Gemini Omni video in the Windows app, and by API on Sume

If you want Gemini Omni video inside your own product rather than in a desktop app, call gemini-omni-flash-1.1 on Sume with POST /v1/videos. It takes 3 to 10 seconds at 360p, 720p, 1080p or 4K, in 16:9 or 9:16, and always returns audio.

Google's September 2026 roundup, read 2026-10-05, mentions generating Gemini Omni videos in the Gemini app for Windows. That is a consumer surface, and the page we read gives no API prices or limits, so we do not quote any from it.

App versus API

A desktop app is built for one person making one clip at a time. An API is for batches, queues and your own storage. On Sume the job is asynchronous: you submit, poll the status, then download from /content, and you can pass a callback_url over HTTPS so your server hears about completion.

The Sume row is also the default for sume/auto, so a request with no pinned model resolves to it (8 seconds, 720p, unless you set otherwise).

What the row takes

Limits and modes from the Video Router docs, prices as list times 1.25 rounded up per job. fal's list comes from its Omni Flash 1.1 page, read the same day.

Modes you cannot get from an app

• Image-to-video with a first frame and an optional last frame.
• Reference-to-video with up to 10 images and up to 3 clips of at most 3 seconds each, addressed in the prompt as <IMAGE_REF_0> and <VIDEO_REF_0>.
• Edit mode, on the Video Router /v1/video-router/generate surface: send a video_url and a prompt, and the source clip sets the output length (no aspect_ratio allowed).
• Idempotent retries with an Idempotency-Key header.

Limits to plan around

The row has no audio toggle: generate_audio: false returns 400, because audio is always on. It rejects size, seed and bitrate_mode, and the longest clip is 10 seconds. For anything longer, split the scene or pick a model with a higher cap such as wan-3.0 (30 seconds). The 4K post covers when the top tier is worth the price.