nphSi

/

Z-Image-Lora

like 127

I did not really follow Anima and there is no support for it in OneTrainer yet because its not available in HF diffusers which is required for OT support. See https://github.com/Nerogar/OneTrainer/issues/1278
I guess output would have a cell shading look at best. Anyway i will test it when support arrives.

I think one of the biggest advantages of Anima is that it uses relatively little VRAM for high-resolution images.

@nphSi

I guess output would have a cell shading look at best.

Anima has realism in its dataset and does it extremely well, especially finetuned. It's nothing like Illustrious.

Looks like the OneTrainer issue has been closed as well.

+1 for looking into it, if you have time. I think you'll be pleasantly surprised by anima.

Did some first tests. I have no clue about Anima inferencing and training so there is probably much room to improve. Samples are made with Euler, Beta on Anima Turbo 10 Steps. This is probably completely wrong too...

• Download:
  https://huggingface.co/nphSi/Z-Image-Lora/blob/main/temp/AN%20Shawnee%20Smith%20(vrtlShawneeSmith).safetensors

• Samples:
  

Maybe Anima is best suited only for characters from video games, anime, cartoons, etc.

With a good finetune for realism i am pretty sure it could beat SDXL. Will check for this later and post if i have anything to show...