AI‑Toolkit WAN LoRA Training Ultimate Pack
Hey everyone! I’ve made a new video for the “ULTIMATE WAN LORA TRAINING AI‑TOOLKIT WEBUI.” Hands‑down the BEST AI model training I have ever made! VIDEOS & IMAGES!!!
Just fricking insane! This really is the best Ai model currently available and Ai-Toolkit makes training a Lora super ultra easy as well, basically the ultimate combination, perfect for complete beginners! With it you can train any LoRA you want for WAN: people, styles, and even full video LoRAs… it’s just nuts!
You can watch the full video here: https://youtu.be/lRg5sPBXTZE
I’ve also make a 1‑click installer for both Local AND RunPod setups just for Patreon supporters right here: https://www.patreon.com/posts/134551276
Link for the manual install: https://pastebin.com/UQNL51Z2
Down below you will also find the WAN_XY-PLOT_LORA_TESTING.json for free for you to use!
Below you’ll find the LoRAs I trained in the video—free for you to play with:
Margot Robbie LoRA
----Taken down (by myself just in case) send me a message for more info
Jojo's Bizarre Adventure Style LoRA
Vlog Style LoRA
----Still in training, come back later....
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😉!
Comments (25)
Hi. How do I start the ai-toolkit program locally after installed it on my pc? Thanks!
XY plot not working for me ttN advPlot images object of type 'NodeOutput' has no len() Tried to troubleshoot with perplexity. Seems to suggest that at this point just not compatible. I am running this on a 5090 Thoughts?
Its honestly kinda unbelievable how willingly Wan 2.x seems to learn. Compared to training on one of the sdxl finetunes - this is just a breeze! Thanks for the awesome guide
Wan 2.2 support? :)
As far as I can see, they are just straight up intercompatible. At least my 2.1 character loras worked fine with 2.2 with no modifications
As I couldn't get the AI Toolkit to run locally, I tried runpod. However, after following all your instructions, when I pasted the 2 lines of code to install the toolkit I get this message "chmod: cannot access 'WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD.sh': No such file or directory bash: ./WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD.sh: No such file or directory root@cdd7502691c6:/workspace# "
Hi, is it better than diffusion pipe?
Hi K, I'm also getting the ModuleNotFoundError: No module named 'dotenv' error. I have downloaded “Build Tools for Visual Studio 2022” from https://visualstudio.microsoft.com/visual-cpp-build-tools/ “Desktop development with C++” is checked and Build Tools is installed and updated. I then did a clean install using your one-click installer in to a new folder. Not sure why this is not working
I tried specifically installing dotenv and adding the path to the system environment variables and restarting but still the same error.
Getting the error "validate_prompt() missing 1 required positional argument: 'prompt'" pm tje Lora Testing workflow every time, I of course, have a prompt in. What am I doing wrong? # ComfyUI Error Report ## Error Details - **Node ID:** 77 - **Node Type:** ttN advPlot images - **Exception Type:** TypeError - **Exception Message:** validate_prompt() missing 1 required positional argument: 'prompt' ## Stack Trace ``` File "/workspace/ComfyUI/execution.py", line 427, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/ComfyUI/execution.py", line 270, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/ComfyUI/execution.py", line 244, in _async_map_node_over_list await process_inputs(input_dict, i) File "/workspace/ComfyUI/execution.py", line 232, in process_inputs result = f(**inputs) ^^^^^^^^^^^ File "/workspace/ComfyUI/custom_nodes/comfyui_tinyterranodes/ttNpy/tinyterraNodes.py", line 2662, in plot plot_image, images, samples = plotter.xy_plot_process() ^^^^^^^^^^^^^^^^^^^^^^^^^
same here
I did exactly as you suggested, but still getting: Traceback (most recent call last): File "C:\AITOOLKIT\ai-toolkit\run.py", line 6, in from dotenv import load_dotenv ModuleNotFoundError: No module named 'dotenv'
Same thing as I said before, this error is due to an incorrect install “Desktop development with C++" really make sure you got it installed correctly. Download “Build Tools for Visual Studio 2022” https://visualstudio.microsoft.com/visual-cpp-build-tools/ During the setup, tick “Desktop development with C++” then click Install Then restart your pc, and redo the install in a new folder.
What image aspect ratio/size should we use for training Lora’s?
Doesn't matter, just make sure your images are of high quality and high resolution , preferably above 1024x1024
Hi K, I think you forgot something very important in the video tutorial : Max Step Saves to Keep, is at 4 by default. Meaning even though you can see the progression with the image samples, you're gonna get only the last 4 epochs so it better not be overtrained... or you are fucked. This is my first lora training using Runpod, it's still going while I'm typing this, we'll see x) I'm training a quite difficult anime style with only 33 images, so far I also noticed something else : The LightX2v lora might help giving very clean results in only 10 steps, but it kills the style. 20 steps without LightX2v is a lot closer to the original style
That is very true and that was a huge mistake on my part, I realized it once the video was already uploaded.... also for the style, I use the fusionx model and lightx lora at 0.3-0.4 weight to add more contrast. I also found that using the normal model with fusionx lora at 1.5 weight and lightx lora at 0.8 gives great looking results depending on the style, so I suggest playing a little bit with the loras weights
Hi, can you guide us how to make the downloading process work? I see alot of people including me couldn't get the model downloaded after after starting the Job. It always got stuck and will not resume at any point. You help will be much appreciated.
what do you mean? what error you get?
I'm curious if anyone tried training a LoRA on a character or concept with images as a single frame, then using it to generate a video?
I mean look at the gif from this page: https://www.patreon.com/posts/1-click-install-134551276 you can train a character lora and then use that lora to either generate a video or generate an image then use image to video to make your video
Is this a JoJo reference 🤭😉
Oh yeah😏
Since I have Linux do I use the rubpod.sh version?
No it's for runpod only, altough it might work on linux if you have an already existing comfyui install