Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

1-Click INSTALL QWEN 20B TEXT TO IMAGE KING ComfyUI WebUI! (Patreon)

Published:
2025-08-11 20:04:34
Edited:
2025-08-11 23:47:18
Imported:
2026-06
Tags:

Downloads

Content

Hey everyone! I've created a 1-click installer for the NEW QWEN IMAGE MODEL! THE NEW BEST TEXT TO IMAGE MODEL ever made! Insane artstyle, incredible prompt following, great text design and just amazing quality. It can do everythingl. BUT...NOT BY ITSELF...The only way to TRULY Make it the best is THE SUPER COMBO TEAM QWEN + WAN 2.2! And My Workflow makes this possible!

You can check out the video right here: https://youtu.be/7P4LHEAEGNg

The workflow: https://www.patreon.com/posts/136261640

The installer of course automates the entire install process for comfyUI and the models downloads!

LOCAL INSTALL!!! RECOMMENDED TO START FROM SCRATCH SINCE YOU NEED THE LATEST COMFYUI TORCH VERSION!!!

Download the QWEN-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat

Run the bat file and select either option 1, 2 or 3 depending on your GPU vram.

Wait for the install to finish

The webui will launch, load the QWEN_COMBO_ULTRA_WORKFLOW.json file and there you go😎 DON'T FORGET TO GO INSIDE THE MANAGER AND CHOSE THE NIGHTLY BUILD FOR COMFYUI TO MAKE SURE THAT THE 8 STEPS LORA WORKS

IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER:

Download the QWEN-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder

Run the bat file and select either option 1, 2 or 3 depending on your GPU vram.

Wait for the install to finish

???

Profit 😎

IF YOU GET AN ERROR WHEN LAUNCHING COMFYUI AFTER USING THIS INSTALLER SAYING "AssertionError: Torch not compiled with CUDA enabled" it's ok don't panic.

Just go inside the python embedded folder, click on the folder path, type cmd press enter, this will bring the command prompt window and there type:

python.exe -m pip uninstall torch

python.exe -s -m pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

--->replace cu128 by your cuda version (if you don't know your cuda version just open a new cmd window and type nvcc --version it will then tell you at the end Cuda compilation tools, release 12.8, V12.8.93 or something different like 12.6, in that case just use cu126 instead of cu128)

IF YOU ARE USING RUNPOD:

Create an account if you haven't already: Runpod

Click on Pod (on the left side) then click deploy

Choose a GPU with at least 24gb of VRAM (4090 is best), click here for the comfyui template (aitrepreneur/comfyui), then edit the template and choose 100gb for both the container and volume disk, then deploy on demand

Go to my pods, wait for everything to finish and then click "connect", then "Connect to HTTP SERVICE port 3000" and click on the manager to update comfyui and restart itThen go to the port 8888, go inside the comfyui folder and then drag and drop the QWEN-WAN-AUTO_INSTALL-RUNPOD.sh file on the left side of the UI then click on the "Terminal" icon on the right side on the UICopy and paste these two lines then press enter:

chmod +x QWEN-WAN-AUTO_INSTALL-RUNPOD.sh

./QWEN-WAN-AUTO_INSTALL-RUNPOD.sh

Wait for everything to be installed

Go back to the port 3000 and click on the manager, choose the Nightly build and restart it

Load the QWEN_COMBO_ULTRA_WORKFLOW.json file and there you go😎

Also use the command:

tail -f comfyui.log

inside the workspace/logs to get the real time logs so that you always know when the install is done. It will then ask you to restart, so click on that restart button, wait for the webui to reboot, click refresh once it's done and there you go.

IF YOU ARE USING RUNPOD AND YOU WANT TO STOP THE POD THEN RESTART IT:

Go inside your pod

SERVICE port 8888

Go inside the Comfyui folder, open a terminal icon and re copy and paste those two lines:

chmod +x QWEN-WAN-AUTO_INSTALL-RUNPOD.sh

./QWEN-WAN-AUTO_INSTALL-RUNPOD.sh

And then in a few seconds it will reinstall all the requirements for the nodes

Go back to comfyui port 3000, click on manager, click update all, wait for everything to be updated then restart.

