Unofficial AnimateDiff v3 motion lora trained on warped noise from https://github.com/Eyeline-Research/Go-with-the-Flow
Use in comfyui (example workflow included) with https://github.com/kijai/ComfyUI-VideoNoiseWarp to generate the warped noise from videos. Reduce lora strength to work with degraded noise, 1.0 strength for no degradation.
Training details:
rank 128
learning rate 1e-5
optimizer AdamW8bit
steps 12000 (0.25 epoch)
resolution 512x512
frames 16
noise offset 0.05
caption dropout 0.1
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
HF Inference API was unable to determine this model's library.