Genjutsu missing from /v1/video-router/models: when it is listed

If higgsfield-genjutsu is missing from GET /v1/video-router/models, it is not a typo on your side: Sume lists Genjutsu Motion Transfer only when its provider is configured, and hides it otherwise. The docs say so in the Video Router page, and the catalog code filters it out the same way. Treat the catalog response as the source of truth for what you can call right now.

This is documented in Video Router and Video generation; the filter itself is in the repo's Video Router catalog.

How do I check whether Genjutsu is available?

List the catalog and look for the id. The response includes capabilities for each model, which the docs tell you to read rather than assume. If the id is present, it also shows the 480p and 720p resolutions and the 4 to 30 second duration range.

curl -s https://api.sume.com/v1/video-router/models \
  -H "Authorization: Bearer $SUME_API_KEY" \
  | jq -r '.. | objects | select(.id? == "higgsfield-genjutsu") | .id'

What does an empty result mean?

No output means the id is not in the response, so a generate call naming it should not be relied on. The Video Router also rejects a model id that is not in its catalog. Nothing in the docs tells you why a provider is not configured, and this post will not guess; the only thing you can act on is that the catalog does not list it.

• Check the same call against the API key you actually use, not a different workspace.
• Do not hard-code Genjutsu into a production flow without a catalog check at startup.
• If you need motion transfer to be deterministic, ask Sume support whether it is enabled for your account.

What can I use instead?

It depends on what you wanted from Genjutsu. For a restyle driven by a text prompt, the Gemini Omni Flash 1.1 edit mode takes one video_url and a prompt. For swapping the people in a clip, H3 Max Recast takes a source of 5 to 30 seconds and 1 to 4 photos. Neither is a Genjutsu equivalent: Genjutsu uses 1 to 8 reference images and preserves the source length and framing, and the docs describe it as motion transfer.

Both alternatives are in the same catalog response, so the one listing call also tells you if they are available.

Does the same apply to /v1/videos?

The Video generation page describes the same catalog behind POST /v1/videos and says Genjutsu is listed only when its provider is configured. Its model discovery endpoint, GET /v1/videos/models, is the place to look from that surface, and the page says to check supported_input_references and the other capability fields before you submit. On /v1/videos Genjutsu takes exactly one video_url and 1 to 8 image_url entries as input_references.

If you pin a model id in a config file, add a startup check that the id is listed and fail loudly if it is not. A silent fallback to a different model changes the output, and for a video job it also changes the bill.