Cookware sizzle video with native audio: Omni Flash 1.1 cost

To make a cookware product video with real sizzle, submit gemini-omni-flash-1.1 to POST /v1/videos with a 9:16 aspect ratio and a duration from 3 to 10 seconds. The model generates a synced audio track with the picture, so the pan, the oil and the sear arrive in one file. At the 720p list rate of $0.10 per second, a 10-second clip bills $1.25.

Why this model fits a sound-driven product shot

Cookware sells on sound as much as on sight. A cast-iron skillet is a sizzle, a lid is a clink, and a nonstick pan is the silence of an egg sliding out. Most video models give you a silent clip and ask you to add sound later. Gemini Omni Flash 1.1 returns audio generated with the frames, which removes one step from a product ad.

The model accepts 3 to 10 seconds at 360p, 720p, 1080p and 4K, in 16:9 or 9:16. It takes image and video input_references but no audio references, so your product photo can anchor the look while the sound comes from the prompt.

The request

Send the pan photo as an image reference and describe the action and the sound in the prompt. Keep the job idempotent with an Idempotency-Key header so a retry does not bill twice.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: skillet-sizzle-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Close-up of the black skillet from the reference photo, steak hits the pan, loud sizzle, steam rises, 9:16",
    "duration": 10,
    "resolution": "720p",
    "aspect_ratio": "9:16",
    "generate_audio": true,
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/skillet.jpg" } }
    ]
  }'

What it costs

Sume prices video at the provider list rate times 1.25. The arithmetic below uses the 720p list rate from the catalog. Read the live figure from GET /v1/videos/models before you budget, because rates change.

The same ten seconds at a higher resolution has a different rate, so pin resolution in the request instead of relying on a default.

Prompting for sound and motion

Because the model writes audio and picture together, name the sounds you want. Say loud sizzle, a lid set down, or quiet room tone. Describe one camera move per clip, such as a slow push-in on the pan, and one subject action. Short, concrete prompts for a single shot work better than a paragraph that tries to cover a whole ad.

If the first take is wrong, change one thing and resubmit under a new Idempotency-Key. Reusing the old key returns the old job by design, so a deliberate retry needs a deliberate new key.

Limits that fail loudly

Ask for 12 seconds and the API will not quietly trim it. The model supports 3 to 10 seconds, so a longer ad is two clips joined on a Timeline, not one oversized request. For the join, see the Timeline 1.0 docs, which bill $0.10 per ceil output minute.

Poll GET /v1/videos/{job_id} until it completes, or pass a callback_url and let the signed webhook tell you. Then fetch the file from /v1/videos/{job_id}/content.