Home Creators Posts Import Register Favorites Logout

Downloads

Content

Ostris AI-Toolkit installs automatically on Windows, RunPod, SimplePod and Massed Compute and is now patched to train on custom single-file base models.

Installer V19 uses Torch 2.13 and CUDA 13 with pre-compiled wheels (xformers, Flash Attention, Sage Attention, TorchAO) for all GPUs on Python 3.12. Cloud installs are fully automatic with Cloudflared access, Windows also needs Node.js 22, and a full training tutorial video is linked.

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 Installer Zip File Torch 2.13, CUDA 13 : AI_Toolkit_SECourses_v19.zip

  • [click here to choose a membership and Join to download zip files]

  • Python 3.12

  • Installed with pre-compiled wheels for : mslk, xformers, flash_attn, sageattention for both Windows and Linux with all CUDA 13 features for Torch 2.13 and they are abi3 for python 3.10, 3.11, 3.12, 3.13

  • Automatic install on Windows, RunPod, Massed Compute (or local linux machines), Windows

    • All Linux installers also updated for Python 3.12 venv

  • Always installs latest version and patches necessary stuff automatically https://github.com/ostris/ai-toolkit

  • Full tutorial how to use and train published 4 December 2025 : https://youtu.be/ezD6QO14kRc

    • Please follow above tutorial to learn how to use AI Toolkit on Windows, RunPod (same on SimplePod) and Massed Compute

  • For Windows you need to also install Node JS for this app : https://nodejs.org/en/download

  • Installation on SimplePod, RunPod and Massed Compute are fully automatic, no manual setup, etc. needed

  • For Windows : 0_Windows_Install.bat > 1_Windows_First_Time_UI_Setup_Run.bat > 2_Windows_Start_App.bat

  • For RunPod and SimplePod : Runpod_SimplePod_AI_Toolkit_Instructions_READ.txt

  • For Massed Compute : Massed_Compute_Instructions_READ.txt

30 September 2026 Update - V19

  • As you know AI Toolkit still does not support single safetensor files as a base model

  • Therefore, normally, you are not able to train on base models hosted on CivitAI, HuggingFace, etc

  • But today, I upgraded our installer to patch AI Toolkit automatically to support custom base models

  • So far tested on Z Image Turbo but should work on all models

  • As you see in below image give full path and start using

    • Make sure accurate model architecture is selected

    • So select everything accurately first and then replace base model path that it puts from templates

  • To make it work, download latest zip file and make a fresh install or overwrite all files and run 3_Windows_Update_App.bat file

12 September 2026 Update - V18

  • All Windows, RunPod, MassedCompute, SimplePod installers upgraded and fixed

    • All tested with actual training

  • Windows update bat file fixed

  • Fresh install is recommended

  • AI Toolkit developer is insisting not fixing TorchAO so I written a patcher and now my installer installed AI Toolkit uses Torch AO 0.19 - compiled by me fresh and it really helps

  • So we currently install with all latest compiled very best libraries

  • Please have installed Python 3.12 and NodeJS 22 : https://nodejs.org/dist/v22.20.0/node-v22.20.0-x64.msi

  • RunPod, Massed Compute and SimplePod installations are fully automatic

    • Just follow Massed_Compute_Instructions_READ.txt , Runpod_SimplePod_AI_Toolkit_Instructions_READ.txt

  • RunPod, Massed Compute and SimplePod now have automatic CloudFlared start feature too to access from secure public URL in your computer

30 July 2026 Update - V17

  • Latest AI Toolkit installers had broken Linux installation and this issue fix

  • Also latest AI Toolkit has some bugs our installers now auto fixes them robustly until the official author fixes

  • We still always install from latest official repo mainstream

11 July 2026 Update - V14

  • Fresh install is very much recommended otherwise you may get errors

  • I have started upgrading all of our apps into latest Torch 2.13 and CUDA 13

  • For this, I have pre-compiled the following wheels with all CUDA 13 features and with all CUDA archs : mslk, xformers, flash_attn, sageattention, torchao

  • All these libraries are properly compiled with abi3 thus works on Python 3.10, 3.11, 3.12 and 3.13

