Animate an image

A video model can begin from a still image you supply. The image goes in whichever field that model uses for an opening frame, given as a public URL, a data URI, or an asset_... ID. In the example below, MiniMax H3 calls that field first_frame.

curl -X POST https://api.protoface.com/v1/run/minimax/minimax-h3 \
  -H "Authorization: Bearer $PROTOFACE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "The sailboat holds course as morning light fills the bay.",
    "duration_seconds": 8,
    "quality": "768p",
    "aspect_ratio": "16:9",
    "first_frame": "https://example.com/sailboat.png"
  }'

The clip lands on video.url when the run completes. A still from Generate an image already has an asset_id on that run’s outputs. Pass it straight into the next run instead of downloading the file. Models constrain which inputs combine. On MiniMax H3, last_frame requires first_frame, and sending it alone returns error.code: "run.input_role_dependency". Neither frame combines with reference, reference_video, or audio, and sending them together returns error.code: "run.input_role_conflict". Many models expose animation as a separate operation, video.animate, which takes the still on image and requires operation. Check the catalog entry for the model you call.

Inference

Realtime

Account

Reference