Re drag and drop the QWEN_COMBO_ULTRA_WORKFLOW.json file inside and there you go😎

As always, supporting me on Patreon allows me to keep creating helpful resources like this for the community. Thank you for your support - now go have some fun😉!

Files

Comments

Brad

Is this able to do uncensored images? or guardrailed?

James

Thanks and well done for making this excellent workflow Aitrepreneur! QWEN is really amazing especially with it's ability to accurately add text! By the way, could you add anything like OmniAvatar to your workflows? It would be awesome to add audio and lipsync to AI generated videos and make our own talking lip-synced avatars made with QWEN/WAN!

phail216

I am getting things like this in the console - lora key not loaded: - Falling back to numpy dequant for qtype: 23 and the generation is pretty slow (2 mins on a 4090). Is this working as intended?

Brumfondl Majikthise

Make sure to switch ComfyUI to the nightly build - I still get the 'falling back to numpy dequant for qtype: 23' message though

Brian Zufelt

I am having the same issue, and I made sure to switch to the nightly build.

Aitrepreneur

it is "censored" like wan was but as always lora allows you to do whatever like always

Aitrepreneur

I was looking into that a few weeks back but unfortunately I couldn't find something that I liked, at least for now but I'm HEAVILY interested in that, so when something good comes out, you'll know

James

Thanks - that would be really good.

HealingPaint

I cannot find the JSON workflow? Where is it?

Vincent Nestler

is there an option for folks using a mac?

HealingPaint

Everything I generate with Qwen (text to image, via the first module in the workflow) comes out burned/blurry/fuzzy...why? Ruins it all. Any ideas? I have a RTX-4090 and am using the default 4-step w/ Lightning Lora settings.

Jurijs Misnakovs

thanks a lot for your work and workflow! what about Loras ? we will need to train separate Loras for Qwen Image, right ? how to do it right now ?

FILM426

you can train quen loras on the updated ostris ai toolkit

FILM426

just simply legendary work! thank you so much

Roni Bliss

It's no good with photo realism well though Wan is still the GOAT in that area. And as a Nigerian scammer by day and hardcore gooner by night I need only convinving images. f you want glossy fantasy images it's the go to. But anything approaching realisitic it can't help spraying its stomach with sticky white AI love piss. It's actually pissing me off how needy it is for attention. I wan't raw amatuer grain straight from the dirt.

Xyzzx

Wow! Given the Wan 2.2 announcements and QWEN, I was like, wouldn't it be great to have a unified comyfui workflow... but I'm inexperienced with anything but constructing the most basic workflows. Can't wait to try this out on Runpod (980ti D:)

Marranu

ok,i must be doing something wrong. my error is "git-setup.exe , is Not compatibly with a 64 bit version of windows. contact vendor to ask if a 64 bit compatibly version is available.version." got any ideas?

Adam Booth

You need to download the git for windows 64 bit version first. https://git-scm.com/downloads/win Thinking about it you might find that even after installing the 64bit version you still get the error in which case perhaps commenting out this line in the script might work: #set "GIT_VER=2.45.0.windows.1"

Xaro

@Aitrepeneur: I love your work man, thanks a lot! 👏🏻 Any plans to integrate a function in your i2v workflow, that when i create like a 5sec Video i could use the last frame and extend the Video again and again? Would be really awesome. Much love to you ❤️✨

phail216

Yeah I did make sure to use the nightly build and now also the Q4 model. I'm still getting the errors but the gen time is down to 25s

quater

I get this error: Unexpected text model architecture type in GGUF file: 'qwen2vl' Do you know why and how I can solve that?

quater

It's ok, i solved it. I thought it would also update the nodes with update All, but it idid not. After updating the gguf nodes it worked.

Arthur

I have the same problem

quater

For me it makes really bad images and I don't get any error messages. the woman in the default images is full of frickles. I also tried to make something like a beach with white dolphins jumping out of the water, palm trees on the beach and the dolphins have white skin shimering in rainbow colors. It was really really bad. At least Wan made dolphins and a good picture as far as it could follow the prompt.

quater

Same for me. Really bad images. The default image is also bad. The woman is full of frickles and the face is also a bit weird.

Frank B

when i try to run it.... size mismatch for model.layers.27.input_layernorm.weight: copying a param with shape torch.Size([3584]) from checkpoint, the shape in current model is torch.Size([2048]). size mismatch for model.layers.27.post_attention_layernorm.weight: copying a param with shape torch.Size([3584]) from checkpoint, the shape in current model is torch.Size([2048]).

