logs as email : monstermmorpg@gmail.com

Scott

In case anyone else sees this, I managed to bypass Stage1 from the workflow. Apparently in Supir there is some kind of Stage1 for captioning or something and one for image processing. Removing the stage1 in my ComfyUI workflow just results in noisier images and I haven't found it useful. I also managed to find some settings in the ComfyUI workflow that make it quite similar to the Gradio version here such as using FluentlyXL 4 and tweaking cfg and some other settings. The main reason I wanted to use the Comfy version was it ran in about 4 minutes for a 1MP to 4MP (2x) upscale vs closer to 7 to 9 minutes for Gradio version. The upscaling time can be annoying on my 4060Ti 16GB if I have a lot of images. I also incorporated captioning by using Microsoft's Florence-2, which probably isn't as accurate as LLAVA, but is a lot smaller and I think is good enough without making my GPU fans go crazy and my computer go slow. If anyone is interested in my workflow I can clean it up and share it. I'm also going to try out ResShift and some others to see if they give almost as good results in less time. I have Topaz Photo AI, which is much faster than SUPIR but the results aren't as good so I typically use SUPIR to get to 4MP and then Topaz does ok enough to expand the photos after getting better details at 4 megapixels and going from 4MP to 10-16 megapixels.

James Osborn

Hi ya, I can't get this to run or complete the installer. I've downloaded the models.rar and put them in the models directory. When trying to install I get the message: ModuleNotFoundError: No module named 'huggingface_hub' When trying to run the program I get the message: ModuleNotFoundError: No module named 'einops' Thanks!

SECourses

hi do you install on windows? probably your python version is inaccurate you need python 3.10.11. give me more details. you must had some error during installation i need logs of them

James Osborn

Hey, thanks for the quick reply, I didn't have extended file names enabled in windows! Seems to be working now thanks.

Steve FX

Hi, i have the same error. I'm using python 3.10.11.

SECourses

hi give more info please. are you on windows? can you record a short video of how you install and error message? i can update installer according to that

SECourses

actually message me from discord i can connect your computer and check quickly

D Marcks

Hello, I've tried installing according to the instructions and video, but when i run the Windows Start BAT, I get the below error. Any ideas on how to fix? Traceback (most recent call last): File "C:\Users\*MYUSERNAME*\Supir_v51\SUPIR\gradio_demo.py", line 14, in import einops ModuleNotFoundError: No module named 'einops' Press any key to continue . . .

SECourses

the error is related to installation. currently working on a fix ty for sending logs

SECourses

just upgraded to v52 it should fix the issue. overwrite previous files with new zip file and update and reinstall please

SECourses

just upgraded to v52 it should fix the issue. overwrite previous files with new zip file and update and reinstall please

SECourses

just upgraded to v52 it should fix the issue. overwrite previous files with new zip file and update and reinstall please

D Marcks

That worked. Thank you.

Albert

Thanks

Bharat Soni

can't get test.py running python3 test.py --img_dir /root/input --save_dir /root/output Namespace(img_dir='/root/input', save_dir='/root/output', upscale=1, SUPIR_sign='Q', seed=1234, min_size=1024, edm_steps=50, s_stage1=-1, s_churn=5, s_noise=1.003, s_cfg=7.5, s_stage2=1.0, num_samples=1, a_prompt='Cinematic, High Contrast, highly detailed, taken using a Canon EOS R camera, hyper detailed photo - realistic maximum detail, 32k, Color Grading, ultra HD, extreme meticulous detailing, skin pore detailing, hyper sharpness, perfect without deformations.', n_prompt='painting, oil painting, illustration, drawing, art, sketch, oil painting, cartoon, CG Style, 3D render, unreal engine, blurring, dirty, messy, worst quality, low quality, frames, watermark, signature, jpeg artifacts, deformed, lowres, over-smooth', color_fix_type='Wavelet', linear_CFG=False, linear_s_stage2=False, spt_linear_CFG=1.0, spt_linear_s_stage2=0.0, ae_dtype='bf16', diff_dtype='fp16', no_llava=True) Loading model from [options/SUPIR_v0.yaml] SUPIR.models.SUPIR_model.SUPIRModel SUPIR.modules.SUPIR_v0.LightGLVUNet sgm.modules.diffusionmodules.wrappers.ControlWrapper sgm.modules.diffusionmodules.denoiser.DiscreteDenoiserWithControl sgm.modules.diffusionmodules.denoiser_weighting.EpsWeighting sgm.modules.diffusionmodules.denoiser_scaling.EpsScaling sgm.modules.diffusionmodules.discretizer.LegacyDDPMDiscretization DPMPP2M Traceback (most recent call last): File "/root/SUPIR/test.py", line 55, in model = create_SUPIR_model('options/SUPIR_v0.yaml', supir_sign=args.SUPIR_sign).to(SUPIR_device) File "/root/SUPIR/SUPIR/util.py", line 50, in create_SUPIR_model model = instantiate_from_config(config.model) File "/root/SUPIR/sgm/util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "/root/SUPIR/SUPIR/models/SUPIR_model.py", line 18, in __init__ super().__init__(*args, **kwargs) File "/root/SUPIR/sgm/models/diffusion.py", line 57, in __init__ instantiate_from_config(sampler_config) File "/root/SUPIR/sgm/util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "/root/SUPIR/sgm/util.py", line 180, in get_obj_from_str module, cls = string.rsplit(".", 1) ValueError: not enough values to unpack (expected 2, got 1)

SECourses

we dont use test.py anymore, so not supported, use the commands and instructions included in the zip file

SECourses

fp 8 + tiled should work up to a certain resolution but can't tell how much. 8 gb is very low

Brett Kelly

Yeah not working for me just stops during the tile process

SECourses

then you need to use cloud. i suggest massed compute 48 gb A6000 - 31 cents per hour

Tranquil

Hmm, it seems like I can't run this offline ? I turn off internet it this bug happen An exception occurred: at Traceback (most recent call last): File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\urllib\request.py", line 1348, in do_open h.request(req.get_method(), req.selector, req.data, headers, File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\http\client.py", line 1282, in request self._send_request(method, url, body, headers, encode_chunked) File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\http\client.py", line 1328, in _send_request self.endheaders(body, encode_chunked=encode_chunked) File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\http\client.py", line 1277, in endheaders self._send_output(message_body, encode_chunked=encode_chunked) File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\http\client.py", line 1037, in _send_output self.send(msg) File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\http\client.py", line 975, in send self.connect() File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\http\client.py", line 1447, in connect super().connect() File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\http\client.py", line 941, in connect self.sock = self._create_connection( File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\socket.py", line 824, in create_connection for res in getaddrinfo(host, port, 0, SOCK_STREAM): File "C:\Users\****\AppData\Local\Programs\Python\Python310\lib\socket.py", line 955, in getaddrinfo for res in _socket.getaddrinfo(host, port, family, type, proto, flags): socket.gaierror: [Errno 11001] getaddrinfo failed

SECourses

we dont do anything from online but i dont know maybe gradio requesting. you can search for solution

Tranquil

I just tested again and have realized that even if we don't use BG restore or Face restore, we must run it when we have internet to download missing files. After that, we can work offline again

Mike

Using this on Massed Compute, the app crashes. Watching NVITOP, I see the GPU is not being used, just RAM and SWAP fill up until the program crashes. What could I be doing wrong?

