frosting/star-atlas-v1

Star Atlas V1

Star Atlas is an animatediff video model based on Fluffyrock Unbound! So prompts that work well there should work well here too.

Make sure you have "use_existing" beta schedule on the animatediff nodes, with zsnr and v-prediction.

Recommended resolutions include 576x576, 704x704, 832x832. You can also add 64 to one of the dimensions and subtract 64 from the other up to 64*3 = 192. So 384x768, 512x896, or 640x1024 would be recommened as well.

Example ComfyUI workflows are included in json format. The txt2vid workflow requires ComfyUI-AnimateDiff-Evolved and comfyui-prompt-control. The img+txt2vid workflow additionally requires ComfyUI-Advanced-ControlNet.

The DomainAdapter model should be put in ComfyUI/models/loras. The MotionModule should go in ComfyUI/models/animatediff_models. The SparseCtrl model should go in ComfyUI/models/controlnet. Fluffyrock Unbound goes in ComfyUI/models/checkpoints as usual, and the boring_e621_unbound embeddings (used in the example workflows) go in ComfyUI/models/embeddings.

You can use this model without any setup at https://frosting.ai :)

The model is licensed for non-commercial use under CC BY NC SA 4.0