Jade Hogg

That is awesome, do you have perhaps a tutorial mate on how to implement and which Lora can be used ?

quater

Add this in the negative: cartoon, anime, static, blurred detailes, subtitles, painting, worst quality, low quality, overexposure, flat cel shading, CGI, plastic look, text overlay, watermarks, extra limbs, bad limbs, cripples limbs, missing limbs, low-res, motion blur, bad quality

Matt H.

I get the same when switching ComfyUI to nightly.

kill vearn

There's something wrong going on with this model (tested with both Qwen Q4 and Q5). In text to image there is a specific prompt (maybe more) that he can't seem to process (lora or not) and gives out a blurry mess. I tought I had something wrong with my setup, but try this prompt and you'll see: "The image shows a young woman sitting on a white furry rug with string lights hanging above her head. She is wearing a white lace bra and matching orange thigh-high stockings. Her blonde hair is styled in loose waves and she is looking directly at the camera with a sultry expression. The overall aesthetic of the image is sensual and feminine."

Marranu

thanks, I knew I was overlooking something.

Frank B

Wait, Does the work flow work for you in ComfyUI Stable? I though it didnt?

Matt H.

I can get decent images intermittently on Stable. I dont think other loras besides the lightning are working properly.

Matt H.

Working on Stable after I updated from comfyui_windows_portable/update/update_comfyui.bat

DrHans Brewery

Can we please get a combined version of the installer and workflow for image and video please.

King Debs

hey, how do you update the gguf nodes specifically ? i had a similar message, now i have a new message after update all. cannot reshape array of size 7802545 into shape (18432,1728)

Aitrepreneur

you can find them on civitai (maybe on huggingface and tensor art as well), put them inside the models/lora folder, then load them using the power lora loader node (pink node) then use the trigger words in the prompt

King Debs

cannot reshape array of size 7802545 into shape (18432,1728) someone please help.

Frank B

Thanks, though given the various issues people are getting with this setup, I might just sleep on it for now - until maybe Aitrep updates it in the future.

Aitrepreneur

This seems like the lora is not loading, probably because you haven't updated comfyui and you haven't changed to the nightly build

Aitrepreneur

Yes, training a lora for Qwen is easy, I already trained a few for it but you need a 5090 for it

Gabo

Why isn't my comparator working? The images below left appear, but nothing appears in the "Combo grain" outputs. Thanks.

Aitrepreneur

hence why for now you need the qwen wan combo to add more realism back but once we get some good realism lora or custom finetuned models, that issue is over

Aitrepreneur

I will make an updated workflow once we get a vace model maybe

Aitrepreneur

This seems like the lora is not loading and you haven't changed to the nightly build

Aitrepreneur

it works though, just update comfyui with the update comfyui bat file inside the update folder and you change to the nightly build inside the manager like showed in the video

Jurijs Misnakovs

yes, sure, I already understood that only option for me is a runpod, so it could be amaizing if you release a 1 click script for a runpod with simple instructions for that :) I`m dumb at all those complicated things, but your guidance is absolutely amazing and easy. your AI toolit for WAN 2.1 lora training is so good, but it works (starts) only for WAN 2.1 training and not for other models, but they are in the list.. different models give different errors on the start..

Aitrepreneur

You can just use the MODELS-NODES_INSTALL.bat files for each installers for that

Aitrepreneur

just update comfyui with the update comfyui bat file inside the update folder and you change to the nightly build inside the manager like showed in the video

kill vearn

link not working. Did it came out right? I can do all kind of images, but when I put that prompt all becomes a random mess barely visible. I even tried reducing it but removing things one at a time to look for triggering words, but I couldn't find any.

Cristo

is there privacy using runpod? and How do I get the images and videos I generated to my PC?

Aitrepreneur

No worries about privacy, for the finished images and videos, you can either save them manually when right clicking on the image and video or you can go inside the output folder then download them from there

Boris

I did both those things and they solved a first issue on the clip loader not having gwen-image as an option, however I'm not getting the same "cannot reshape array of size 31073314 into shape (18944,2016)" error as the post above. I've run both update_stable and update.bat just in case but still not working :s

kill vearn

not working yet, anyway I've uploaded a test to show the issue: https://postimg.cc/XpwgPj0k literally the only thing changing is the prompt, using lora 8steps (but it doesn't make a difference 4-8 or no lora).