SECourses

did you get A6000 ALT or normal? ALT config has 24 GB RAM it may cause that

Mike

That fixed it, thanks

SECourses

yes i have seen CCSR + SUPIR combined is good but it would slow down even further :D maybe we can add it optionally

Dave

I see, yes, it will slow everything down. But as you mentioned, maybe you could make it optional with a checkbox, similar to what you did with LLaVa. I have another question: do you know how to use the batch processing feature on RunPod? For example, I created a folder inside "workspace" called "Images," and I put all the images I want to process in there. Inside "Images," I created another folder called "Edited," where all the edited images will be saved. What should I specify for the input and output folders for batch processing? I’ve tried using "workspace/Images," but it doesn’t seem to work. Thanks!

SECourses

input = /workspace/Images and output = /workspace/Images/Edited

Philipp Capetian

Hi. Great work! But unluckily the Kaggle Notebook keeps on crashing because it runs out on memory. I tried to upsacle 1024x1024 1,5x. I'm using 2xT4 - is that correct?

SECourses

yes it is limited. also get P100 it has 1 GB extra VRAM

Philipp Capetian

Ok! So the Supir Kaggle Notebook cannot use the dual T4s? P100 is only one GPU, isn't it? Has the choice of fp16 or bf16 or fp32, etc. vor the decoder an impact?

SECourses

kaggle doesnt support bf16, fp16 should work fine

Philipp Capetian

Ok, it works now. It takes quite looong. And I am stuck to 2x resize. I tried to split it into once 2x and once 1,5x, but of course it is rather limited by the image size per se. So unfortunately no 4k Background through Kaggle. Well, in the end it is free, so one cannot complain. I will now move on to Massed Compute

Robert Arsene

Hey! I often experience that the faces upscled and visible in restored faces tab is way better and detailed than what is shown then in the final upscaled image. I trid downloading and just paste them in Phtotoshop, but then there is always a size issue and does not fit the original image, What is causing SUPIR not to use those very detailed and sharp upscaled images of faces?

SECourses

sadly still at some images paste back not working. i couldnt fix that issue yet but will look at it hopefully soon

Ec Jep

I have tried updated to v52 and v54 but both keeping showing this error after installation, when trying to run with options 1, 1, 2, 1 Enter your choice (1-2): 1 no module 'xformers'. Processing without... no module 'xformers'. Processing without... Traceback (most recent call last): File "D:\SD\supir-V52\SUPIR\gradio_demo.py", line 24, in from SUPIR.models.SUPIR_model import SUPIRModel File "D:\SD\supir-V52\SUPIR\SUPIR\models\SUPIR_model.py", line 9, in from SUPIR.utils.tilevae import VAEHook File "D:\SD\supir-V52\SUPIR\SUPIR\utils\tilevae.py", line 76, in import xformers.ops File "D:\SD\supir-V52\SUPIR\venv\lib\site-packages\xformers\ops\__init__.py", line 8, in from .fmha import ( File "D:\SD\supir-V52\SUPIR\venv\lib\site-packages\xformers\ops\fmha\__init__.py", line 10, in from . import ( File "D:\SD\supir-V52\SUPIR\venv\lib\site-packages\xformers\ops\fmha\triton_splitk.py", line 110, in from ._triton.splitk_kernels import _fwd_kernel_splitK, _splitK_reduce File "D:\SD\supir-V52\SUPIR\venv\lib\site-packages\xformers\ops\fmha\_triton\splitk_kernels.py", line 632, in _fwd_kernel_splitK_autotune[num_groups] = autotune_kernel( File "D:\SD\supir-V52\SUPIR\venv\lib\site-packages\xformers\ops\fmha\_triton\splitk_kernels.py", line 614, in autotune_kernel kernel = triton.autotune( TypeError: autotune() got an unexpected keyword argument 'use_cuda_graph' Press any key to continue . . .

SECourses

hello what is your python? gpu model? for some reason your xformers installation failed. do you have your install logs? so we can see error reason

Ec Jep

python 3.10.11 and RTX4090. I am running the update requirements again so if it something changes and will let you know shortly. I have been using previous version of supir (ex v 47) just fine..

SECourses

ah please do a fresh install. you can move model files.

Ec Jep

fresh install with v55 and updated to cuda 12.4 and all working again (I move the models every time to avoid downloading). thx!

Peter Folvarcik

experiencing the same thing. faces in restored faces tab looking perfect but final image is just not using them. is there any workaround or something I can try or is it just the type of image on which it doesn't work yet? like images with more than one face?

pis (edited)

Comment edits

2026-09-07 10:00 Hello, I just downloaded the latest version and I get an error in the first cell and the execution process stops :(
2024-11-19 23:09

SECourses

hello. you are using on windows? what is the error you are getting please give me more info

4401 4401

Hi, just installed latest version on runpod, when I ran batch it went fine until Step 2 and got the error: Processing images (Stage 2) Loading model from [options/SUPIR_v0.yaml] An exception occurred: 401 Client Error. (Request ID: Root=1-673dddb3-5dbb995e36a488001d8f1c2a;9517ca1b-8919-4e0c-b042-346d281f395e) Repository Not Found for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main. Please make sure you specified the correct `repo_id` and `repo_type`. If you are trying to access a private or gated repo, make sure you are authenticated. Invalid username or password. at Traceback (most recent call last): File "/workspace/SUPIR/venv/lib/python3.11/site-packages/huggingface_hub/utils/_http.py", line 406, in hf_raise_for_status response.raise_for_status() File "/workspace/SUPIR/venv/lib/python3.11/site-packages/requests/models.py", line 1021, in raise_for_status raise HTTPError(http_error_msg, response=self) requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main

SECourses

hi you are using inaccurate template. i see it is python 3.11. so something during your installation failed because we download everything from our own repository. please use this template : RunPod Pytorch 2.2.0 runpod/pytorch:2.2.0-py3.10-cuda12.1.1

4401 4401

Hi, I restarted a new pod using that template and now getting this error.Processing 1 images... Processing images (Stage 2) Loading model from [options/SUPIR_v0.yaml] An exception occurred: 401 Client Error. (Request ID: Root=1-673f8091-6899551c161324b20996fd3c;9ed684cf-cd55-4667-ba4e-8280edeae2df) Repository Not Found for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main. Please make sure you specified the correct `repo_id` and `repo_type`. If you are trying to access a private or gated repo, make sure you are authenticated. Invalid username or password. at Traceback (most recent call last): File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/utils/_http.py", line 406, in hf_raise_for_status response.raise_for_status() File "/workspace/SUPIR/venv/lib/python3.10/site-packages/requests/models.py", line 1021, in raise_for_status raise HTTPError(http_error_msg, response=self) requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main The above exception was the direct cause of the following exception: Traceback (most recent call last): File "/workspace/SUPIR/gradio_demo.py", line 783, in start_single_process _, result = batch_process(img_data, **values_dict) File "/workspace/SUPIR/gradio_demo.py", line 1156, in batch_process last_result = supir_process(img_data, a_prompt, n_prompt, num_samples, upscale, edm_steps, s_stage1, s_stage2, File "/workspace/SUPIR/gradio_demo.py", line 904, in supir_process load_model(model_select, ckpt_select, diff_dtype, sampler, progress=progress) File "/workspace/SUPIR/gradio_demo.py", line 384, in load_model model = create_SUPIR_model(model_cfg, weight_dtype, supir_sign=selected_model[-1], device=device, ckpt=checkpoint_use, File "/workspace/SUPIR/SUPIR/util.py", line 50, in create_SUPIR_model model = instantiate_from_config(config.model) File "/workspace/SUPIR/sgm/util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "/workspace/SUPIR/SUPIR/models/SUPIR_model.py", line 18, in __init__ super().__init__(*args, **kwargs) File "/workspace/SUPIR/sgm/models/diffusion.py", line 61, in __init__ self.conditioner = instantiate_from_config( File "/workspace/SUPIR/sgm/util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "/workspace/SUPIR/sgm/modules/encoders/modules.py", line 89, in __init__ embedder = instantiate_from_config(embconfig) File "/workspace/SUPIR/sgm/util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "/workspace/SUPIR/sgm/modules/encoders/modules.py", line 457, in __init__ model_file = get_model(SDXL_CLIP1_PATH) File "/workspace/SUPIR/SUPIR/utils/model_fetch.py", line 12, in get_model snapshot_download(model_repo, local_dir=os.path.join(models_folder, model_folder), local_dir_use_symlinks=False) File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn return fn(*args, **kwargs) File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/_snapshot_download.py", line 232, in snapshot_download raise api_call_error File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/_snapshot_download.py", line 155, in snapshot_download repo_info = api.repo_info(repo_id=repo_id, repo_type=repo_type, revision=revision, token=token) File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn return fn(*args, **kwargs) File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/hf_api.py", line 2756, in repo_info return method( File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn return fn(*args, **kwargs) File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/hf_api.py", line 2541, in model_info hf_raise_for_status(r) File "/workspace/SUPIR/venv/lib/python3.10/site-packages/huggingface_hub/utils/_http.py", line 454, in hf_raise_for_status raise _format(RepositoryNotFoundError, message, response) from e huggingface_hub.errors.RepositoryNotFoundError: 401 Client Error. (Request ID: Root=1-673f8091-6899551c161324b20996fd3c;9ed684cf-cd55-4667-ba4e-8280edeae2df) Repository Not Found for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main. Please make sure you specified the correct `repo_id` and `repo_type`. If you are trying to access a private or gated repo, make sure you are authenticated. Invalid username or password.

