AI frame interpolation API: 24 to 60 fps on Sume

Sume's docs do not describe an AI frame interpolation model. Two documented paths touch frame rate: Timeline 1.0 accepts output.fps of 24, 25, 30 or 60 and meets a different source rate by repeating or dropping frames, and the video filter allowlist includes the ffmpeg framerate and fps filters. Slow motion by retiming (setpts) is not allowed.

Topaz Labs' September 2026 page lists Frame Interpolation, Slow Motion and SDR to HDR under its video workflows; it gives no further detail, so nothing about how those work is claimed here. Sume facts are from Timeline 1.0 and Video filter, read 2026-10-01.

What does Timeline do when fps differs from the source?

The docs say an output.fps that differs from a source's rate is met by repeating or dropping a frame every few frames, which gives judder on motion, and the render reports output_fps_resamples_sources. If you omit output.fps, the render keeps the rate the sources already run at. So 24 to 60 on Timeline is a container-rate change, not new in-between frames.

What does the video filter allow?

The filtergraph allowlist in the compiler has fps, framerate, tblend, tmix and tpad under a Time heading. You can validate a program for free with POST /v1/video-filter/check before paying the per-encode rate, and the check can return valid: false with diagnostics rather than a bare 400. Test on a short clip and inspect the result yourself; the docs make no quality claim for any of these filters.

Why is slow motion refused?

The filter docs list trim and setpts as not allowed and send range cuts to video trim. The compiler comment explains there is one way to cut a range and it is not a filtergraph string. Retiming for slow motion would need setpts, so plan it outside Sume.

How do I get 60 fps output for a social upload?

Set output.fps to 60 on the Timeline render and review motion for judder. For a related comparison on a rival tool, see Runway Enhance frame rate, 60 fps alternative.