King Debs

hi, i had doen all of tha before. updated comfyui, swittched to nighly build liek video. ran the log issue on chatgpt, still cant fix the issue.

kill vearn

Found the culprit. Something to do with the text interpretation (ironic). "The image shows" makes a mess. If I type "Photo of a" it gets better.

kill vearn

it's the default parameters, I even reinstalled it with <12 and >12gb options. Literally changing the prompt fixes the result. Only thing I change is the steps based on the lora. I don't know how to give you the parameters, I can screenshot but it's barelly visible.

Aitrepreneur

That just means that your 4 steps lora is not loading. Probably because you have haven't updated comfyui and you haven't selected the nightly build. Use the update comfyui bat file inside the update folder

kill vearn

nope, I updated and installed nightly. And the issue is present with 4 step -lora 8 step lora and no lora. You can see here that changing the prompt changes the result without changing anything else: https://postimg.cc/XpwgPj0k

Aitrepreneur

The prompt is not the issue since I litteraly used your prompt to make the image you see above

kill vearn

here https://imgbox.com/YkckysTm I hope is somewhat readable, just tested it again. Without changing the parameters, I change the prompt and the image is fine.

Aitrepreneur

https://imgbox.com/XWBVE7eP I'm telling you it's not the prompt. The lora is probably not loading right, either the file is corrupted or you have not correctly changed to the nightly build. What do you see in the cmd window? Any lora errors?

kill vearn

I don't know what to say, I'm creating awesome images, just can't get a nice image with that prompt alone. I get sometime but not always this message: CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16 lora key not loaded: transformer_blocks.0.attn.add_k_proj.alpha lora key not loaded: transformer_blocks.0.attn.add_k_proj.lora_down.weight lora key not loaded: transformer_blocks.0.attn.add_k_proj.lora_up.weight lora key not loaded: transformer_blocks.0.attn.add_q_proj.alpha with a bunch of errors, but only sometime and not always with that prompt. I also get more often this: Falling back to numpy dequant for qtype: 23 Falling back to numpy dequant for qtype: 23 Falling back to numpy dequant for qtype: 23 Falling back to numpy dequant for qtype: 23 I don't think is a corruption because I deleted the folder and tried again. And I get the same result even disabling loras and increasing steps.

Frank B

Was on nightly, but will try the update- thanks.

kill vearn

I know it makes no sense, but it's just that prompt that makes the image become a blurry mess

Aitrepreneur

please don't spam here, I can't read all comments, just dm me all info

Speedy2023

In my case changing to the 8-step QWEN lora and changing the value of the steps in the corresponding KSampler(s) helped.

HealingPaint

Quality for terrible, so I updated everything, shut it down and restarted....How do we fix the issue with wan_image not loading in the cliploader(gguf)? I've updated everything, including the GGUF loader, restarted completely and now it won't work. wan_image is not an available option (I am using the latest beta)

HealingPaint

I fixed the wan_image issue...and quality issues by running update_comfyui.bat from the update folder...don't trust the GUI update all to do the job folks...

Raven Shadow29

Found a way to get a clean and sharp image for this workflow once again update ComfyUI from the update folder not the manager update comfyui and python from here sorts out the out of focus and blar in the image, just as a side note 2 parts of the python update is now looking for python 3.13 on your system install this first or 2 parts will fail to update should work as intended after this worked for me on my pc

HealingPaint

That wasn't the case. I did all that to no avail. What fixed it finally was updating comfy from the UPDATES folder after closing it all out. Update All from the GUI w/ restarting/refreshing wasn't doing it. I tried that a couple of times before this. You should mention that in your videos! Please.

xpatjohn

Apologies in advance, but I get this error when trying to run the workflow: Prompt outputs failed validation: UnetLoaderGGUF: - Value not in list: unet_name: 'Qwen_Image_Distill-Q4_K_S.gguf' not in ['Qwen_Image_Distill-Q8_0.gguf', 'Wan2.2-T2V-A14B-LowNoise-Q8_0.gguf'] I know I probably did something wrong, but I have no idea what.