  • So I have updated the Ostris AI Toolkit installers with newest libraries

  • Moreover, I have improved the installation process thus now it is many times faster than before

  • It will automatically generate Python 3.12 venv and install everything there, if you don't have 3.12, it will use your default python

    • Make sure to have 3.12.10 it is best

    • Make sure to have nodejs 22+ installed - 22 preferred and tested

  • Follow requirements tutorial (below) and latest AI Toolkit tutorial : https://youtu.be/ezD6QO14kRc

  • For JSON prompt generation we have new tutorial and tool that is useful for Ideogram 4 model training

Windows Requirements

Ostris AI Toolkit Training Tutorial (for Z Image Turbo and others)

Video link : Full tutorial how to use and train published 4 December 2025 : https://youtu.be/ezD6QO14kRc

  • This section is made for Qwen Image training comprehensive tutorial video

  • First of all follow below requirements tutorial and make sure all requirements properly installed or updated

  • Download the latest zip file from either attachments (down below) or Latest Zip File (at the top of the post)

  • The follow the tutorial video every step

Auxiliary tools are as below

Click on Images to See Full Size - Screenshots

Related posts

All posts in this topic: LoRA and Fine-Tuning Training: Kohya, Musubi Tuner, OneTrainer, AI Toolkit · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

Z-Image Turbo LoRA training with AI Toolkit and Z-Image ControlNet Full Tutorial for Highest Quality

