How it feels to have a lightweight anime model that doesn't use SDXL as a base.
This thing trained surprisingly fast AND well. Use the tag @shadman just like any other anima artist tags.
Check out my other style LORA (Anima version coming soon): https://civarchive.com/models/1430075/illustrious-blackwhiplash-bwl-style
Description
FAQ
Comments (7)
If it ain't SDXL, what's the base model of Anima? Thank you 🙌
It's actually NVIDIA's Cosmos-Predict2-2B-Text2Image!
@Cryphe Thanks!
What trainer did you use? I really don't wanna have to install Linux if it's something like diffusion pipe
I used this branch of SD-Scripts https://github.com/kohya-ss/sd-scripts/tree/feat-anima-polish
great LORA!
just wondering how you managed to train it using sd-scripts? i tried their branch https://github.com/kohya-ss/sd-scripts/tree/feat-anima-polish using a previous data set i used for illustrious but the output lora didn't seem to load in comfyui.
not sure if theres anything special you have to do other than setting --network_module networks.lora_anima and the paths to the modle/vae/qwen3
Tbh I just followed the docs. This specific branch has a docs page that explains how to setup your training args.