Len

Thank you!!! I had same issue, this fixed it!

Osvaldo Alfaro

I keep getting "Missing Node Types. When loading the graph, the following node types were not found" for a node called "MagCache" .

Osvaldo Alfaro

Ah! Looks like I managed to go in and install it manually in the Manager. :)

Speedy2023

Thank you very much. That definitely helped.

Anthony Sparks

when running i get this error: KSampler mat1 and mat2 shapes cannot be multiplied (703x3584 and 4096x5120) is there a way to fix this??

Anthony Sparks

did this ever get fixed??? I have the same problem, ran the update file and im on the nightly build

J

Do you or anyone in the comments have any experience with running all of these various models such as QWEN, WAN, Flux, etc... on linux? Any recommendations if so?

Len

QWEN is pretty great, BUT I found it's actually terrible with Text. Maybe because of the distilled model, but I ran several "movie poster" type generators, and it messed every single one somewhere... I did a quick comparison between several image generators so see for yourself: http://funny.name/misc/AOF50GM.jpg Prompt was: "Minimal but striking comedy-horror movie poster for “Attack of the 50-Foot Grandma.” A giant, angry-looking elderly woman in a pink floral dress and pearl necklace towers over a small suburban neighborhood, clutching an enormous rolling pin like a baseball bat. She looks angry and her eyes glow ominously. Tiny cars and panicked people scatter on the street below her feet. The sky is bright blue with a few fluffy clouds, giving the scene an absurdly cheerful tone despite the chaos. The title “ATTACK OF THE 50-FOOT GRANDMA” dominates the bottom in huge, bold retro 1950s sci-fi style letters. Tagline: “Dinner’s Ready… and So Is She.” Clean composition, high detail, photorealistic with exaggerated scale. Printed on slightly crumpled, faded paper, with subtle wear, fold marks, and vintage poster texture, colors slightly muted as if aged."

quater

I did the same and it worked, but it is no way near as good as it should be.

Roni Bliss

I just got an RTX Pro 6000 with 96gb VRAM - Now trying to pull off creating Loras and hopefully make some fine tuned checkpoints I have 63,000 photo dataset I want to create from.

dattrax

I'm running on AMD, so would like the script to ask before trampling my torch/torchaudio/torchvision as a feature request

Brudolf Wikswolle

Can someone help me? I have a problem I cant run the workflow. Every time I try to run an error stating: Clip loader gguf unexpected text model architecture type in GGUF file "quen2vl". Does anyone know a solution for that?

Danilo Ramscheid

OSError: [WinError 127] Não foi possível encontrar o procedimento especificado. Error loading "D:\Qwenimage\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\lib\torch_cuda.dll" or one of its dependencies.

BlazingSandwich

Help! I can't use the ComfyUI auto install. Says it is not compatible with 64-bit. Is there something I am missing?

Roni Bliss

Qwen Lora training now available on the amazing AI-Toolkit!! - Check out Ostris the creators YouTube channel for Tutorial. The results already amazing, tested fast Lora creation on my RTX Pro 96gb VRAM

Adele Kaoki

IF you have low quality or lora not loaded errors after switching to nightly build, close it, go to the folder > update > update_comfyui.bat and let it roll until end. The manager update is sometimes halting midway, so this fixes it. I am, though, getting a lot of "Falling back to numpy dequant for qtype: 23" messages, you guys know any reason for that?

Xyzzx

Wait, wtf is an RTX Pro 96GB and how do you have one?

Xyzzx

Set up the 100gb storage in advance maybe, that's one place I know you can click on encrypt

Xyzzx

Any ideas about black output? I am hitting some NSFW filter?

Jaina Shop

anyone with this error? KSampler mat1 and mat2 shapes cannot be multiplied (312x3584 and 4096x5120), thank you Jaina

I Bea

In plain terms, the LoRA was trained against a different architecture / naming scheme than the UNet (or text encoder) you’ve loaded, so ComfyUI can’t find where to plug that LoRA weight.

