Replace one prop in a video with AI: bottle to apple edit prompt
Swap one object in a finished clip with gemini-omni-flash-1.1 on Sume: the docs example prompt, fields you cannot set, and an 8-second price.

To replace one object in an existing video, send the clip as video_url to gemini-omni-flash-1.1 and prompt the swap in one sentence plus a keep clause. The Sume Video Router docs use this exact request as the model's edit example: "Replace the bottle with an apple. Keep everything else the same." at 720p. The output follows the source clip, so a prop swap leaves the camera, the timing and the people as they were, and you pay for an edit job and not for a new generation.
This page walks through that example field by field, because each field in the edit shape either does something or is refused, and the refusals are what trip people up.
The request, field by field
The edit request is the smallest request on the video surface. Only four fields matter, and each of the other video fields is either ignored for edit or rejected.
| Field | Use in edit mode | Note |
|---|---|---|
| model | gemini-omni-flash-1.1 | Edit is an input shape of this one id |
| prompt | The swap plus a keep clause | Required |
| video_url | Public HTTPS URL of the source clip | This selects the edit mode |
| resolution | 360p, 720p, 1080p or 4K | Default 720p; 4K may be sent as 4K or 4k |
| duration | Reserve hint only | Never sent to the provider; default 8 s |
| aspect_ratio | Do not send | 400 if given |
| generate_audio | Do not send false | 400; audio is always native and synced |
{
"model": "gemini-omni-flash-1.1",
"prompt": "Replace the bottle with an apple. Keep everything else the same.",
"video_url": "https://example.com/source.mp4",
"resolution": "720p",
"duration": 8
}What the swap can and cannot depend on
The prompt is the only input that describes the new object. A reference photo is not allowed with video_url, so "an apple" has to be specific enough in words: say "a red apple" or "a green apple with a stem", not just "fruit". If the new object has to be an exact product, a new generation with reference images is the right route, and an edit is not.
The old object also needs to be unambiguous. If three bottles are on the table, name the one you mean by position or color: "the bottle on the left". A clear target matters more than a long prompt.
Why the duration field is only a hint
The provider does not return an output length for edits, so Sume cannot know the final length when the job is admitted. It reserves from the duration you send, and falls back to 8 seconds when you send nothing. The Video Router doc marks the exact meaning of this as something that still needs verification against the provider, so treat the reserve as an upper estimate and not as a promise of the settled price.
The practical rule is to send the true length of your source clip. A 6 second clip reserved as 8 seconds reserves about a third more than the clip needs. Sending a duration above your clip's real length does not make the output longer.
What it costs at an 8 second hint
Sume bills the provider list times 1.25 on each row, rounded up to the cent for a job. The table shows the 8 second default hint, which is what a request reserves when it sends no duration. At the default 720p resolution an 8 second edit reserves $1.00. For the arithmetic on other lengths, see Omni edit billing and the 8 second reserve.
| Resolution | Provider list per second | Sume per second | 8 second pass |
|---|---|---|---|
| 360p | $0.03 | $0.0375 | $0.30 |
| 720p | $0.10 | $0.125 | $1.00 |
| 1080p | $0.15 | $0.1875 | $1.50 |
| 4K | $0.30 | $0.375 | $3.00 |
When to re-roll and when to edit
A prop swap is also the easiest edit to judge. You know where the object was, so you can scrub to the frames where a hand touches it and see at once whether the new object holds its shape. Check those frames before the rest.
Use an edit when the scene is right and one object is wrong. Use a new take when the camera move or the action is wrong, because an edit cannot fix those, and a clip with the wrong action is a candidate for a new take at a lower resolution first. The deciding question is whether the thing you want to change is the same thing the source clip has already settled. For the three routes side by side, read fixing one detail in a finished clip.
Sources
Related posts
More in Models
- Square 1:1 ad video: which Sume models accept it, and 4:5
Seedance, Kling, Wan and MiniMax accept 1:1 on Sume; Gemini Omni takes only 16:9 and 9:16. No video model lists 4:5, so Feed needs a crop. Table and steps.
- MiniMax H3 Max on Sume: video, lip-sync and recast, which id to call
Three Sume ids carry the MiniMax H3 Max name: minimax-h3-max for video, a lip-sync route and h3-max-recast. What each takes, its window and its rate.
- Which Sume image models are text-to-image only, with no photo edits
Five Sume image models reject reference images: Higgsfield Soul, Imagen 4 Fast, Imagen 4 Ultra, Recraft V4 and Qwen Image Max. Prices, ratios and edit options.
- An OpenRouter-compatible video API: sume/auto or a pinned model
Sume's POST /v1/videos follows OpenRouter's video generation API field for field. Let sume/auto pick the model, or pin a catalog id like seedance-2.5.
Written by Sume