Avatar clip frame rate: Griffin 25 fps vs Fabric vs H3 Max

Tavus's Griffin page says one latent is 8 frames at 25 fps, which is 320 ms, and Griffin-Lite generates 720p video in 320 ms chunks in real time (read 2026-10-05). Sume's H3 Max lip-sync notes say Fabric is 25 fps and that you must measure H3 Max on the first real clip. POST /v1/video-inspect returns a probe for exactly that.

Why the frame rate matters

If you assemble several talking clips into one timeline, the output frame rate is locked once. Mixing sources at different rates forces a conversion. That is why Sume's packet guidance says to use one lip-sync model for each run: the model sets the assemble output.fps lock.

Griffin's number is a property of a streaming system. You do not receive Griffin files from Sume, so the comparison is about expectation, not interchange.

Measure your own clip

Import the finished clip into your workspace first, then inspect it. The default mode is sync and waits up to 30 seconds. If it takes longer you get a 202 and poll the job. frames: false skips stills when you only want the probe.

curl -X POST https://api.sume.com/v1/video-inspect \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: probe-h3-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/talk.mp4",
    "frames": false
  }'

What to do with the result

Read the frame rate from probe, write it next to the model name in your run notes, and keep every clip in that run on the same model. If a later clip goes to the other model because it falls outside the 5 to 14.8 second window, convert deliberately instead of finding the mismatch in the final export.