HunyuanVideo LoRA
This can generate fluid art styled animations, trigger words are flu1dart and in a fluidart style. Tested at weights between 0.8 and 1.0.
Example Prompt:
(neon glow melting over a planet like saturn, galactic explosions behind the flaming crescent backlit, in a fluidart style, flu1dart)
beautiful combination of multiple colors, green yellow red blue amber all glowing in streaks like a complex woodgrain pattern or layers of a planet like Jupiter, fluid art mixtures and radiant colors fast moving liquidTrained with https://github.com/tdrussell/diffusion-pipe/, using the same image dataset as the one for the Fluid Art Flux LoRA.
flux LoRA
Alternative versions trained at different steps are available and may produce different results, v1's at 1800 steps works very well at higher strengths.
v1.1 works with Schnell and Dev. Tested on latest ComfyUI & Forge.
Article with ComfyUI workflow & training information: https://civarchive.com/articles/6706
Description
FAQ
Comments (12)
Very nice, thank you, may I ask you how many images one needs to train such model?
Thank You! For this I used 54 images, and trained with ai-toolkit - https://github.com/ostris/ai-toolkit
There's a good tutorial here also: https://www.youtube.com/watch?v=HzGW_Kyermg
@neurodiculous Thank you!
Created an article that went through a few of the training details if interested, but it was mostly through that tutorial video above: https://civitai.com/articles/6706
can you share a workflow? i can't get V1 working, and i am using the trigger
Here's an article with a workflow attached, hope you get it working! Let me know if it still doesn't work after this, Thanks! - https://civitai.com/articles/6706
Looks really cool. I've only done character model Loras, any advice on how to do a decent style model?
Thanks! I just published article on the process I took, although I'm not sure if I'm the best to provide suggestions, this is my first Flux LoRA, there's probably a lot of room for improvement 😅 - https://civitai.com/articles/6706
Preview of a new version to be released soon, v1.1: https://civitai.com/posts/5474672
does it work with schnell?because i cant trigger any lora on comfy ui
It was trained specifically for Dev, but I just tried a new version I'm about to release with the default Schnell and it actually worked! Make sure you're on the latest version of ComfyUI and place a Load LoRA node before the "BasicGuider" and "CLIP Text Encode (Prompt)" when using these examples: https://comfyanonymous.github.io/ComfyUI_examples/flux/
Let me know if you have any issues and I can upload a workflow later of the same thing.
Also since this wasn't ever intended for Schnell, it might produce unusual results, but at least it may work a little! I'll see if I can train a Schnell version later at some point as well.
Just published v1.1-alpha to try, I may eventually remove the -alpha, but I also tested it with Schnell and it works! Let me know if you have any issues with the new version.