daniel mendoza

I spent about 2 hours installing, until in the end it gave me an error: WARNING: Failed to find MSVC. PD: I already solved it by installing Windows SDK (10.0.26100) for Windows 11. You should put it within the requirements

pis (edited)

Comment edits

2026-09-07 10:00 Hi there, I just tried installing locally and to my surprise it worked, it took 30 minutes to a photo. I have a laptop with 6 VRAM. My question is the following: I don't know if I chose the correct options when running the AI, since I noticed that I got a CUDA error. Maybe it's not recognizing the GPU properly and that's why it took 30 minutes, or is it normal with a GPU with that amount of VRAM? I don't know, but I would like to know why that error appeared. I'll leave it here below. Use both VRAM optimizations and it uses around 12GB VRAM GPU FP8 optimization uses around 8GB VRAM If you have over 30 GB VRAM, you can start both full Params and no Tiled-VAE Modify this file and add --share if you want Gradio share Please select an option: 1. Start As BF16/FP16 - Uses Lesser VRAM Preferred 2. Start As FP8 - Uses More Lesser VRAM - Very Good For 12GB and Below GPUs 3. Start As Full Precision Enter your choice (1-3): 2 Please select an option: 1. Start Using Tiled VAE - Uses Lesser VRAM Preferred 2. Start Without Tiled VAE Enter your choice (1-2): 1 Please select an option: 1. Start With Auto Move To CPU - Useful When Using LLaVA - Moves model to CPU to save VRAM for LLaVA 2. Keep Models Always In GPU Enter your choice (1-2): 2 Please select an option: 1. Start With Light Theme 2. Start With Dark Theme Enter your choice (1-2): 2 WARNING: Failed to find CUDA. Running on local URL: http://127.0.0.1:7860
2024-12-05 02:11

Neon Harmonic

Is there any way to use the app via an API call for automation?

SECourses

You can start gradio and user gradio api feature. other than that nope

SECourses

sadly 6 GB not enough. use either paid services like massed compute or runpod or free kaggle notebook

SECourses

i dont know your error exactly but nevertheless 6gb wont work :D paid platforms as far as i know doesnt care as long as you dont do anything illegal

Art

Hi Furkan... tried ti run it on kaggle (version 5.6) but alìt the fourth step it gives me the error: /// /kaggle/temp/SUPIR Traceback (most recent call last): File "/kaggle/temp/SUPIR/gradio_demo.py", line 24, in from SUPIR.models.SUPIR_model import SUPIRModel File "/kaggle/temp/SUPIR/SUPIR/models/SUPIR_model.py", line 5, in from pytorch_lightning import seed_everything File "/opt/conda/lib/python3.10/site-packages/pytorch_lightning/__init__.py", line 27, in from pytorch_lightning.callbacks import Callback # noqa: E402 File "/opt/conda/lib/python3.10/site-packages/pytorch_lightning/callbacks/__init__.py", line 14, in from pytorch_lightning.callbacks.batch_size_finder import BatchSizeFinder File "/opt/conda/lib/python3.10/site-packages/pytorch_lightning/callbacks/batch_size_finder.py", line 26, in from pytorch_lightning.callbacks.callback import Callback File "/opt/conda/lib/python3.10/site-packages/pytorch_lightning/callbacks/callback.py", line 22, in from pytorch_lightning.utilities.types import STEP_OUTPUT File "/opt/conda/lib/python3.10/site-packages/pytorch_lightning/utilities/types.py", line 42, in from torchmetrics import Metric File "/opt/conda/lib/python3.10/site-packages/torchmetrics/__init__.py", line 37, in from torchmetrics import functional # noqa: E402 File "/opt/conda/lib/python3.10/site-packages/torchmetrics/functional/__init__.py", line 14, in from torchmetrics.functional.audio._deprecated import _permutation_invariant_training as permutation_invariant_training File "/opt/conda/lib/python3.10/site-packages/torchmetrics/functional/audio/__init__.py", line 36, in if _SCIPI_AVAILABLE: File "/opt/conda/lib/python3.10/site-packages/lightning_utilities/core/imports.py", line 198, in __bool__ self._check_available() File "/opt/conda/lib/python3.10/site-packages/lightning_utilities/core/imports.py", line 164, in _check_available self._check_requirement() File "/opt/conda/lib/python3.10/site-packages/lightning_utilities/core/imports.py", line 130, in _check_requirement pkg_version = Version(_version(req.name)) File "/opt/conda/lib/python3.10/site-packages/packaging/version.py", line 264, in __init__ match = self._regex.search(version) TypeError: expected string or bytes-like object

