Raw-to-plated recipe clip: first and last frame on Sume
Pin the raw-ingredients photo as the first frame and the plated dish as the last. Wan 3.0, Omni Flash 1.1 and H3 take both; the 8-second prices are inside.

For a raw-to-plated recipe clip, send the raw-ingredients photo as the first frame and the finished-dish photo as the last frame in one POST /v1/videos request. On Sume, wan-3.0, gemini-omni-flash-1.1, minimax-h3, kling-3 and the Seedance ids all list end-frame support in the catalog, so the clip is generated between your two photos instead of toward whatever the prompt invents. An 8-second 720p clip costs $1.00 on Wan 3.0 or Omni Flash 1.1, and $0.60 on MiniMax H3 at its native 768p.
The request shape is the OpenRouter-style one documented on Video generation: frame_images with frame_type set to first_frame and last_frame. The limits per model come from the Video Router page, read on 2026-10-06.
What does the request look like?
Two frame_images entries and a short prompt that describes the motion between them. The API infers image-to-video from frame_images, so you never declare a mode. Keep both photos at the same framing (same overhead angle, same board), because the model has to travel from one to the other in the time you give it.
curl -sS -X POST https://api.sume.com/v1/videos \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "wan-3.0",
"prompt": "Raw ingredients on a board come together into the plated dish, overhead, steam rising",
"duration": 8,
"resolution": "720p",
"frame_images": [
{"type": "image_url", "image_url": {"url": "https://example.com/raw.jpg"}, "frame_type": "first_frame"},
{"type": "image_url", "image_url": {"url": "https://example.com/plated.jpg"}, "frame_type": "last_frame"}
]
}'What does an 8-second recipe clip cost?
The table prices the same 8-second request on the models that take a last frame. Omni Flash 1.1 makes 16:9 or 9:16 clips only and always produces synced audio; H3 also has native audio with no toggle.
| Model id | Resolution | Length | Sume price | Note |
|---|---|---|---|---|
| wan-3.0 | 720p | 8 s | $1.00 | |
| gemini-omni-flash-1.1 | 720p | 8 s | $1.00 | 16:9 or 9:16 only |
| minimax-h3 | 768p | 8 s | $0.60 | 768p is native; no 720p |
| kling-3 | 720p | 8 s | $1.68 | with sound; silent is $1.12 |
| seedance-2-fast | 720p | 8 s | $2.42 | |
| seedance-2.5 | 720p | 8 s | $4.63 |
Which one should I try first?
Sume does not rank output quality, so decide by constraint. If the two photos are 4:3, skip Omni Flash and Kling, which do not list that ratio. For a throwaway test, run it at 480p on wan-3.0 for $0.50 before paying for the 720p pass. Nothing on Sume accepts a seed, so a re-run will not repeat the clip; budget one or two takes.
Sources
Related posts
More in Use cases
- Real-estate agent, 25 listings a month: staged photos and voiceover
25 listings with six AI-edited photos, a voiceover and a music bed cost about $14.60 a month on Sume. Unit prices and the sum.
- AI listing video disclosure line: put it in the Sume output schema
Carry the AI disclosure and the original-photos link with each listing video: a typed output_schema field, what input does not carry, and a check.
- AI video generator for real estate: which Sume Format?
Sume's 27 catalog Formats include none for property listings. What to call instead for a listing video, and the disclosure test agents face.
- Real estate listing video with an AI avatar: fair housing script check
Describe the home, not the buyer. A fair housing word check in Python, a listing-photo scene on Sume Avatar 1.0, and the cost of a 45-second listing clip.
Written by Sume