Gemini Omni Flash prompt guide: one shot, audio, text

Google's Gemini Omni Flash prompt guide gives four habits: ask for a single continuous shot when you want one scene, put what you do not want inside the prompt as a plain sentence, describe the audio you want, and write out any on-screen text. Omni Flash tries to make several shots by default, so a one-shot request has to say so.

The tips below are quoted from Google's Gemini API page on Omni Flash, last updated 2026-09-23 and read 2026-09-29. They are Google's advice for its model. Sume lists the same-named gemini-omni-flash-1.1 (Video Router); its prompt goes in the prompt field, and where Sume differs the section says so.

How do I get one continuous shot?

Google says that by default Omni Flash will try to create a video with a few different shots, and that if you need a single scene you must prompt for it. Its example phrases are below.

How do I say what I do not want?

Write it in the prompt. Google's page says negative prompts are not supported as a separate setting, and that you can put your negatives in the regular prompt, for example "Do not do X". Its guide adds that simple negative prompts such as "No dialogue" help when a video contains things you do not want.

How do I control the sound?

Google says the model tries to generate an appropriate audio track by default, and that you describe the audio you want in the prompt, especially music. On Sume this is the only lever: gemini-omni-flash-1.1 always generates native synced audio, and generate_audio: false is rejected. If you want no dialogue, say "No dialogue" in the prompt, as Google suggests.

How do I get readable text in the video?

Google says you can prompt for text and that Omni will render it correctly and readably, and that if text will appear naturally, even in background elements, it helps to define what it should say. Its examples spell out the exact strings, such as a street sign that says a given phrase. Quote the exact words in your prompt and keep them short.

Does the prompt language matter?

Google's page says English is fully supported and other languages have not been evaluated, so they may work but results can vary. Write in English when the result matters.

How do I send one of these prompts on Sume?

Put the prompt in prompt on POST /v1/videos with the model id. Duration is 3–10 seconds, and the aspect ratio is 16:9 or 9:16.

curl -X POST "https://api.sume.com/v1/videos" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: omni-one-shot-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "In a single continuous shot, a tabby cat sits on a sunny windowsill. Gentle breeze, distant bird chirps. No dialogue.",
    "resolution": "720p",
    "aspect_ratio": "9:16",
    "duration": 6
  }'