SECourses

doing a fresh install now to test

Michael Liu

hi Furkan when i run supir it give me this error message Repository Not Found for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main. Please make sure you specified the correct `repo_id` and `repo_type`. If you are trying to access a private or gated repo, make sure you are authenticated. Invalid username or password. at Traceback (most recent call last): File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\utils\_http.py", line 406, in hf_raise_for_status response.raise_for_status() File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\requests\models.py", line 1021, in raise_for_status raise HTTPError(http_error_msg, response=self) requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main The above exception was the direct cause of the following exception: Traceback (most recent call last): File "C:\SUPIR_v56\SUPIR\gradio_demo.py", line 783, in start_single_process _, result = batch_process(img_data, **values_dict) File "C:\SUPIR_v56\SUPIR\gradio_demo.py", line 1156, in batch_process last_result = supir_process(img_data, a_prompt, n_prompt, num_samples, upscale, edm_steps, s_stage1, s_stage2, File "C:\SUPIR_v56\SUPIR\gradio_demo.py", line 904, in supir_process load_model(model_select, ckpt_select, diff_dtype, sampler, progress=progress) File "C:\SUPIR_v56\SUPIR\gradio_demo.py", line 384, in load_model model = create_SUPIR_model(model_cfg, weight_dtype, supir_sign=selected_model[-1], device=device, ckpt=checkpoint_use, File "C:\SUPIR_v56\SUPIR\SUPIR\util.py", line 50, in create_SUPIR_model model = instantiate_from_config(config.model) File "C:\SUPIR_v56\SUPIR\sgm\util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "C:\SUPIR_v56\SUPIR\SUPIR\models\SUPIR_model.py", line 18, in __init__ super().__init__(*args, **kwargs) File "C:\SUPIR_v56\SUPIR\sgm\models\diffusion.py", line 61, in __init__ self.conditioner = instantiate_from_config( File "C:\SUPIR_v56\SUPIR\sgm\util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "C:\SUPIR_v56\SUPIR\sgm\modules\encoders\modules.py", line 89, in __init__ embedder = instantiate_from_config(embconfig) File "C:\SUPIR_v56\SUPIR\sgm\util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "C:\SUPIR_v56\SUPIR\sgm\modules\encoders\modules.py", line 457, in __init__ model_file = get_model(SDXL_CLIP1_PATH) File "C:\SUPIR_v56\SUPIR\SUPIR\utils\model_fetch.py", line 12, in get_model snapshot_download(model_repo, local_dir=os.path.join(models_folder, model_folder), local_dir_use_symlinks=False) File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\utils\_validators.py", line 114, in _inner_fn return fn(*args, **kwargs) File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\_snapshot_download.py", line 232, in snapshot_download raise api_call_error File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\_snapshot_download.py", line 155, in snapshot_download repo_info = api.repo_info(repo_id=repo_id, repo_type=repo_type, revision=revision, token=token) File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\utils\_validators.py", line 114, in _inner_fn return fn(*args, **kwargs) File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\hf_api.py", line 2748, in repo_info return method( File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\utils\_validators.py", line 114, in _inner_fn return fn(*args, **kwargs) File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\hf_api.py", line 2533, in model_info hf_raise_for_status(r) File "C:\SUPIR_v56\SUPIR\venv\lib\site-packages\huggingface_hub\utils\_http.py", line 454, in hf_raise_for_status raise _format(RepositoryNotFoundError, message, response) from e huggingface_hub.errors.RepositoryNotFoundError: 401 Client Error. (Request ID: Root=1-67659ec4-14fd49f2179d7b825f458b3a;9933cb22-ed1b-44b8-9963-03f93f7f2bd5) Repository Not Found for url: https://huggingface.co/api/models/models/clip-vit-large-patch14/revision/main. Please make sure you specified the correct `repo_id` and `repo_type`. If you are trying to access a private or gated repo, make sure you are authenticated. Invalid username or password.

SECourses

hi your model downloads failed. please rerun Windows_Resume_Fix_Model_Downloads.bat file

Damjan Žakelj

Hey, just installed the new version of SUPIR and after I start making my first upscale, I'm stuck here for more than 10 minutes, original image size is 2 MB, upscaled to 1,7 (6,8 MB):This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from Terminal to deploy to Spaces (https://huggingface.co/spaces) Processing 1 images... Processing LLaVA You are using a model of type llava to instantiate a model of type llava_supir. This is not supported for all configurations of models and can yield errors. Loading vision tower: openai/clip-vit-large-patch14-336 Loading checkpoint shards: 0%| | 0/2 [00:00

Michael Liu

Thanks! After downloading tons of model, it works now

SECourses

awesome and yes it needs a lot of models

SECourses

hi your disk is very slow or you dont have sufficient RAM. can you list me both? how much RAM you have? did you set virtual RAM? what is your disk?

s s

When using Batch Input Folder to process multiple photos, turn on BG restoration and Face restoration, Restored Faces only shows one photo with restored face. I hope to be able to display all the restored face photos in batch processing

SECourses

yes for some faces face restore doesnt work sadly.

Neil Rhodes

Is it important to up date PIP in 2025 mine is still 23.0.1

SECourses

if working not important. mine is : pip 24.3.1 from C:\Python310\lib\site-packages\pip (python 3.10)

Neil Rhodes

thanks i was updating to v58 and had to reinstall, it broke and said i dont have python installed. I did but i tried to uninstall Python and resintall python 3.10 but when i run SUPIR 58 it keeps complaing no Python? What can i check to get it working again, thanks, i tried the updates requirements file and now reports 'pip' is not recognized as an internal or external command, operable program or batch file. What has gone wrong it was working earlier For the record my paths broke to PIP and Python, trying to fix paths broke too. So installed a new Python and used that , massive learning troubleshooting curve

James

Hi I had it working fine on previous version updated to this new one and now it fails with error before it even gets started: ERROR:asyncio:Exception in callback _ProactorBasePipeTransport._call_connection_lost(None) handle: Traceback (most recent call last): File "C:\Users\theVR\AppData\Local\Programs\Python\Python310\lib\asyncio\events.py", line 80, in _run self._context.run(self._callback, *self._args) File "C:\Users\theVR\AppData\Local\Programs\Python\Python310\lib\asyncio\proactor_events.py", line 162, in _call_connection_lost self._sock.shutdown(socket.SHUT_RDWR) ConnectionResetError: [WinError 10054] An existing connection was forcibly closed by the remote host

SECourses

hi this looks like network / browser / Gradio related error. therefore, looks like your system related. also you are using AppData\Local\Programs\Python\Python310 which i dont recommend. so please install python and everything like this and reinstall supir : https://youtu.be/DrhUHnYfwC0

3D_Jedi

just installed latest windows version on Nvidia 4070 Super Ti and receiving this error? To create a public link, set `share=True` in `launch()`. IMPORTANT: You are using gradio version 4.19.0, however version 4.44.1 is available, please upgrade. -------- An exception occurred: stat: path should be string, bytes, os.PathLike or integer, not NoneType at Traceback (most recent call last): File "C:\SUPIR\SUPIR\gradio_demo.py", line 783, in start_single_process _, result = batch_process(img_data, **values_dict) File "C:\SUPIR\SUPIR\gradio_demo.py", line 1096, in batch_process ckpt_select = get_ckpt_path(ckpt_select) File "C:\SUPIR\SUPIR\gradio_demo.py", line 288, in get_ckpt_path if os.path.exists(ckpt_path): File "C:\Kohya-Python\lib\genericpath.py", line 19, in exists os.stat(path) TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType Exception in callback _ProactorBasePipeTransport._call_connection_lost(None) handle: Traceback (most recent call last): File "C:\Kohya-Python\lib\asyncio\events.py", line 80, in _run self._context.run(self._callback, *self._args) File "C:\Kohya-Python\lib\asyncio\proactor_events.py", line 165, in _call_connection_lost self._sock.shutdown(socket.SHUT_RDWR) ConnectionResetError: [WinError 10054] An existing connection was forcibly closed by the remote host Exception in callback _ProactorBasePipeTransport._call_connection_lost(None) handle: Traceback (most recent call last): File "C:\Kohya-Python\lib\asyncio\events.py", line 80, in _run self._context.run(self._callback, *self._args) File "C:\Kohya-Python\lib\asyncio\proactor_events.py", line 165, in _call_connection_lost self._sock.shutdown(socket.SHUT_RDWR) ConnectionResetError: [WinError 10054] An existing connection was forcibly closed by the remote host An exception occurred: stat: path should be string, bytes, os.PathLike or integer, not NoneType at Traceback (most recent call last): File "C:\SUPIR\SUPIR\gradio_demo.py", line 832, in start_batch_process result, _ = batch_process(img_data, **values_dict) File "C:\SUPIR\SUPIR\gradio_demo.py", line 1096, in batch_process ckpt_select = get_ckpt_path(ckpt_select) File "C:\SUPIR\SUPIR\gradio_demo.py", line 288, in get_ckpt_path if os.path.exists(ckpt_path): File "C:\Kohya-Python\lib\genericpath.py", line 19, in exists os.stat(path) TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType

SECourses

please make a fresh install into c:\supir_ai currently you have totally unrelated messages C:\SUPIR\SUPIR\ C:\Kohya-Python\lib\

SECourses

hi model downloads failed. run installer again. likely your pod is broken

Hoàng Giang Sơn Trương

import flash_attn_2_cuda as flash_attn_gpu ImportError: DLL load failed while importing flash_attn_2_cuda: The specified procedure could not be found.

