Product turntable video from one photo with Gemini Omni

A turntable clip from one photo is possible, but the back of the product is invented. Send the photo as a first frame, ask for a slow 180-degree orbit on a plain surface, and the model will imagine the side and rear you never photographed. On Sume an 8-second 1080p Omni clip is $1.50; treat the result as a mood clip, not as proof of what the back looks like.

If the rear matters (labels, ports, ingredients), photograph it and use the two-frame method below.

Single-photo orbit

Use frame_images with the product photo as first_frame. Name the camera move with plain words, and say what must stay fixed.

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "The camera orbits slowly 180 degrees around the object on a plain light-grey surface, smooth constant speed, single continuous shot. The product keeps its shape, color and label. Soft studio light. Audio: faint room tone. No music. No dialogue.",
  "duration": 8,
  "resolution": "1080p",
  "aspect_ratio": "16:9",
  "frame_images": [
    { "type": "image_url", "image_url": { "url": "https://example.com/bottle-front.jpg" }, "frame_type": "first_frame" }
  ]
}

Two photos make the back real

Omni supports a first and a last frame on Sume (and in Google's own Omni guide, read 2026-10-07). Photograph the front and the back, send them as first_frame and last_frame, and ask for a rotation between. The model still invents the middle, but both ends are real, which covers what a buyer reads. This costs the same per second as the single-photo version.

Prices for an 8-second orbit

Sume bills list times 1.25 per second, rounded up to the cent per job. fal's list for Omni Flash 1.1 (recorded 2026-08-28) is $0.03, $0.10, $0.15 and $0.30 per second at 360p, 720p, 1080p and 4K.

Checks and a fallback

Pause at 0, 4 and 8 seconds and compare the shape with your photo; the video frames endpoint can pull stills from a Sume-hosted clip at times you name. Look for a warped logo, a changed cap, or a label that turns into gibberish. If the object is changing, shorten to 5 seconds and slow the orbit to 90 degrees. If the product must be exact, a rotating mockup built from real photos in a timeline will be more accurate than any generated orbit, and Sume does not claim a true 3D scan.

Reading the result

Judge three things: whether the label is readable at the start and end, whether the shape of the object stays stable while the camera moves, and whether the surface under it stays the same. Orbit shots are especially hard on objects with text. If text is the key selling point, shoot or render the label separately and composite it in a timeline. For a clean product page, a short loop of 4 to 5 seconds with a slow 90-degree move usually survives better than a full 8-second rotation.