Timmy

Not sure if this is the same error or not, but I know I sometimes I will get an error similar to this one, often with the values being different. It looks like the one you posted but not exactly. I found this to be caused due to: 1. A mismatch between the Clip and Model. - Verify you are using the correct Clip (Text Encoder) and Model files. I have found multiple Clip (Text Encoder) / Model files are named identically but are not the ones I needed. There are multiple Text Encoders that have identical names but are not for the "Clip" format, which is the one you need. 2. The clip or model files are corrupt and did not download correctly. -If they are the correct files (particularly when using the auto downloader bat) the files became corrupt during downloading. Re-downloading them directly without using the auto installer can fix the issue. You can look inside the batch file to get their download links. Your error looks similar to the one I get when the above issues occur, but not the same. Might be worth a try to look into my suggestions above.

Timmy

I also get spammed with "Falling back to numpy dequant for qtype: 23". I used the node batch file provided to install the files as I already had ComfyUI installed. The following are the model/clip/vae/lora I am using: Power Lora Loader(rgthree): "Qwen-Image-Lightning-4steps-V1.0.safetensors" Unet Loader (GGUF): Qwen_Image_Distill-Q4_K_S.gguf CLIPLoader (GFUF): Qwen2.5-VL-7B-Instruct-UD-Q4_K_S.gguf Load VAE: qwen_image_vae.safetensors Am I doing something wrong?

Jason G (edited)

Comment edits

2026-09-08 20:00 Can someone point me to some good AI tools for generating prompts for image and video generation?
2025-08-14 20:28

Aitrepreneur

you probably haven't selected the correct model in the model node, just click on the model name and select the model that was downloaded for your gpu

Aitrepreneur

you are using another workflow since there is no magcache nodes in this workflow

Aitrepreneur

you are probably using incorrect models or resolutions, you can dm me screenshots of the parameters

Aitrepreneur

except on runpod, no I do not but it should be the same

Aitrepreneur

try maybe with the 8 steps lora and a different more precise prompt

Aitrepreneur

update comfyui correctly using the update comfyui bat file inside the update folder, then re drag and dropping the workflow inside comfyui and making sure that qwen is selected in the clip loader node

Aitrepreneur

update comfyui correctly using the update comfyui bat file inside the update folder or reinstall comfyui

Roni Bliss

They are NVIDIA blackwell cards - Prosumer, made for AI workstations, there are many places to buy them, even on Amazon lol https://www.amazon.com/Blackwell-Professional-Workstation-Simulation-Engineering/dp/B0F7Y644FQ Pretty cheap for the power also: Our card cost $6,800 - £11,000 with Nthe full VIDIA workstation, but you can put them in most PC towers, they hardly consume much power, and unlike most Prosumer graphics cards they only take up 2 SLOTS! .. Amazing card for AI ... Whatever you throw at it, original untouched (BF16 / FP16) Qwen image around 5 second image time, Flux1.dev 2-3 seconds, insane power! - Its good for the next 10 years without having to upgrade i reckon

Tibby

Which QWEN IMAGE quantization? 1) Q4_K_S ΓÇô GPUs < 12 GB VRAM RECOMMENDED 2) Q5_K_S ΓÇô GPUs 12-24 GB 3) Q8_0 ΓÇô Best quality, GPUs 24 GB+ Enter 1, 2 or 3: 1 -------- Checking prerequisites -------- -------- Downloading ComfyUI -------- % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0 25 2028M 25 520M 0 0 2608k 0 0:13:16 0:03:24 0:09:52 1781k curl: (18) end of response with 1581838371 bytes missing Download failed. Press any key to continue . . . Not working bud

EX01

Grok 4 does amazing too. However this workflow just refuses to do NSFW (asking for a friend :-)) Im still using last weeks WAN2.2 which does amazing quality.

zedd

download is not working for me. keeps failing

jhon

i got the same for wan 2.2 video... i think you might have already installed comfy ui and are trying to install again in seperate folder. this worked for me (download the nodes file from the page corresponding to what you want to install, below is what worked for me since i already installed qwen, so for you download nodes file from qwen post and place in previous comfy ui folder and run it) IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER: (yours is QWEN-ULTRA-MODELS-NODES_INSTALL.bat) Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) . Wait for the install to finish. i got cuda error after that but fixed following instructions on how to fix on wan 2.2 ultra video instruction download post...ater that i could not generate videos until i turned my windows paging file size set to let windows decide.

Mark Mynhier

Is there a good tutorial for using ComfyUI? I'm used to 1111.