SECourses

sorry about this mistake. please make a fresh install of V65 or delete your VENV folder and use new installer from V65.

Adrian Kajzer

When I change the settings of Supir Options, it only works on one setting Model moved to CPU All moved to CPU Processing 1 images... Processing images (Stage 2) Processing image 1/1... Upscaling image (pre)... Final upscale factor: 2.50 Batchifying sample... Instantiating sampler. An exception occurred: BaseDiffusionSampler.__init__() got an unexpected keyword argument 'restore_cfg' at Traceback (most recent call last): File "/workspace/SUPIR/gradio_demo.py", line 951, in start_single_process _, result = batch_process(img_data, **values_dict) File "/workspace/SUPIR/gradio_demo.py", line 1493, in batch_process last_result = supir_process(img_data, a_prompt, n_prompt, num_samples, upscale, edm_steps, s_stage1, s_stage2, File "/workspace/SUPIR/gradio_demo.py", line 1327, in supir_process samples = process_sample(model, lq, caption) File "/workspace/SUPIR/gradio_demo.py", line 1252, in process_sample samples = model.batchify_sample(input_data, caption, num_steps=edm_steps, restoration_scale=s_stage1, File "/workspace/SUPIR/venv/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context return func(*args, **kwargs) File "/workspace/SUPIR/SUPIR/models/SUPIR_model.py", line 154, in batchify_sample self.sampler = instantiate_from_config(self.sampler_config) File "/workspace/SUPIR/sgm/util.py", line 175, in instantiate_from_config return get_obj_from_str(config["target"])(**config.get("params", dict())) File "/workspace/SUPIR/sgm/modules/diffusionmodules/sampling.py", line 722, in __init__ super().__init__(*args, **kwargs) File "/workspace/SUPIR/sgm/modules/diffusionmodules/sampling.py", line 425, in __init__ super().__init__(*args, **kwargs) TypeError: BaseDiffusionSampler.__init__() got an unexpected keyword argument 'restore_cfg'

SECourses

let me test all configs and fix. thanks for letting me know

SECourses

please update. v65 fixed this issue. sorry for that. just run update bat

Adrian Kajzer

I'm using RunPod, the EDM sampler is already working fine, but there is a problem with the sampler DPMPP2M Processing 1 images... Processing images (Stage 2) Processing image 1/1... Upscaling image (pre)... Final upscale factor: 2.50 Instantiating sampler. Instantiated sampler. Seed set to 1369342914 Encoding first stage with denoise... Encoded first stage with denoise... Decoded first stage... Encoded first stage... Loading denoiser. Loaded denoiser. Sampling... TiledRestoreDPMPP2MSampler An exception occurred: DPMPP2MSampler.sampler_step() got an unexpected keyword argument 'eps_noise' at Traceback (most recent call last): File "/workspace/SUPIR/gradio_demo.py", line 951, in start_single_process _, result = batch_process(img_data, **values_dict) File "/workspace/SUPIR/gradio_demo.py", line 1493, in batch_process last_result = supir_process(img_data, a_prompt, n_prompt, num_samples, upscale, edm_steps, s_stage1, s_stage2, File "/workspace/SUPIR/gradio_demo.py", line 1286, in supir_process samples = process_sample(model, face, [caption], face_resolution, is_face=True) File "/workspace/SUPIR/gradio_demo.py", line 1252, in process_sample samples = model.batchify_sample(input_data, caption, num_steps=edm_steps, restoration_scale=s_stage1, File "/workspace/SUPIR/venv/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context return func(*args, **kwargs) File "/workspace/SUPIR/SUPIR/models/SUPIR_model.py", line 226, in batchify_sample _samples = self.sampler(denoiser, noised_z, cond=c, uc=uc, x_center=z_stage1, control_scale=control_scale, File "/workspace/SUPIR/sgm/modules/diffusionmodules/sampling.py", line 500, in __call__ _x, _old_denoised = self.sampler_step( TypeError: DPMPP2MSampler.sampler_step() got an unexpected keyword argument 'eps_noise'

SECourses

thank you so much this was an important bug. just fixed with v66. run your update bat file

Glen Carpenter

EDIT - I did a complete wipe of Python 3.10, FFmpeg, CUDA 12.8 or above, cuDNN 9.4 or above, C++ Tools, MSVC and re-installed everything. I am now running this with CUDA 12.4, it is back to being fast. Will probably try with CUDA 12.8 later, but the root cause seemed to be the multiple versions of CUDA on my system. I had CUDA 11.8, 12.1, 12.4, and 12.8 all installed, there was potentially conflicts. I am having some issues with v6.8 It seems there is an incompatibility with the current version of `xformers`in requirements.txt I did follow the instructions for installing prerequisites and updating PATH variables Windows_Install.bat: ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. xformers 0.0.28.post3 requires torch==2.5.1, but you have torch 2.7.0.dev20250311+cu128 which is incompatible. Windows_Direct_Start_No_Options.bat: Traceback (most recent call last): File "D:\repos\SUPIR_v68\SUPIR\gradio_demo.py", line 23, in import pillow_avif # Import the AVIF plugin ModuleNotFoundError: No module named 'pillow_avif' Press any key to continue . . . I did get it to run by manually installing with pip install pillow-avif-plugin But it is SLOOOW - 10 minutes for a 450x450 image a 1x upscale with an RTX 4090

SECourses

something wrong with you. 450x450 to 1 mega pixel takes few seconds on my rtx 5090. i am fixing AVIF thing also thanks for reporting.

SECourses

i am fixing Windows_Update_Requirements.bat as well thanks for reporting

io io

V69. I get something it look the same during install : ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. xformers 0.0.28.post3 requires torch==2.5.1, but you have torch 2.7.0.dev20250311+cu128 which is incompatible.

Neil Rhodes

i updated to latest veriosn today using Windows Update requirements this is a wonderful app and i use it daily but... Traceback (most recent call last): File "A:\SUPIR_v58\SUPIR\gradio_demo.py", line 23, in import pillow_avif # Import the AVIF plugin ModuleNotFoundError: No module named 'pillow_avif' Press any key to continue . . . how to fix?

SECourses

do a git pull and install again. it should fix it. i updated requirements.

