12 Sume video ids: which need a source video, which take text

Sume's video router lists twelve ids. Nine start from a text prompt. grok-imagine-video-1.5 needs a first-frame image, and higgsfield-genjutsu and h3-max-recast need a source video plus photos, so none of those three can start from a prompt alone. gemini-omni-flash-1.1 additionally has an edit mode that takes a video_url.

The twelve

Durations come from the docs and the model catalog; confirm them live with GET /v1/videos/models.

Why it matters

If you wrap the API in a picker, filter on the request you can actually build. A form with only a prompt box should hide grok-imagine-video-1.5, higgsfield-genjutsu and h3-max-recast; the catalog constraints say each requires an image or a source video, so a prompt alone will not produce a clip.

• Prompt-only forms: show the nine text-capable ids.
• Tools that upload footage: add h3-max-recast and higgsfield-genjutsu; tools that upload a still can add grok-imagine-video-1.5.
• Always read supported_input_references from the catalog instead of hard-coding this table.

Reading the catalog in code

The live source for this table is the model list. A short script can print every id with its durations and input types, which makes a good CI check when the catalog changes:

import os, requests
r = requests.get("https://api.sume.com/v1/videos/models",
                 headers={"Authorization": "Bearer " + os.environ["SUME_API_KEY"]})
for m in r.json()["data"]:
    d = m["supported_durations"]
    print(m["id"], d[0], d[-1], m["supported_input_references"])