Lip sync API: turn an image and audio into a talking clip

To lip sync a still image to audio with the Sume API, send POST /v1/veed/fabric-1.0 with the picture in image_url, a Sume-hosted audio_url, and the audio's measured length in duration_seconds. For 5–14.8 seconds of audio, POST /v1/minimax/h3-max/lip-sync takes the same body. Both run as jobs and return a talking clip.

Facts come from the Models overview and the request schemas in the Sume API reference, read on 2026-09-26. Prices are read from Sume's pricing code, the Fabric rate from the code behind API pricing. If you have a script rather than audio, the talking avatar video API takes a ready avatar and the script directly.

What inputs does a lip sync request need?

Both endpoints take the same inputs:

• One visual source: a public HTTPS still in image_url, or a ready avatar in avatar_id or avatar_handle, resolved server-side to the avatar's identity still. They are mutually exclusive.
• The models guide prefers image_url of a generated, inspected posed still, and avatar_handle only when the user named that avatar.
• Audio on the Sume media host, typically a TTS segment. Other hosts are rejected, and the file can be at most 10 MB. The text to speech API can produce one clip per sentence.
• The audio's measured length in duration_seconds.

What does a lip sync request look like?

Send an Idempotency-Key so a retry returns the original job instead of creating a second one. This request lip syncs a presenter photo to one voice line at 720p:

curl -X POST https://api.sume.com/v1/veed/fabric-1.0 \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: fabric-line-001" \
  -d '{
    "image_url": "https://example.com/presenter.png",
    "audio_url": "https://media.sume.com/artifacts/example/line-1.wav",
    "duration_seconds": 7.6,
    "resolution": "720p"
  }'

How do VEED Fabric 1.0 and H3 Max Lip Sync differ?

MiniMax H3 Max Lip Sync is the explicit alternative to Fabric inside its 5–14.8 second audio window. For audio outside that window, use Fabric, which takes 1–300 seconds.

Can a video model lip sync to my voice-over instead?

No. Sume's models guide says video models do not lip-sync to generated TTS or to a later voice-over, so a talking face is never a video-model clip with narration laid underneath. The guide builds a speaking shot from an accepted still plus TTS on Fabric, and leaves wordless beats, B-roll, and product motion to image and video generation.

To animate a still with the motion of a driving video instead of a voice, see the motion control API guide.

What does lip sync cost?

VEED Fabric 1.0 costs $0.1875 per audio second (720p), plus a 5.5% agent fee by default. It is priced per audio second, rounded up, so the 7.6-second line above is billed as 8 seconds: $1.50 at 720p before the fee. duration_seconds is what reserves that amount at submit.

MiniMax H3 Max Lip Sync costs $0.10 per audio second (768p), plus the same agent fee. That rate is the provider's list price × 1.25, billed per audio second and rounded up, and GET /v1/catalog carries the 480p and 1080p rates. Failed jobs, and jobs canceled before generation starts, are refunded before capture.

What are the limits?

The schemas and the models guide set these rules:

• Exactly one visual source per request, and an audio_url on the Sume media host of at most 10 MB.
• Fabric takes 1–300 seconds of audio. H3 Max Lip Sync takes 5–14.8 seconds, and its output length follows the audio.
• Provider queue ids are selected by Sume and are not accepted in either body.
• POST /v1/avatar-1.0/image-to-video and POST /v1/models/sume/avatar-1.0/image-to-video/runs are deprecated, retiring aliases that still accept the same body. Migrate by changing only the URL to POST /v1/veed/fabric-1.0 or POST /v1/models/veed/fabric-1.0/runs.
• On hosted MCP, the Fabric body goes inside payload on the avatar-image-to-video_create tool.