SECourses

Run windows update and then run windows install again it should fix without redownload

Neil Rhodes

pull what from git? sorry its been a while since i did anything with gitpull. I ran the windows up and windows install, nothing changed. What do i need to do wth git please remind me, thanks

SECourses

that should fix it. if you still couldnt fix make a fresh install and move models :D

SECourses

yes then it will fix it itself :D

Neil Rhodes

I reinstalled and now the resulting image looks like a brown granular blurred depth map every time

SECourses

something seems wrong. which image you are trying can you email me? did you try other images? default config? monstermmorpg@gmail.com

James Bond

Hi! I have been trying to install SUPIR locally using v6.9. Getting some errors in installation process, had to abandon the install. See a copy of one of the error messages below: "ERROR: Exception: Traceback (most recent call last): File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_vendor\urllib3\response.py", line 438, in _error_catcher yield File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_vendor\urllib3\response.py", line 561, in read data = self._fp_read(amt) if not fp_closed else b"" File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_vendor\urllib3\response.py", line 527, in _fp_read return self._fp.read(amt) if amt is not None else self._fp.read() File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_vendor\cachecontrol\filewrapper.py", line 98, in read data: bytes = self.__fp.read(amt) File "C:\Python310\lib\http\client.py", line 466, in read s = self.fp.read(amt) File "C:\Python310\lib\socket.py", line 705, in readinto return self._sock.recv_into(b) File "C:\Python310\lib\ssl.py", line 1274, in recv_into return self.read(nbytes, buffer) File "C:\Python310\lib\ssl.py", line 1130, in read return self._sslobj.read(len, buffer) TimeoutError: The read operation timed out During handling of the above exception, another exception occurred: Traceback (most recent call last): File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\cli\base_command.py", line 106, in _run_wrapper status = _inner_run() File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\cli\base_command.py", line 97, in _inner_run return self.run(options, args) File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\cli\req_command.py", line 67, in wrapper return func(self, options, args) File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\commands\install.py", line 386, in run requirement_set = resolver.resolve( File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\resolution\resolvelib\resolver.py", line 179, in resolve self.factory.preparer.prepare_linked_requirements_more(reqs) File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\operations\prepare.py", line 554, in prepare_linked_requirements_more self._complete_partial_requirements( File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\operations\prepare.py", line 469, in _complete_partial_requirements for link, (filepath, _) in batch_download: File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\network\download.py", line 184, in __call__ for chunk in chunks: File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\cli\progress_bars.py", line 55, in _rich_progress_bar for chunk in iterable: File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_internal\network\utils.py", line 65, in response_chunks for chunk in response.raw.stream( File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_vendor\urllib3\response.py", line 622, in stream data = self.read(amt=amt, decode_content=decode_content) File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_vendor\urllib3\response.py", line 560, in read with self._error_catcher(): File "C:\Python310\lib\contextlib.py", line 153, in __exit__ self.gen.throw(typ, value, traceback) File "D:\SUPIR_v69\SUPIR\venv\lib\site-packages\pip\_vendor\urllib3\response.py", line 443, in _error_catcher raise ReadTimeoutError(self._pool, None, "Read timed out.") pip._vendor.urllib3.exceptions.ReadTimeoutError: HTTPSConnectionPool(host='download.pytorch.org', port=443): Read timed out."

SECourses

hi it is related to your internet but i just updated to official torch version. so try v70. also if you still get error turn off your VPN if you have. if you dont have VPN try Warp VPN from Cloudlfare works amazing : https://youtu.be/0RSaYlmmblc

James Bond

Hi and thank you for your reply. Yes, that was related to my connection, my computer power settings caused PC to 'sleep' during the slow install (my internet is not as fast as yours, :)). That error did not repeat when I reinstalled SUPIR after changing PC power settings, but another error cameup and I am not sure how to fix that issue. Below, I am quoting the error message, would appreciate your help. " ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. xformers 0.0.28.post3 requires torch==2.5.1, but you have torch 2.7.0.dev20250311+cu128 which is incompatible. " Is it because CUDA12.8 is presently my default? (I have CUDA 11.8 and 12.8, installed as per your post and video tutorial advice)

SECourses

glad you solved. now i recommend this. delete venv and install v70 installer and pay attention to install logs that none fails. it should fix

Error_404_unknown

Hello Dr Furkan, really sorry I just reinstalled the new version of SUPIR after not using for a long time. It seems to be saving the upscaled images as small squares in the top left corner of the original image. Not sure how to stop it doing this and just save the final image?

SECourses

it may happen with face restore. please send me your image. i guess it happens with very low resolution images but not sure i have to debug. email me : monstermmorpg@gmail.com

Fernando Martin

Hi dear Furkan, i tried the latest version and when i try any image i get this error: File "C:\Users\Baldur\Downloads\SUPIR_v71\SUPIR\venv\lib\site-packages\transformers\tokenization_utils_base.py", line 2032, in from_pretrained raise EnvironmentError( OSError: Can't load tokenizer for 'C:\Users\Baldur\Downloads\SUPIR_v71\SUPIR\models\clip-vit-large-patch14'. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure 'C:\Users\Baldur\Downloads\SUPIR_v71\SUPIR\models\clip-vit-large-patch14' is the correct path to a directory containing all relevant files for a CLIPTokenizer tokenizer. How can i fix it? i know its about the token from Hugging face, but dont know where to put it in the app to get it work

SECourses

hi easy. run downloader again. some models failed to download for some reason

Hipno

EDIT: IT WORKS!!! I didn't read the beginning where you said to download the models from the zip file. Thank you so much!! Hey Dr, https://www.patreon.com/posts/99176057 Supir isn't working for me, i'm using Massed compute and have followed the instructions but get this error. It looks like the models aren't downloading so i've tried fresh installing 2 more times but same issue. An exception occurred: stat: path should be string, bytes, os.PathLike or integer, not NoneType at Traceback (most recent call last): File "/home/Ubuntu/Downloads/SUPIR_v71/SUPIR/gradio_demo.py", line 1242, in start_singleprocess , result = batch_process(img_data, **values_dict) File "/home/Ubuntu/Downloads/SUPIR_v71/SUPIR/gradio_demo.py", line 1752, in batch_process ckpt_select = get_ckpt_path(ckpt_select) File "/home/Ubuntu/Downloads/SUPIR_v71/SUPIR/gradio_demo.py", line 632, in get_ckpt_path if os.path.exists(ckpt_path): File "/home/Ubuntu/.pyenv/versions/3.10.16/lib/python3.10/genericpath.py", line 19, in exists os.stat(path) TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType Message #errors-and-solutions

Neil Rhodes

upgraded to 5090 thought i did the 5090 updates ok, what should i do now errors, ComfyUI execution error: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

SECourses

yes delete everything and use latest installer and install fresh. make sure to have cuda 12.8 installed

Eclipse Beast - TF

I get an error while using the dpmpp2m sampler. TiledRestoreDPMPP2MSampler An exception occurred: Boolean value of Tensor with more than one value is ambiguous at Traceback (most recent call last): File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\gradio_demo.py", line 1242, in start_single_process _, result = batch_process(img_data, **values_dict) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\gradio_demo.py", line 1822, in batch_process last_result = supir_process(img_data, a_prompt, n_prompt, num_samples, upscale, edm_steps, s_stage1, s_stage2, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\gradio_demo.py", line 1656, in supir_process samples = process_sample(model, lq, caption) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\gradio_demo.py", line 1555, in process_sample samples = model.batchify_sample(input_data, caption, num_steps=edm_steps, restoration_scale=s_stage1, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\venv\Lib\site-packages\torch\utils\_contextlib.py", line 116, in decorate_context return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\SUPIR\models\SUPIR_model.py", line 226, in batchify_sample _samples = self.sampler(denoiser, noised_z, cond=c, uc=uc, x_center=z_stage1, control_scale=control_scale, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\sgm\modules\diffusionmodules\sampling.py", line 817, in __call__ eps_noise = noise_sampler(s_in * sigmas[i], s_in * sigmas[i + 1]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\venv\Lib\site-packages\k_diffusion\sampling.py", line 114, in __call__ return self.tree(t0, t1) / (t1 - t0).abs().sqrt() ^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\venv\Lib\site-packages\k_diffusion\sampling.py", line 87, in __call__ t0, t1, sign = self.sort(t0, t1) ^^^^^^^^^^^^^^^^^ File "S:\AI\Programy\Upscare\SUPIR_v71\SUPIR\venv\Lib\site-packages\k_diffusion\sampling.py", line 84, in sort return (a, b, 1) if a < b else (b, a, -1) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ RuntimeError: Boolean value of Tensor with more than one value is ambiguous

ApathyJones

Seems to be a gradio error with the latest version when trying to load a video

SECourses

yes video not working. i am working on a way better video upscaler.