Z-Image Turbo LoRA training with Ostris AI Toolkit + Z-Image Turbo Fun Controlnet Union + 1-click to download and install the very best Z-Image Turbo presets. In this tutorial, I will explain how to setup Z-Image Turbo model properly in your local PC with SwarmUI and download models and use them with highest quality via ready presets. Moreover, I will show to install Z-Image Turbo Fun Controlnet Union to generate amazing quality images with ControlNet preprocessors. Furthermore, I will show how to 1-click install AI Toolkit from Ostris and train Z-Image Turbo model LoRAs with highest quality configs made for every GPU like 8 GB GPUs, 12 GB GPUs, 24 GB GPUs and so on. I did a massive research to prepare these Z-Image Turbo model training configurations. 👇 Links & Resources Mentioned: Download SwarmUI & Models: [ https://www.patreon.com/posts/Download-SwarmUI-Models-114517862 ] Ostris AI Toolkit & Configs (SECourses Version): [ https://www.patreon.com/posts/Ostris-AI-Toolkit-140089077 ] Ultimate Batch Image Processing App: [ https://www.patreon.com/posts/Ultimate-Batch-Image-Processing-App-120352012 ] SwarmUI with ComfyUI Backend Windows Tutorial: [ https://youtu.be/c3gEoAyL2IE ] SwarmUI with ComfyUI Backend RunPod and Massed Compute Cloud Tutorial: [ https://youtu.be/bBxgtVD3ek4 ] ⏱️ Video Chapters: 0:00 Introduction to Z-Image Turbo Model 0:54 FP8 Scaled Version 5.7GB for Low VRAM 1:10 ControlNet Union with Z-Image Turbo 1:30 LoRA Training with Ostris AI Toolkit 2:00 Default vs Custom Training Preset Quality Comparison 3:00 RunPod Cloud Training Preview 3:40 MassedCompute Cloud Training Preview 4:16 Downloading Z-Image Models via SwarmUI 5:00 Z-Image Turbo Core Bundle & ControlNet Files 5:58 FP8 Scaled Model & Musubi Tuner Converter 7:13 Updating ComfyUI for Sage & Flash Attention 8:13 Updating SwarmUI & ControlNet Preprocessors 8:52 Updating & Importing Latest SwarmUI Presets 9:20 Generating with Quality 2 Fast Preset 10:48 Generating with Quality 1 Upscale Preset 11:35 Quality 1 vs Quality 2 Visual Comparison 12:13 Setting up ControlNet Input & Aspect Ratio 13:41 ControlNet Strength Settings & Canny Test 15:26 Using Depth Preprocessor with Z-Image 15:58 Coloring Lineart Drawings with ControlNet 16:58 Lineart Preprocessing Comparison 17:50 Ostris AI Toolkit Installation Prerequisites 19:12 Installing Ostris AI Toolkit on Windows 20:02 First Time UI Setup & Launching Interface 21:04 Loading Custom Training Configs 21:38 Creating a New Dataset Structure 22:24 Ultimate Batch Image Processing App Install 23:17 Dataset Prep Stage 1: Auto-Zooming with SAM2 26:08 Dataset Prep Stage 2: Resizing to Exact Resolution 28:12 How to Select Best Training Images 30:24 Importance of Emotions & Angles in Datasets 31:44 Z-Image Resolution & Aspect Ratio Rules 33:21 Configuring Training Parameters & Epochs 36:52 Resolution Impact on Training Speed 37:46 Starting the Training Job on Windows 38:39 Monitoring Training Progress & VRAM 39:43 Checkpoint Generation Settings 40:40 Resuming Training from Last Checkpoint 42:09 Training Speeds on RTX 5090 vs 4090 vs 3060 43:01 Training Quality: Default vs Custom Preset Comparison 44:21 Testing LoRAs with SwarmUI Grid Generator 46:04 Fixing ControlNet Error in Grid Generation 47:09 Comparing Generated LoRA Checkpoints 47:38 Using Trained LoRA with ControlNet Union 48:10 RunPod: GPU Selection & Template Setup 50:32 RunPod: Port 8675 Config & Initialization 51:36 RunPod: Uploading Installation Files 52:01 RunPod: One-Click Installation Command 54:07 RunPod: Starting AI Toolkit & Proxy Connection 54:38 RunPod: Uploading Dataset via Interface 55:32 RunPod: Starting the Training Job 56:24 RunPod: Speed & Cost Analysis 57:28 RunPod: Auto-Stop Command Setup 58:24 MassedCompute: GPU Selection & Coupon Code 1:00:16 MassedCompute: ThinLinc Client Setup 1:01:21 MassedCompute: Transferring Files to Shared Folder 1:02:55 MassedCompute: Installation Command 1:05:49 MassedCompute: Connecting via Public URL 1:06:54 MassedCompute: Starting Training Job 1:08:43 Downloading Checkpoints & Stopping Instance 🚀 Master Z-Image Turbo & LoRA Training: The Ultimate Guide! In this comprehensive tutorial, I show you how to generate ultra-realistic images in seconds using the lightweight Z-Image Turbo model. We cover everything from 1-click installation on SwarmUI (ComfyUI backend) to mastering ControlNet Union for precise image control. But that’s not all! I also reveal how to train your own high-quality Z-Image Turbo LoRAs using the Ostris AI Toolkit. I have developed a custom training preset that significantly outperforms the default settings—you have to see the comparison to believe it! Whether you are on a local PC, RunPod, or MassedCompute, this guide has you covered. 🔥 What You Will Learn: Z-Image Turbo Setup: How to run this fast, 6GB model (FP8 included) on almost any GPU. ControlNet Mastery: Use Canny, Depth, and Lineart to control your generations perfectly. LoRA Training: Step-by-step guide

Comments

Hockey

I can't start the software with Windows_First_Time_UI_Setup_Run: AI Toolkit UI - First Time Setup ============================================ This will install dependencies and build the UI for the first time. This may take several minutes... Installing Node.js dependencies... 'npm' is not recognized as an internal or external command, operable program or batch file. ERROR: Failed to install dependencies! Please check your Node.js installation and internet connection. Press any key to continue . . .

Roni Bliss

YESSS! You did it, finally you did it! ... THE BEST Training script of all time gets the SEC treatment! BIG LOVE for this brother! Thanks a million!

