There are scripts that I use when I create images by using anima in diffusers or make models.
These scripts are based on hdae/diffusers-anima.
I thank hdae and claude.
When you make images by using hdae/diffusers-anima module, you have errors to load lora and checkpoint into the pipeline.
There was a small mistake in their code. Then I modified hdae/diffusers-anima/src/diffusers_anima/loaders/lora_pipeline.py and hdae/diffusers-anima/src/diffusers_anima/pipelines/anima/loading.py.
When you overwrite their codes in modyfied codes, you can images by using hdae/diffusers-anima module.
When you don't know it well, please run modify.py.
I make svd_merge_lora.py of kohya-ss/sd-scripts to run for anima model.
I make extract_lora_from_models.py of kohya-ss/sd-scripts to run for anima model.
It is a script that burns loras in a checkpoint.
It is a script that merge anima checkpoints. This script supports block merge, but I do not understand it well.
It is a script that the ckpt that doesn't be loaded in diffusers_anima makes to be loaded. If you modified diffusers_anima by my suggestion, this is not necessary.
I introduced hdae/diffusers-anima module to mokucola. But you need an abundance of time to make images in Google Colab (you need ~6 minutes to make a 1024-1024 image.), because T4 GPU doesn't support bfloat16.