SECourses

yes i guess best to stick other one until we fix it. ty for report

Eclipse Beast - TF

Another one? Is there a way to somehow add others? Because there are only two — this one and the other one that sticks strictly to the photo. Plus, I have another issue: when I upload any other model from Civia, for example a pony one, the whole program crashes. What it outputs after upscaling is blurry and broken, while the models you’ve already added seem to work.

SECourses

sadly we have only these 2 atm. but our default config working really really amazingly good

Max Gonzales

Why do I keep getting this error when trying to start supir with my rtx 5090? The procedure entry point could not be located in the dynamic link library D:\Dropbox\Resources\AI\SUPIR_v73\SUPIR\venv\Lib\site-pack ages\xformers\_C.pyd. ^ error dialogue Please select an option: 1. Start Using Tiled VAE - Uses Lesser VRAM Preferred 2. Start Without Tiled VAE Enter your choice (1-2): 2 Please select an option: 1. Start With Auto Move To CPU - Useful When Using LLaVA - Moves model to CPU to save VRAM for LLaVA 2. Keep Models Always In GPU Please select an option: 1. Start With Light Theme 2. Start With Dark Theme Enter your choice (1-2): 2 W0527 10:12:16.530000 2772 venv\Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects . py : 29] NOTE: Redir ects are currently not supported in Windows or MacOs. WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for: PyTorch 2.7.0. dev20250311+cu128 with CUDA 1208 (you have 2.7.0+cu128) Python 3.10.11 (you have 3.10.11) Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers) Memory-efficient attention, SwiGLU, sparse and more won't be available. Set XFORMERS_MORE_DETAILS=1 for more details Running on local URL: http://127.0.0.1:7860 To create a public link, set 'share=True' in 'launch()'. IMPORTANT: You are using gradio version 4.19.0, however version 4.44.1 is available, please upgrade. Enter your choice (1-3): 3 Enter your choice (1-2): 1 ^Powershell

SECourses

hi please delete venv folder and run installer again it should fix

Max Gonzales

Are there recommended settings for the Rtx 5090, I keep eating into shared vram and my gpu seems to only run at about 80watts?

SECourses

FP8 and non tiled upscale. but that will limit your max resolution

SECourses

83 mega pixel upscale probably impossible

Eclipse Beast - TF

Error, the new version is broken, I’m getting an error I noticed that during installation, it also throws an error: ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt C:\Users\stare\AppData\Local\Programs\Python\Python310\python.exe: can't open file 'S:\\AI\\Programy\\Upscare\\SUPIR_v74\\gradio_demo.py': [Errno 2] No such file or directory

SECourses

install error. and what is S drive usb etc? install into C drive and send me logs : monstermmorpg@gmail.com

Eclipse Beast - TF

I won’t install this on drive C because it’s not possible. Drive S is simply the second partition on my system, but that doesn’t matter anywaym the issue is caused by missing or incomplete files. Because of the lack of this, he practically can't install anything. Check the file structure, because I was able to install previous versions normally using Windows_Install, and now everything is breaking due to missing files

Eclipse Beast - TF

I fixed it , the antivirus on Windows 11 was blocking and deleting something

SECourses

ok great finding thanks for letting me know

Adrian Kajzer

Where is the problem? Processing 1 images... Processing LLaVA You are using a model of type llava to instantiate a model of type llava_supir. This is not supported for all configurations of models and can yield errors. An exception occurred: Using `bitsandbytes` 8-bit quantization requires Accelerate: `pip install accelerate` and the latest version of bitsandbytes: `pip install -i https://pypi.org/simple/ bitsandbytes` at Traceback (most recent call last): File "/workspace/SUPIR/gradio_demo.py", line 1242, in start_single_process _, result = batch_process(img_data, **values_dict) File "/workspace/SUPIR/gradio_demo.py", line 1797, in batch_process last_result = llava_process(img_data, temperature, top_p, qs, save_captions, progress=progress, skip_llava_if_txt_exists=skip_llava_if_txt_exists) File "/workspace/SUPIR/gradio_demo.py", line 1308, in llava_process load_llava() File "/workspace/SUPIR/gradio_demo.py", line 742, in load_llava llava_agent = LLavaAgent(llava_path, device=LLaVA_device, load_8bit=args.load_8bit_llava, File "/workspace/SUPIR/llava/llava_agent.py", line 27, in __init__ tokenizer, model, image_processor, context_len = load_pretrained_model( File "/workspace/SUPIR/llava/model/builder.py", line 102, in load_pretrained_model model = LlavaLlamaForCausalLM.from_pretrained(model_path, low_cpu_mem_usage=True, **kwargs) File "/workspace/SUPIR/venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 3024, in from_pretrained hf_quantizer.validate_environment( File "/workspace/SUPIR/venv/lib/python3.10/site-packages/transformers/quantizers/quantizer_bnb_4bit.py", line 62, in validate_environment raise ImportError( ImportError: Using `bitsandbytes` 8-bit quantization requires Accelerate: `pip install accelerate` and the latest version of bitsandbytes: `pip install -i https://pypi.org/simple/ bitsandbytes`

SECourses

hi testing now sorry for late reply

SECourses

sorry for this error fixed with v76 make sure to select Runpod Pytorch 2.2.0 runpod/pytorch:2.2.0-py3.10-cuda12.1.1-devel-ubuntu22.04 and set 150 gb disk space

Krzysztof Madej

In what order should I run the following scripts to ensure that the installation proceeds correctly? Windows_Install.bat Windows_Resume_Fix_Model_Downloads.bat Windows_Update_Requirements.bat Windows_Update_Version.bat

SECourses

Windows_Install.bat is sufficient for new install. you need Windows_Resume_Fix_Model_Downloads.bat only if some of the model downloads fails. the others are needed to update later

Error_404_unknown

just a tip if anyone is interested. If your runing on massed compute generate the LLava captions locally and put them with with your images before uploading. It does work but think I'm going grey it's so slow. Upscaling is super quick though on an ADA 6000 compared to my 3090.

Error_404_unknown

Hello Dr Furkan not sure if this is possible but when processing large batches if there is and error like OOM is it possible to just skip that file and go to the next rarther than terminating the whole batch and comming back in the morning to find it did almost nothing?

SECourses

unfortunately not possible with how pytorch working. perhaps a new threads must be spawned each time and it may work but that can also cause model loading each time. if you just restrict your max resolution i think you shouldnt get OOM. simply it is not easy i need to code and test.

SECourses

yes RTX A6000 ADA is great. moreover SeedVR2 optimized upscaling coming soon hopefully

Error_404_unknown

Omg thanks, didn't know about this just tried in comfy on massed compute it is so fast.

Regina Ciira

im trying to install but it keeps saying access denied..what should i do

SECourses

hello that is your windows installation related. if you upgrade gold member i can connect your pc and possibly fix

Philipp Ludwig

Nothing happens after install. Models downloaded. Its a first install. The bat says something about venv/activation but couldn't find any bat

Error_404_unknown

Big thank you on the update! Now when you realise you messed something up and cancel it doesn't take 45min to stop! It almost instant now, this is great.

