Gemini Omni reference video length: 3 seconds per clip on Sume
On Sume, each Gemini Omni Flash reference video can be at most 3 seconds, with up to 3 clips. Trim longer footage first, then address clips as VIDEO_REF tags.
On Sume, every reference video sent to gemini-omni-flash-1.1 can be at most 3 seconds long, and a request can carry up to 3 of them (reference_video_urls, each 3 seconds or less). For longer footage, cut the moment you need into a short clip first, then pass that clip.
Google's video page tells Gemini API users to use Gemini Omni Flash as the default model for video generation (read 2026-09-30). The per-clip limit below comes from Sume's Video Router docs.
What are the reference limits for Gemini Omni on Sume?
The reference-to-video capability takes images, video clips, or both. Native audio is always on, and reference_audio_urls does not exist for this model.
| Input | Limit | Tag in the prompt |
|---|---|---|
reference_image_urls | Up to 10 | <IMAGE_REF_0> |
reference_video_urls | Up to 3, each 3 s or less | <VIDEO_REF_0> |
reference_audio_urls | Not accepted | None |
How do I point the prompt at a reference clip?
Use the tags in your prompt. Tags are 0-based and follow list order, so the first video in reference_video_urls is <VIDEO_REF_0> and the second is <VIDEO_REF_1>. Reorder the list if you want a different clip to be <VIDEO_REF_0>.
How do I shorten a longer clip to fit?
Video trim 1.0 returns a new MP4 holding only [start, end) and leaves the source untouched. Send video_url (a media.sume.com file in your workspace), start, and exactly one of end or duration to POST /v1/video-trim, with an Idempotency-Key. Use duration: 3 or less, then pass the returned video_url as the reference. Read the trim docs for the full body.
If the clip is the only thing that matters, also see Gemini Omni timecode prompts for steering cuts inside one generation.
What happens to a video I want to edit instead?
Editing is a separate mode. video_url is the edit source, not a reference, and the prompt describes the edit. The 3-second reference cap applies to reference_video_urls, not to that edit mode.
How do I confirm the limits before I send a job?
Read the catalog with GET /v1/video-router/models/{model_id}. The limits are listed there per model, and they differ between models, so check the id you plan to use.
Sources
Related posts
More in Models
- Gemini TTS API: which engine Sume's TTS routes to
Gemini 3.8 TTS is not in Sume's TTS Router, which lists Cartesia Sonic ids only. What Sume's TTS takes for voice, engine and length, with a checklist.
- Generative Extend any aspect ratio: Sume video ratios
Premiere's Generative Extend now takes any aspect ratio. Sume's gemini-omni-flash-1.1 makes 16:9 or 9:16 clips; for other shapes, conform with Timeline fit.
- Genjutsu Object Swap API: Sume serves Motion Transfer only
Higgsfield pairs Motion Transfer with Object Swap in Genjutsu. Sume's Genjutsu row serves Motion Transfer only; Object Swap is rejected by name.
- GPT Image 2.5 cached input price: does Sume pass it on?
OpenAI lists cached input at $1.25 text and $2 image per million tokens for Sunburst. Sume's docs give estimates and reserved cost, not a cache discount.
Written by Sume