Edit a clip, swap a person, or move motion onto photos: 3 Sume models

Match the tool to what you already have. To change things inside a clip with a text instruction, use Gemini Omni Flash 1.1 in edit mode. To replace the people in a clip with people from 1 to 4 photos, use H3 Max Recast. To drive 1 to 8 reference images with a source video's motion, use Higgsfield Genjutsu Motion Transfer. All three need a source video_url, and none of them does text-only generation.

Three inputs, three tools

The constraints below are from the Sume catalog and the Video Router page, read 2026-10-08.

What each costs for a 10-second result

Sume bills the catalog list times 1.25, rounded up to the cent. The Recast and Genjutsu durations are the inspected source length. For Omni edit I use a 10 s result, because Sume bills Omni per output second.

Choosing between them

• If the change is a prompt ("replace the bottle with an apple, keep everything else"), Omni edit is the cheapest of the three at 720p.
• If the change is who appears, Recast keeps the source length, motion, camera, cuts and soundtrack and replaces the people. One photo per new person, left to right by default.
• If you have stills and want them moved by a reference video, Genjutsu preserves the source length and framing. Check that the model is in your catalog, because it is listed only when its provider is configured.
• Recast needs source shots of 15 s or less. A single 16-second take must be split first.

How to call it

Send Omni edit through the Video Router with model: "gemini-omni-flash-1.1", prompt and video_url, plus an optional resolution. On /v1/videos, Recast and Genjutsu take one video_url entry plus image_url entries in input_references. Because Recast and Genjutsu durations must equal the inspected source length, run video inspect first so you do not pay for a wrong guess.

Check it before you run it

Every figure above is a catalog list price times 1.25, rounded up to the cent, as of 2026-10-08. Catalogs change, so before a large batch, read the current model entry in the docs and recompute the one line that matters for your case. Write the arithmetic next to the job in your own notes: list rate, seconds or characters, multiplier, rounding. If the result differs from the wallet charge by more than a cent, the catalog entry has changed, and the docs page is the place to find out why.

Run one small job first. Submit a single request with the pinned model id and the settings in the tables, poll the returned polling_url until it finishes, and compare the charge with your estimate. Then scale up. Using a pinned id for the test matters, because sume/auto never names the family, so you cannot tie its charge to the row you priced.