Hockey

I apologize.

mastermk

Here it is again, always the same mistake: Error running job: The size of tensor a (36) must match the size of tensor b (16) at non-singleton dimension 1 ======================================== Result: - 0 completed jobs - 1 failure ======================================== Traceback (most recent call last): File "S:\ai_toolkit\ai-toolkit\run.py", line 120, in main() File "S:\ai_toolkit\ai-toolkit\run.py", line 108, in main raise e File "S:\ai_toolkit\ai-toolkit\run.py", line 96, in main job.run() File "S:\ai_toolkit\ai-toolkit\jobs\ExtensionJob.py", line 22, in run process.run() File "S:\ai_toolkit\ai-toolkit\jobs\process\BaseSDTrainProcess.py", line 2010, in run self.sample(self.step_num) File "S:\ai_toolkit\ai-toolkit\extensions_built_in\sd_trainer\DiffusionTrainer.py", line 288, in sample super().sample(step, is_first) File "S:\ai_toolkit\ai-toolkit\jobs\process\BaseSDTrainProcess.py", line 370, in sample self.sd.generate_images(gen_img_config_list, sampler=sample_config.sampler) File "S:\ai_toolkit\ai-toolkit\venv\lib\site-packages\torch\utils\_contextlib.py", line 120, in decorate_context return func(*args, **kwargs) File "S:\ai_toolkit\ai-toolkit\toolkit\models\base_model.py", line 605, in generate_images img = self.generate_single_image( File "S:\ai_toolkit\ai-toolkit\extensions_built_in\diffusion_models\wan22\wan22_14b_i2v_model.py", line 83, in generate_single_image output = pipeline( File "S:\ai_toolkit\ai-toolkit\extensions_built_in\diffusion_models\wan22\wan22_pipeline.py", line 262, in __call__ latents = self.scheduler.step( File "S:\ai_toolkit\ai-toolkit\venv\lib\site-packages\diffusers\schedulers\scheduling_unipc_multistep.py", line 992, in step model_output_convert = self.convert_model_output(model_output, sample=sample) File "S:\ai_toolkit\ai-toolkit\venv\lib\site-packages\diffusers\schedulers\scheduling_unipc_multistep.py", line 628, in convert_model_output x0_pred = sample - sigma_t * model_output RuntimeError: The size of tensor a (36) must match the size of tensor b (16) at non-singleton dimension 1

Draxel

I love trying out new trainers. I don't know why though. :-) Thanks for the work on this.

ExpSoft

Hi, do you think it's possible to add ramtoch : https://github.com/lodestone-rock/RamTorch to use the CPU offloading in AI toolkit ? this allow to run this tool for trining with 10gb Vram cards from what i understood ? Thank you for the courses !

SECourses

hi AI toolkit added it already just run installer again to upgrade / update :)

Pham Don

i got this error when start train with flux may i ask how do i fix this? 'Detected no triton, on systems without Triton certain kernels will not work ostris ai toolkit'

SECourses

i just made a fresh install and triton installed. where do you see this message? did you try training any errors?

Pham Don

Hello, thank you for your respond, I tried to delete everything and re installed I think it worked. I think I was missing the requirement.txt when I run the script. I just started the model training now and didn’t get that error anymore it in the transformer step and downloading now. However I noticed it downloaded really fast around 400-600Mbs however every time it get around 80-90 percent it slow down to couple kilobytes I checked my internet and I still get avg about 400-600Mbs, may I ask if it is because of my internet or the huggingface hub? Thank you for your time

SECourses

sorry for late reply. it can be 100% related to hugging face sadly :(

Play Music

hi, have you ever thought about implementing Polynomial learning rate scheduler instead of constant? it can be done with a little edit on scheduler.py (then editing the config in advanced tab) and in my tests, it's giving slightly better results. i start with lr:0.0003 and end with 0.0001 or 0.000075, generally in 7000-8000 steps. i have around 200 images in dataset and i generally start getting good loras around 6000. since you like testing for best results, i thought you might wanna know. thanks for your hard work.

