Change the background behind a person in a video with AI (Omni edit)
Swap the backdrop behind a person in an existing clip with Gemini Omni Flash 1.1 edit on Sume: a keep-the-subject prompt, limits, and a 10-second price table.

To change the background behind a person in an existing video on Sume, send the clip as video_url to gemini-omni-flash-1.1 and write a prompt that describes the new backdrop and keeps the person fixed: "Change the background to a sunlit loft office. Keep the person, her clothing and her movement the same." This is the route to try first when the footage is good and only the place is wrong. If you need an exact location from a photo, this route cannot take the photo, and a different route does.
Sume's docs set the boundaries: edit mode takes a prompt and a source clip, and it does not take references or an aspect ratio. Everything below follows from those limits.
Prompt: describe the place in words
Because an edit cannot receive a reference photo of the new location, the prompt carries all of the description. Give the model enough to build a place and not a label.
- Name the place type and the light: a bright loft office with large windows and soft morning light.
- Keep the lighting direction consistent with the person, or say what the light should do: warm light from the left.
- Add one prop or detail to anchor the scene, such as a plant on a shelf, and stop there.
- End with the keep clause: Keep the person, her clothing and her movement the same.
Keep the subject steady: what to write and what to skip
The keep clause is the instruction that matters most in a background edit, since the person is the one thing you do not want regenerated. Name the parts that must not change: the person, the clothing, the movement and the camera. Skip adjectives about the person's appearance that are not in the source, because they invite a change.
Pick source footage that helps. A clip where the subject is clearly separated from the background, in steady light, gives the edit less to guess. A clip where the subject's hair blends into a busy backdrop, or where several people overlap, is harder to hold steady. This is practice advice from how the edit route is documented, and it is not a Sume guarantee.
Three routes for a new background
An edit is not the only way to change where a person stands. The right route depends on whether you need a specific place, a specific person, or just a plausible one.
| Route | Input | Choose it when | Limit |
|---|---|---|---|
Omni edit (video_url) | Source clip + prompt | The footage is right, the place is wrong | No reference photo, no aspect_ratio |
| Reference-to-video on Omni | Images and short clips as references | You have a photo of the place and can reshoot the action | A new generation, not an edit; up to 10 images and 3 clips of 3 s |
| Kling 3.0 motion control | A still of the person in the new place + a motion clip | You want a stand-in performing the original moves | Output length follows the motion video, 1 to 30 s |
What to check in the result
The edges of the subject are the weak point of any background change. Look at hair, hands and moving fabric against the new backdrop, and look for a halo or a flicker as the person moves. Check that the shadow on the floor still fits the new light. If any of those fail, re-run the same prompt once before you rewrite it, since each run differs. The model also generates native synced audio, so listen to a room tone that does not fit the new place, because you cannot switch the audio off in this mode.
If the clip has a plain green backdrop, the issue is simpler, and the stored guide on replacing a green screen background compares three routes.
A 10 second pass
Billing is the provider list times 1.25 per row, rounded up to the cent. Send a duration equal to your source length so the reserve matches the clip. The table is for a 10 second source, the longest clip that the Gemini Omni Flash 1.1 generation modes accept.
| Resolution | Provider list per second | Sume per second | 10 second pass |
|---|---|---|---|
| 360p | $0.03 | $0.0375 | $0.38 |
| 720p | $0.10 | $0.125 | $1.25 |
| 1080p | $0.15 | $0.1875 | $1.88 |
| 4K | $0.30 | $0.375 | $3.75 |
{
"model": "gemini-omni-flash-1.1",
"prompt": "Change the background to a sunlit loft office. Keep the person, her clothing and her movement the same.",
"video_url": "https://example.com/source.mp4",
"resolution": "720p",
"duration": 10
}Sources
Related posts
More in Use cases
- Change the print on a t-shirt in a model photo with a design reference
Swap the graphic on a shirt in a model photo: photo first, new design as a second reference to ideogram/ideogram-v4.5. $0.075 per photo at medium on Sume.
- Customer onboarding email series with an avatar: 5 clips, cost by tier
Five 20-second onboarding clips from one Sume avatar cost $18.40 on standard, $24.50 on plus and $55.00 on max. How to plan the series and when to re-render.
- How do I automate a daily morning briefing audio with a TTS API?
A daily briefing of 2,200 characters is one TTS job: 11 cents a day, $3.30 for 30 days on Sume, plus a 1-cent concat if you add a music sting.
- How do I make AI announcer intros for conference speakers?
Ten 25-second speaker intros with one shared sting cost about 43 cents on Sume: 2 cents of TTS and a 1-cent join each, plus a $0.125 track made once.
Written by Sume