Put a specific person in AI video after the Sora API shutdown

The short answer

OpenAI's deprecations page lists the Videos API and the sora-2 models with a shutdown date of September 24, 2026. If your Sora workflow put a specific person on screen, Sume has three routes: Kling motion control at $0.1575 a second, H3 Max Recast at $0.375 a second at 768p, and reference images on MiniMax H3 or Gemini Omni Flash.

What changed

OpenAI's deprecations page says developers were notified on March 24, 2026, and that the Videos API and sora-2, sora-2-pro and their dated snapshots shut down on September 24, 2026. The page names no replacement. This post only covers the one job where the replacement choice is not obvious: keeping a particular person consistent.

Pick by where the movement comes from

The three routes differ in where the movement comes from.

When each one fits

Choose motion control when you can film the move yourself. Choose Recast when the video already exists, such as a shot from a previous Sora run or a stock clip, and you want to swap the person in it. The source must run 5 to 30 seconds with no single shot longer than 15 seconds. Choose reference images when you want the model to invent the scene and only need the person to look right.

Reference-image request

Reference images ride in input_references on /v1/videos. The same call works for either model; only model changes.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: person-001" \
  -d '{
    "model": "minimax-h3",
    "prompt": "The person in the reference image walks through a market",
    "duration": 8,
    "resolution": "768p",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/person.png"}}
    ]
  }'

Consent

Use photos of people who agreed to appear in the video. For a migration of the plain submit-and-poll flow, see the Sora smoke test.