Infill with I2V-A14B?

#17
by noisefloordev - opened

The I2V section of the paper (5.1.1) mentions "we incorporate multiple tasks, i.e., image-to-video generation, video continuation, first-last frame transformation, and random frame interpolation". Does frame interpolation work with the I2V-A14B model?

Comfy does I2V (start frame only) and start-to-end interpolation natively, but I've never been able to get random frame interpolation to work, eg. providing a full video and masking just a range of frames. Is there any working code or Comfy workflow that does this?

I know that the Wan2.2-VACE-Fun-A14B can do this natively. But that model is based on the T2V model, not the I2V model, so you can't use a LoRA trained for the I2V with it for editing. Being able to use the same model for frame replacement and joining clips would be really useful.

Sign up or log in to comment