SECourses

it can be added but i don't plan to develop it :D i just had to make a quick fix until the author fixes the block swapping issue otherwise low VRAM configs were not working

Oui

Sadly I got better result using near basic preset than with yours :/ (for the 24GB) Previous setup (same dataset) : rank 32 / lower resolution 532 to 1024 / timestep : sigmoid (clearly an important part) / I've show a better and faster understanding of the face (and the whole was a fcking lot faster) ofc it could be an error on my end just wanted to share my situation

Shing Ming Lin

I'm currently learning how to train z-image lora. Following your method, I'm training it on Runpod, but I'm getting the following message. What should I do next? I believe many people have this problem. It would be best if you could release a video tutorial. [WORKER] TOOLKIT_ROOT: /workspace/ai-toolkit [WORKER] Cron worker started with interval: 1000 ms [UI] ▲ Next.js 15.1.7 [UI] - Local: http://localhost:8675 [UI] - Network: http://192.168.6.2:8675

Vicent Guardiola

Hi! I intalled your version, followed your youtube tutorial but the training is not starting. Its like "Starting job..." all the time and no logs visible anywere, do you know what is happening? I have all the other things running correctly, like comfyui, kohya and so on. Thanks so much!

Vicent Guardiola

Sorry! Now It's working, I deleted all folders related to ostris and re-installed and it's ok! thanks!!

Play Music

thanks for the video, but your settings are too heavy for my rtx 3090, because of rank 128 and 1536px. so it's not just the VRAM thing to consider i guess. on paper yes it works but with around 200 sec/iter which makes no sense. i have hi-res dataset with 200 images and descriptive captions and your settings are impossible for me. so i lowered them to rank 64 and 1280px only. this way i can reach around 12 sec/iter speed, which is still slow but doable. with default settings (resolutions), i trained my lora with 2sec/iter speed but the output was terrible with 8000 steps. i don't think z-image will ever be as good as qwen when it comes to character loras. anyway, time will tell after the base model is released. one last thing, Sigmoid should be the first choice when you train for a realistic character. i tested that too many times.

SECourses

true z image is weaker than qwen atm. their base model we will see. 200 second / iteration only happened because it used shared vram. so either your dataset aspect ratios or you use vram. therefore you need to use cpu offloading. its speed should be around 15 second / it

Taiga

error: subprocess-exited-with-error × git clone --filter=blob:none --quiet https://github.com/jaretburkett/easy_dwpose.git 'z:\Temp\pip-req-build-vgb8rgh4' did not run successfully. │ exit code: 128 ╰─> No available output. note: This error originates from a subprocess, and is likely not a problem with pip. ERROR: Failed to build 'git+https://github.com/jaretburkett/easy_dwpose.git' when git clone --filter=blob:none --quiet https://github.com/jaretburkett/easy_dwpose.git 'z:\temp\pip-req-build-vgb8rgh4'

Taiga

god I wish you tutorials/lectures wern't over 30 minutes long

aprisma

Hi. May I ask with what tool did you make the picture of "optimal GPU conf training tiers"?

SECourses

I used Nano Banana 2 - the Google model that is best atm for info graphics

SECourses

i recommend to check chapters to learn the parts you need quickly

Play Music

thanks for the tip. i've never had to use cpu offloading in ai-toolkit so i always ignored that option. after testing a few options, i ended up with 36% cpu offloading with 1536px. now it trains with 12sec/iter speed which is so acceptable for that resolution and 128 rank with 3090. using ostris' de-turbo model helped with my lora results as well. this model seems better than turbo+adapter at training. still not a qwen 2509 though, but surely looking more life-like. i finally started to see big potential in z-image, but not for the upcoming base model, maybe its next version. in this version, whatever lora i apply, i can see an 'asian person' like a shadow behind the character somehow.

SECourses

so you are using on massed compute? well we didnt change anything since video but i will test it. are you sure you are selecting SECourses image? you are trying z image turbo?

100Sachen80

Hi Furkan, how can i prevent the generated image to have the whole person put into the face again, when using ? i use the preset 1 with upscaling. i fiddled around with the upscaling and segmentation settings but did not get it fixed. the segmentation always puts the whole person into to face, seems to happen only when using upscaling and segmentation together. Just realized that i have same problem also in wan 2.2 image generation with the wan image generation preset. Happens when image generation with trained loras.

SECourses

probably runpod template error. replying from discord

Moarr Social

In RunPod, there is already a template for OstrisAI. Why not just use that instead of using the script that you have provided?

SECourses

i install latest fixed version of the app every single time. that template probably getting outdated so quickly. moreover we use official pytorch template extremely lightweight immediately starts

Moarr Social

Is there a way to speed up the training? With the default config, 4,000 steps on an RTX 4090 is estimated to take ~6 hours (~5.3 s/iter). Would upgrading the GPU meaningfully reduce this, or is the bottleneck elsewhere?

