New ad in the style of last quarter's spot: reference video
Use a past brand clip as a reference input on gemini-omni-flash-1.1 so a new product spot keeps the look. Reference-to-video accepts image and video, not audio.

To make a new product spot that looks like your last one, pass the old clip as a video_url entry in input_references on POST /v1/videos with gemini-omni-flash-1.1, and add the new product photo as an image_url reference. The model accepts image and video references but no audio references. It conditions the new generation on them.
Reference or edit
These are two different jobs. An edit changes the pixels of one clip and keeps its length. A reference-to-video generation makes a new clip that borrows look and motion from the reference. For a new product with an old brand look, you want the second.
Sume infers the mode from the fields. input_references present means reference-to-video. frame_images wins if you send both. A single reference image with no frame is reference-to-video, the same rule as for Seedance.
The request
Send the previous spot and the new product photo. Say in the prompt what carries over, such as colour grade, camera speed and set, and what changes.
curl -X POST https://api.sume.com/v1/videos \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: q4-spot-style-ref-001" \
-d '{
"model": "gemini-omni-flash-1.1",
"prompt": "Same warm grade and slow push-in as the reference clip, now featuring the product in the photo",
"duration": 8,
"aspect_ratio": "9:16",
"input_references": [
{"type": "video_url", "video_url": {"url": "https://example.com/last-quarter.mp4"}},
{"type": "image_url", "image_url": {"url": "https://example.com/new-product.jpg"}}
]
}'Limits
Audio references are not accepted on this model, so the soundtrack comes from the prompt or from your own bed on Timeline. Output is 3 to 10 seconds, 16:9 or 9:16, at 360p, 720p, 1080p or 4K. Billing follows the catalog list rate times 1.25 for the resolution you pick.
| Reference type | Accepted on gemini-omni-flash-1.1 |
|---|---|
| image_url | Yes |
| video_url | Yes |
| audio_url | No |
Own what you reference
Only reference footage you have the rights to use. A brand's own past spots are the safe case. Then compare the new clip with the old one side by side before approval, since a reference guides the look and does not guarantee it.
Sources
Related posts
More in Use cases
- New Year countdown video music: build to midnight with a timed prompt
Write a timed Lyria 3.5 prompt that builds for 50 seconds and hits at midnight. $0.125 a take on Sume, then a Timeline render at $0.10 under your countdown.
- Nine product angles as references: MiniMax H3 Max on Sume
How to send up to nine product photos as reference images to minimax-h3-max on Sume, what the vendor limit is, and where Sume's own limits live.
- One 90-second ad, three voice beats: three TTS jobs and a concat
Splitting a 1,350-character read into hook, proof and call-to-action jobs costs 8 cents plus 1 cent to join, against 7 for one job. When that is worth it.
- One Black Friday master, three markets: EU, UK, Australia cards
One clean master, a different price card per market: EU prior price, UK total with delivery, Australian end date. Three $0.20 jobs. Read 2026-10-05.
Written by Sume