It seems that currently only 512x512x16 videos can be generated.

[xianienie]

“ultralytic” in requirements.txt should be “ultralytics”
The example you provided is based on sd1.5 iclight animatediff, but it seems that xformer memory optimization is not being used. Currently, it can only generate 512x512x16 videos, which require 23GB of VRAM. Have you considered developing it into a plugin for comfyui or webui? Also, have you tried using the xl model for animatediff or other sd1.5 models?

[YujieOuO]

Thank you for your attention and for pointing out the typo in requirements.txt, We have already fixed it.
light-a-video currently supports animatediff with 16 frames. We have also tested it with CogVideo and LTX-Video, which work but still require some optimization on time efficiency. We highly appreciate your valuable suggestions and will consider more VDM models in future versions.

[YujieOuO]

Light-A-Video suppots Wan2.1 now, have fun.