SECourses

yes reduce training resolution to 1024x1024 and it will become like 3x 4x faster

abaj006

Does your method use the de-distillation training adapter? Thanks

SECourses

yes it is using adapter v2 atm in configs

Joshua Thomas

Cannot get it to run properly, Says torchaudio is not installed and even after installing the requirements.txt it gives the same error, When using the easy installer released by the tookit dev I do not get the error. When following your steps with the easy install toolkit I get this error "AttributeErrorAttributeError: : 'FakeTextEncoder' object has no attribute 'config''FakeTextEncoder' object has no attribute 'config'" when training is about to fully finish. No matter what installer I use I cannot get your stuff to work

SECourses

so easy. first follow this video to properly install python it is mandatory : https://youtu.be/DrhUHnYfwC0 have python 3.10 installed then follow this video to properly install ai toolkit and make a z image turbo training : https://youtu.be/ezD6QO14kRc i literally made a fresh install and test training in total 5 minutes

sh fung

Hello, I can't start the software with Windows_First_Time_UI_Setup_Run: AI Toolkit UI - First Time Setup _____________________________________________________ ⚠ `devIndicators.buildActivity` is deprecated and no longer configurable. Please remove it from next.config.ts. ⚠ Warning: Next.js inferred your workspace root, but it may not be correct. We detected multiple lockfiles and selected the directory of B:\AI_Toolkit_SECourses_v10\package-lock.json as the root directory. To silence this warning, set `outputFileTracingRoot` in your Next.js config, or consider removing one of the lockfiles if it's not needed. See https://nextjs.org/docs/app/api-reference/config/next-config-js/output#caveats for more information. Detected additional lockfiles: * B:\AI_Toolkit_SECourses_v10\ai-toolkit\ui\package-lock.json ▲ Next.js 15.5.9 - Experiments (use with caution): · serverActions · middlewareClientMaxBodySize: "100gb" Creating an optimized production build ... ⚠ Compiled with warnings in 5.0s ./node_modules/systeminformation/lib/cpu.js Module not found: Can't resolve 'osx-temperature-sensor' in 'B:\AI_Toolkit_SECourses_v10\ai-toolkit\ui\node_modules\systeminformation\lib' Import trace for requested module: ./node_modules/systeminformation/lib/cpu.js ./node_modules/systeminformation/lib/index.js ./src/app/api/cpu/route.ts ✓ Compiled successfully in 5.8s Skipping validation of types ✓ Linting ⨯ Next.js build worker exited with code: 3221225477 and signal: null ________________________________________________________ It seems the Next.js is not working to start the AI Toolkit. May I know if there is any solution please? I have reinstall the Next.js with the latest version already, but no luck to start the AI toolkit. Thank you!

sh fung

Some more error message when I start the ai toolkit and hope this helps: Starting application... > ai-toolkit-ui@0.1.0 start > concurrently --restart-tries -1 --restart-after 1000 -n WORKER,UI "node dist/cron/worker.js" "next start --port 8675" [WORKER] TOOLKIT_ROOT: B:\AI_Toolkit_SECourses_v10\ai-toolkit [WORKER] Cron worker started with interval: 1000 ms [UI] ▲ Next.js 15.5.9 [UI] - Local: http://localhost:8675 [UI] - Network: http://10.5.0.2:8675 [UI] [UI] ✓ Starting... [UI] ⨯ Failed to load next.config.ts, see more info here https://nextjs.org/docs/messages/next-config-error [UI] ReferenceError: False is not defined [UI] at Object. (B:\AI_Toolkit_SECourses_v10\ai-toolkit\ui\next.config.compiled.js:13:24) [UI] next start --port 8675 exited with code 1 [UI] next start --port 8675 restarted [UI] ▲ Next.js 15.5.9 [UI] - Local: http://localhost:8675 _________________________________________ Thank you!

Shing Ming Lin

The Z Image base has been released. Is the training method the same as Turbo?

