HeyGen lipsync batches of 100 vs Sume Fabric one clip per job

HeyGen's changelog lists POST /v3/lipsyncs/batches, which takes payloads identical to the single lipsync call, with up to 100 ids readable at once through a bulk status endpoint. Sume's closest route is VEED Fabric 1.0 at POST /v1/veed/fabric-1.0, and it creates one job per call, so a batch is a loop of submits.

What HeyGen lists

From the HeyGen API changelog, read 2026-10-02:

• This is lip-syncing onto existing video input at HeyGen; check its docs for the input fields.

What Sume's still plus audio route takes

The Models page lists VEED Fabric 1.0 as a talking still plus audio route (veed/fabric-1.0). The body needs audio_url, a measured duration_seconds, and exactly one visual source: image_url or an avatar_handle. A related MiniMax H3 Max lip-sync route takes the same still plus audio body, with audio of 5 to 14.8 seconds.

Note the input differs: Sume's routes animate a still to your audio, not a replacement mouth on an existing video. Sume's models page also says video models do not lip-sync to generated narration laid underneath, which is why talking shots go through this route.

Running 100 clips on Sume

There is no batches path to call. Submit each clip as its own job with a unique Idempotency-Key, then follow Jobs and results. Webhook mode saves you polling 100 jobs.

• Keep a table of idempotency key to job id.
• Retry a failed submit with the same key, never a new one.
• Measure each audio file's duration before you send it; the field is required.

Which to choose

If your source is existing video that needs new lip movement, HeyGen's batch lipsync is the matching feature. If your source is a still and a recording, Sume's Fabric route fits, one job at a time.