Which Sume video models take a source video_url? Edit, transfer

Three Video Router rows take a video_url source clip: gemini-omni-flash-1.1 for prompt-based edits, higgsfield-genjutsu for motion transfer, and h3-max-recast for swapping the people in a clip. No other row does, and sume/auto never routes to the last two.

Everything else, from Seedance 2.5 to Kling 3, generates a new clip from text, a frame or references. If you want to change an existing video, this is the whole list.

The three rows

The Video Router doc spells out the contract. video_url is the source of an edit, a motion-transfer or a recast job, and it cannot be combined with image_url, end_image_url or reference_*_urls. To condition a new generation on a clip, use reference_video_urls instead.

Edit with Gemini Omni Flash 1.1

The edit request sends a prompt and the source clip. The API sends no aspect ratio on this path and returns a 400 if you give one, and generate_audio: false is also rejected because the model always produces synced audio. If you send a duration with video_url it is only a reserve hint (the default hint is 8 seconds); the source clip sets the output.

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "Replace the bottle with an apple. Keep everything else the same.",
  "video_url": "https://example.com/clip.mp4",
  "resolution": "720p"
}

Why Genjutsu may be missing

higgsfield-genjutsu appears in GET /v1/video-router/models only when its provider is configured for the environment. If your catalog call does not list it, that is expected, not a bug in your client.

Where this surface lives

video_url is a Video Router field. The OpenRouter-shaped POST /v1/videos body does not carry the edit mode, so use POST /v1/video-router/generate for these three. The legacy surface creates the same jobs and the same billing, at provider list times 1.25.

• Edit: gemini-omni-flash-1.1, source length sets the output.
• Motion transfer: higgsfield-genjutsu, needs reference images, no text-only mode.
• Recast: h3-max-recast, one photo per new person, the output keeps the source motion, camera, cuts and sound.