Hedra Omnia: one image plus audio, and Sume's lip-sync route

If you want a single image plus an audio file to become a speaking video, both Hedra and Sume take that pair of inputs. Hedra lists Omnia (Fast Alpha) as full-body and facial animation from one image and audio. Sume's POST /v1/minimax/h3-max/lip-sync takes image_url (or a saved avatar_id / avatar_handle) plus a Sume-hosted audio_url and returns a lip-synced video. The difference you must plan for is the scope: the Sume route is a lip-sync route, and the docs describe it as a still plus audio job, not a full-body motion model.

What each side lists

The Hedra row comes from the vendor's own announcements page, which is dated June 1, 2026, so check it again before you build on it. The Sume row comes from the repo API doc for the route.

A quick way to test both

Use the same portrait and the same 8-second line for both services. The point is to see what moves, not to rank them.

• Render the line once with Sume at 480p so you pay the lowest rate: ceil(8) seconds at $0.0625 is $0.50.
• Check that the still's width divided by height is between 0.4 and 2.5; Sume rejects other stills.
• Upload the audio to Sume first, because audio_url must be a Sume-hosted file.
• Send duration_seconds that matches the audio, as a number from 5 to 14.8.
• Poll GET /v1/jobs/:id/status, then read /result when it is done.

What Sume does not do

Sume's lip-sync route does not turn a still into a full-body performance. If you need a body to move, the docs point to Kling motion control, which takes a still plus a motion video rather than audio. I could not find a Sume route that takes one image and audio and animates the whole body, so treat that as a gap against the Hedra listing.

The route also does not clamp. A duration_seconds of 15 is refused, not trimmed, and speed_tier is accepted and ignored. The strict body rejects model, endpoint and provider_endpoint.

How to choose

Pick by the shot you need. A head-and-shoulders line read by a presenter fits a lip-sync job, and you pay a known per-second rate with a reserve at admission and a refund if the job fails. A shot where the character walks or gestures needs a motion source. Whatever you choose, pin the model for the whole run: the Sume packet guidance is to use one lip-sync model per run so faces and timing stay consistent across segments.