What's this?

Most video generation AIs output at 16 or 24fps. But in anime production, a single drawing is often held for 2 or 3 frames.

This isn’t just about saving labor — animating on twos or on threes can create a very different rhythm, sometimes even more dynamic than full 24fps.

So, 24fps isn’t always a superior version of 12fps or 8fps.

I built a workflow that converts animation into "on twos" or "on threes." Instead of lowering the frame rate of the whole video (which just looks choppy), this workflow applies the effect only to the subject, while keeping everything else smooth.

Workflow

• The input video is split into background / subject / optional foreground layers.
• The subject (and optional foreground) are segmented with Florence-2 + SAM 2.1, then refined with SDMatte for clean alpha matting.
• For the background, masked areas are filled using Wan2.1 VACE inpainting.