Product video from four reference photos with Omni Flash 1080p

A 6-second 1080p product video from reference photos costs $1.125 on Sume with Gemini Omni Flash 1.1, at $0.1875 per second. Omni accepts up to 10 reference images and 3 to 10 seconds, in 16:9 or 9:16.

Small sellers have photos but not footage. Reference images make the model use your product's shape and colour instead of inventing one. The costs are small enough to try a few angles, but references do not guarantee a faithful copy, so check every frame. Rates come from the Sume pricing code and the video catalog (read 2026-10-10).

What is the endpoint chain?

One job, with photos prepared first.

• Photograph the product on a plain surface from four sides. Use public HTTPS URLs, or Sume-hosted ones.
• Optional: POST /v1/rmbg-1.0/remove on each photo at $0.0225, to give the model a clean object. This costs $0.09 for four.
• POST /v1/videos with gemini-omni-flash-1.1, duration: 6, resolution: 1080p, aspect_ratio: 9:16 and the images as references; read the exact field name in the Video Generation page.
• Poll /v1/jobs/{id}/status or use callback_url, then take the video_url from the result.

What does it cost?

Resolution sets the price per second; references add no charge in the fixtures.

What are the limits?

Omni takes image and video references and no audio, so add sound in Timeline. It has a 3 to 10 second range; a longer video is a Wan 3.0 or Seedance 2.5 job, which accept up to 30 seconds. References guide the look but do not lock it: logos and small print may still shift. Never publish a video that shows a label you have not checked.

A cutout from Remove Background is a PNG with a transparent background of a real photo. Sume does not generate transparent images from text, so make cutouts from photographs.

Start at 360p ($0.225 for 6 seconds) and look for deformed shapes before you pay for 1080p. See still-to-clip with a first frame for the case where one image is enough.

Before you post

Compare the video with the product in your hand. Check colour, proportions and any text on the item, and regenerate if the model altered them. The 480p proof loop fits here, and the logo sting uses Omni's end frame if the shot should land on a lockup.

Pre-publish checklist

Check the video against the real product frame by frame for shape, colour and any text on the item. Reference images guide the model but do not lock it.

If the first draft is wrong, change the photos or the prompt before raising resolution; resolution only makes the same mistake sharper.

• Use four clean views: front, back, side and top.
• Keep the background plain and the lighting even.
• Run 360p first, then 1080p.