Part Time Poker Pro

curl: (18) end of response with 491319331 bytes missing Download failed. Press any key to continue . . . this happens to me everytime. I deleted my comfyui folder to start from scratch. keeps failing

Part Time Poker Pro

yup, me too it keeps failing curl: (18) end of response with 491319331 bytes missing Download failed. Press any key to continue . . . this happens to me everytime. I deleted my comfyui folder to start from scratch. keeps failing

Len

Pretty much any prompt longer than 3 words has issues. Increasing steps helps but doesn't solve it completely...

Bradford

Well that means you can use the AI video models at full quality, without having to find tricks to get them to work with less VRAM. But $7000+ is a lot of money to spend on generating AI videos that maybe no one cares about and maybe no one would pay money for. I thought I was horrible for spending over $500 on Runpod credits and maybe I was. GPU tech is advancing rapidly so it should come down in price drastically in the next few years. Now I finally have a 3090, and I am fortunate that I have been able to get most of the things from AItrepreneur to work. There are often some little tricks you need to figure out tho, to optimize the quality and the VRAM usage.

sai ram

Which model is best for anime image generation: Kontext, Qwen, or Wan2.2, without anatomical distortion? And LoRA training?

Part Time Poker Pro

I deleted my entire COMFYUI folder in order to install with a clean/fresh wipe. But now the auto install .bat file errors halfway thru downloading. did I delete/uninstall incorrectly?

Cristo

When I try to use runpod, it says "memory 100%" then it crashes, I cant see the result of what im creating. How do I solve this?

adamaave

Can this model be used to create girls like in Flux? I tried to make a new model (with a detailed prompt), but instead I just get a black image. When I use a simple, random prompt, it generates normally. Any tips? I’m mainly working on NSFW content. Maybe I should look into a different course? I’d really appreciate any advice.

Rahul Rabha

even Iam getting black image, Iam not using NSFW but trying to generate some hindu dieties. I don't know what Iam doing wrong?

VINCENT ALONZO

I got it running fine but the results are not great for some reason when comparing the exact input image/prompt to wan 2.2...

BlazingSandwich

This was from further down: You need to download the git for windows 64 bit version first. https://git-scm.com/downloads/win Thinking about it you might find that even after installing the 64bit version you still get the error in which case perhaps commenting out this line in the script might work: #set "GIT_VER=2.45.0.windows.1"

Anna

I don't get this one, everything I got out of it, on suggested settings, was SD era quality. At least for "photos".

Ian Brumley

Got his error after installing everything Prompt outputs failed validation: UnetLoaderGGUF: - Value not in list: unet_name: 'flux1-dev-Q8_0.gguf' not in ['Wan2.2-T2V-A14B-LowNoise-Q4_K_S.gguf'] DualCLIPLoaderGGUF: - Value not in list: clip_name1: 'ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors' not in [] - Value not in list: clip_name2: 't5-v1_1-xxl-encoder-Q8_0.gguf' not in [] VAELoader: - Value not in list: vae_name: 'ae.safetensors' not in ['qwen_image_vae.safetensors', 'wan_2.1_vae.safetensors', 'taesd', 'taesdxl', 'taesd3', 'taef1']

Ian Brumley

After installing the models (This should be included with the Auto Installer, but I figured it out so I guess it's ok...), I get the error: mat1 and mat2 shapes cannot be multiplied (512x4096 and 10240x4096)

Frank B

Dude your last few All in one installs for Quen and Wan are simply not downloading and installing the models for me, is this because Im not installing to a c:/ path?

Andries van der Wijk

Fastfilmgrain and FastLaplacianSharpen nodes are missing. Also, updating ComfyUI breaks the whole thing, so don't do that.

ti

Same here. Any solution?

Visalyar

I'd like an answer too, using the default prompt. Apparently Qwen gives black outputs with sage attention but I disabled and still getting them. Edit: ok I feel dumb now, the 4 step lora wasn't downloaded. It ran fin but skipped the lora as it wasn't there. Check to see if you have the correct lora name

Charles Hopper

Its working for me beautifully except the images never come out clear for me. Extremely grainy, but other than looking like something blown up to much in MS paint, the output follows the prompt perfectly. I know its got to me something on my end, I'm a newb after all, but I have reinstalled following the video 3 times.