SECourses

I am working to add it to kohya will research there hopefully In our musubi trainer app

BigVI

I could not get this to run on SimplePod, I used this one "AI_Toolkit_SECourses_v15.zip" it seems like everything installed fine, but when I ran your start script "npm run start" it kept throwing this error in a loop: "[UI] TOOLKIT_ROOT: /workspace/ai-toolkit [WORKER] TOOLKIT_ROOT: /workspace/ai-toolkit [UI] AI Toolkit UI: http://localhost:8675 (4 file server workers) [WORKER] Cron worker started with interval: 1000 ms [UI] ▲ Next.js 15.5.21 [UI] - Local: http://localhost:8675 [UI] - Network: http://localhost:8675 [UI] [UI] ✓ Starting... [UI] [Error: ENOENT: no such file or directory, open '/workspace/ai-toolkit/ui/.next/prerender-manifest.json'] { [UI] errno: -2, [UI] code: 'ENOENT', [UI] syscall: 'open', [UI] path: '/workspace/ai-toolkit/ui/.next/prerender-manifest.json' [UI] } [UI] Next.js exited with code 1, shutting down file server [UI] node dist/cron/fileServer.js start --port 8675 exited with code 1 [UI] node dist/cron/fileServer.js start --port 8675 restarted" Thanks!

SECourses

fixed with v16. ty for reporting

BigVI

Thank you sir

Xyzzx

Maybe I don't understand how patreon works but why do all of your "newest" posts having comments from up to and over a year ago?!?!

BigVI

I received this error when starting training: "Running 1 job /workspace/ai-toolkit/venv/lib/python3.12/site-packages/torchao/quantization/quant_api.py:1745: SyntaxWarning: invalid escape sequence '\.' * regex for parameter names, must start with `re:`, e.g. `re:language\.layers\..+\.q_proj.weight`. W0731 14:35:04.560000 12508 venv/lib/python3.12/site-packages/torch/utils/_pytree.py:630] is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release. W0731 14:35:04.870000 12508 venv/lib/python3.12/site-packages/torch/utils/_pytree.py:630] is an Enum subclass and is now natively supported by torch.compile as an opaque value type. Calling register_constant() on Enum subclasses is deprecated and will be an error in a future release. Unable to import `torchao` Tensor objects. This may affect loading checkpoints serialized with `torchao` Error running job: No module named 'torchaudio' Error running on_error: cannot access local variable 'job' where it is not associated with a value ======================================== Result: - 0 completed jobs - 1 failure ======================================== Traceback (most recent call last): File "/workspace/ai-toolkit/run.py", line 147, in main() File "/workspace/ai-toolkit/run.py", line 132, in main raise e File "/workspace/ai-toolkit/run.py", line 119, in main job = get_job(config_file, args.name) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/ai-toolkit/toolkit/job.py", line 28, in get_job from jobs import ExtensionJob File "/workspace/ai-toolkit/jobs/__init__.py", line 1, in from .BaseJob import BaseJob File "/workspace/ai-toolkit/jobs/BaseJob.py", line 5, in from jobs.process import BaseProcess File "/workspace/ai-toolkit/jobs/process/__init__.py", line 6, in from .TrainVAEProcess import TrainVAEProcess File "/workspace/ai-toolkit/jobs/process/TrainVAEProcess.py", line 20, in from toolkit.data_loader import ImageDataset File "/workspace/ai-toolkit/toolkit/data_loader.py", line 21, in from toolkit.config_modules import DatasetConfig, preprocess_dataset_raw_config File "/workspace/ai-toolkit/toolkit/config_modules.py", line 7, in import torchaudio ModuleNotFoundError: No module named 'torchaudio' Job process terminated with exit code 1." Thanks!

SECourses

sorry for this error. i added torchaudio with v17. i will test trainings as well but havnet yet

BigVI

Great, thanks!

SECourses

v17 works just tested on simple pod with z image model

BigVI

Yes it works! Thanks for testing also!

nurtu

Any news as to how to train MinimaxH3 Loras or finetunes? For ex I want to train a character replacer lora

SECourses

character replacer logic same as previous edit lora trainings. so you would need input image + a prompt + a video = character replaced video in training dataset.