SECourses

awesome. there are some visual things i will try to fix now

Neil Rhodes

Suggestion: I am a huge fan of Supir, however, there used to be something in stable Diffusion called Tile upscale. Taking a small image and upscaling chunks and stitching to make a massive image, is there a way to tile upscale like the old SD used to do it. If there is a very large image already... I used a manual method in SUPIR where i broke a very large image down into pieces and 1:1 enhanced the details without upscale and then manually stitched them all back again, which worked well but a lot of work. The 1:1 no-upscale was a great way to just enhance that large image, and smooth out some textures and blotches. Could this be a future feature?? So it can handle larger images but the "sharpen" them ? it was such a great way to enahnce it :)

SECourses

sadly it cant be a feature with this. but i am gonna add tiled SeedVR2 and it is really good and fast and it will have full tiled upscale feature

Neil Rhodes

I think that is good news then?

Rafał Potoczny

Hey, I've got error.. any cloe? "LLaVA processing completed Clearing LLaVA. LLaVA cleared. LLaVA unloaded Processing images (Stage 2) Loading model from [options/SUPIR_v0.yaml] ERROR: Exception in ASGI application + Exception Group Traceback (most recent call last): | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/uvicorn/protocols/http/httptools_impl.py", line 412, in run_asgi | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/uvicorn/middleware/proxy_headers.py", line 69, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/fastapi/applications.py", line 1054, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/applications.py", line 123, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/middleware/errors.py", line 186, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/middleware/errors.py", line 164, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/middleware/cors.py", line 85, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/middleware/exceptions.py", line 65, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/_exception_handler.py", line 64, in wrapped_app | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/_exception_handler.py", line 53, in wrapped_app | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/routing.py", line 756, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/routing.py", line 776, in app | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/routing.py", line 297, in handle | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/routing.py", line 77, in app | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/_exception_handler.py", line 64, in wrapped_app | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/_exception_handler.py", line 53, in wrapped_app | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/routing.py", line 75, in app | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/responses.py", line 258, in __call__ | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/anyio/_backends/_asyncio.py", line 772, in __aexit__ | exceptiongroup.ExceptionGroup: unhandled errors in a TaskGroup (1 sub-exception) +-+---------------- 1 ---------------- | Traceback (most recent call last): | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/responses.py", line 261, in wrap | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/responses.py", line 253, in stream_response | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/_exception_handler.py", line 50, in sender | await send(message) | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/_exception_handler.py", line 50, in sender | await send(message) | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/starlette/middleware/errors.py", line 161, in _send | File "/workspace/SUPIR/venv/lib/python3.10/site-packages/uvicorn/protocols/http/httptools_impl.py", line 465, in send | uvicorn.protocols.utils.ClientDisconnected +------------------------------------ "

SECourses

looks like temporary pod related error. i recommend to install into another pod or turn off machine and turn back on and try again

SECourses

ye that is biggest issue of runpod. i recommend massed compute

Rafał Potoczny

I tried use MassedCompute, but after few generations/some time I got everything laggy and slow.. any tip? that's why I'm using currently RunPod :)

Rafał Potoczny

I wrote wrong, sorry :) my problem is: I'm having an issue with SwarmUI on MassedCompute. When I launch it through the GUI and start generating images, after a few generations the whole system gets super laggy - mouse moves slowly, screen becomes unresponsive, etc. I read somewhere that I should access SwarmUI through Cloudflare icon instead of the desktop interface, but haven't tried it yet. What's the proper way to run SwarmUI to avoid these slowdowns? Should I be using the Cloudflare link instead of opening it through the GUI? Thanks for any help!

SECourses

Yes please use cloud flare icon that is what I use. Open the url on your pc and will work great

Shing Ming Lin

It is recommended to upgrade to Python 3.12, as many new programs now support it and the system is also updated.

SECourses

ye but this program works perfect with python 3.10.11. so have 3.10.11 in your system as my installer will use it

Cemil Hacimahmutoglu

hocam kaggle yüklemesinde * Serving Flask app '__main__' * Debug mode: off /kaggle/temp/SUPIR Traceback (most recent call last): File "/kaggle/temp/SUPIR/gradio_demo.py", line 25, in import pillow_avif # Import the AVIF plugin ^^^^^^^^^^^^^^^^^^ ModuleNotFoundError: No module named 'pillow_avif' hatalarını alıyorum manual olarak pillow_awif isteme eklemem rağmen aynı hata devam ediyor yardımcı olursanız sevinirim.

SECourses

fixed with v97 zip file. use new kaggle notebook

matte

I get this error no matter what I do: File "D:\SUPIR_v97\SUPIR\SUPIR\util.py", line 74, in create_SUPIR_model load_to_device(config.get('SDXL_CKPT')) File "D:\SUPIR_v97\SUPIR\SUPIR\util.py", line 65, in load_to_device state_dict = load_state_dict(checkpoint_path, tgt_device) File "D:\SUPIR_v97\SUPIR\SUPIR\util.py", line 27, in load_state_dict state_dict = safetensors.torch.load_file(ckpt_path, device=location) File "D:\SUPIR_v97\SUPIR\venv\lib\site-packages\safetensors\torch.py", line 383, in load_file result[k] = f.get_tensor(k) torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.00 MiB. GPU 0 has a total capacity of 11.99 GiB of which 0 bytes is free. Of the allocated memory 10.74 GiB is allocated by PyTorch, and 487.42 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

SECourses

hello it is not an error. it is out of VRAM. run as FP8 + tiled. also if you still getting error that means your target resolution is too big for your 12 GB GPU

SECourses

it will scan Lora sub folder inside models folder. you cant set anything else at the moment.

SECourses

you are right : 21 September 2025 Update V100 Now you can give custom models path and loras path For custom models path (SDXL models) add this args --ckpt_dir e.g. : --ckpt_dir "C:\temp\b" For custom loras (SDXL Loras) path give this args --loras_dir e.g. : --lora_dir "C:\temp\c" To add this args edit Windows_Start_SUPIR_app.bat and add there You will also see an example case that has REM at the beginning REM means it is commented - not executed Just execute Windows_Install_or_Update.bat file to update

SECourses

it is --outputs_folder "C:\temp\o"

Neil Rhodes

What practical use is a custom model or LoRA can you advise a use case?

SECourses

well custom models can certainly impact quality of output but i didnt find any for Loras. added it since people requested.

SECourses

ye your lora iss fp8. get proper fp16 loras. fp8 loras doesnt make sense at all also

100Sachen80

Congratulations to the 100th release of the zip file :D

Rafał Potoczny

Hello Furkan, I need your help: "Enter your choice (1-2): 2 App startup: is_processing = False IMPORTANT: You are using gradio version 4.19.0, however version 4.44.1 is available, please upgrade. -------- Running on local URL: http://127.0.0.1:7862 Could not create share link. Please check your internet connection or our status page: https://status.gradio.app." I tried update gradio, change GPT to 4090, A6000, A40, etc.. even perplexity didn;t help :( but after update gradio, venv and another commands still the same... btw the most up to date is gradio 5.49.0 i think..? root cause is that: it doesnt create gradio link...

SECourses

gradio link is needed for cloud machines. where do you running on runpod? if so delete pod and get a new one.

Rafał Potoczny

Yes, I use runpod. Supir v100. I did this - different pod as I mentioned before on rtx 4090, 6000. I have network volume with all models. I did re-install and stil the same. Any ideas? 💡 Thanks.

SECourses

keep trying again. gradio live is failing due to network of the runpod sadly. but you can use runpod proxy. add proxy ports 7860 7861 7862 7863 and then connect via proxy port

Robert Arsene

I've been trying to upscale pictures and I can no longer get that crisp sharpness. Very noticeable on the faces as they get a grainy look. Do you know why it may be?

SECourses

sadly i dont know. but we didnt change the app it should be same as before. perhaps your configs or used base model not same?