Solar roof before and after clip: Wan 3.0 first and last frame, 7 s
Send the bare roof as first_frame and the finished array as last_frame to wan-3.0. A 7-second clip costs $0.88 at 720p on Sume. Rules for matching the photos.

Send the bare-roof photo as a first_frame and the finished-array photo as a last_frame in one wan-3.0 job, and ask for a seven-second time-lapse between them. At 720p that costs $0.88 on Sume (7 x $0.10 x 1.25 = $0.875, rounded up). The model invents the middle, so the two endpoints are the only frames you control.
| Model and resolution | 7 s clip |
|---|---|
| wan-3.0 480p | $0.44 |
| wan-3.0 720p | $0.88 |
| wan-3.0 1080p | $1.75 |
| seedance-2.5 480p | $1.89 |
| seedance-2.5 720p | $4.05 |
Request shape
Sume's /v1/videos wire takes frame_images with frame_type first_frame and last_frame. Wan 3.0 is listed with end-frame support in the catalog, and Seedance 2.5 accepts start and end frame as well. Duration for wan-3.0 can be 2 to 30 s, and Seedance 2.5 starts at 4 s, so a 7 s clip works on both.
Shoot the two photos as a pair
- Use the same camera position, height and lens. A drone hovering a few metres off will read as a camera move, not a time-lapse.
- Match the light as closely as the jobs allow. If the before photo is overcast and the after photo is sunny, ask the prompt for the shift rather than hoping for it.
- Keep the roof edges, chimney and neighbouring houses in the same place in both frames.
- Use the real after photo. The array in the last frame should be the one actually installed, since the clip is shown to the customer.
What you cannot combine
On the /v1/videos wire, if you send both frame_images and input_references, frame_images controls the mode and the job runs as image-to-video. The Video Router's own endpoint is stricter: it rejects first/end frame fields together with reference_*_urls. So a logo reference plus an end frame is not available; add the logo in an editor.
The prompt should describe what happens between the frames, such as panels appearing in rows from the left, rather than describing the end state again. Keep the clip to one idea. For a quote walkthrough, make a second clip from a different photo pair and stitch them.
Prompt wording that helps the middle
Name what is allowed to change and what is not: the roof, chimney and trees stay fixed, and only the panels and the light change. Ask for a locked-off camera. If the clip is meant to read as a time-lapse, say so and give the order in which panels appear, because the model has no other information about the installation sequence.
A Seedance 2.5 take of the same 7 s costs more: $1.89 at 480p and $4.05 at 720p, against $0.44 and $0.88 for Wan 3.0. The model card lists end-frame input for Seedance 2.5 too, so the choice is about the look you prefer. Wan costs less per take, so more takes fit in a budget.
Cost of trying options
At $0.88 a take at 720p, five takes cost $4.40. At 480p, 7 x $0.0625 = $0.4375, rounded up to $0.44, so five drafts are $2.20.
Sources
Related posts
More in Use cases
- How do I add a spoken tagline and sound sting to the end of an ad?
Make a 3-second spoken end tag: one 1-cent TTS job, a $0.125 music sting and a $0.01 concat to attach it to any ad read. About 14.5 cents on Sume.
- Spreadsheet to AI video: Wan 3.0 xls input vs chart images on Sume
Alibaba lists xls as a Wan 3.0 input, Sume does not. Export each chart as a PNG and send up to 10 reference images or a first and last frame.
- Swap the presenter per market: one ad, two Recast jobs on Sume
Recast keeps the motion, camera, cuts and sound of one source ad and replaces the people. How to make one version per market, the limits and the price.
- 10 static ad variants in one Sume image request: n and its cap
The Sume Image API takes n from 1 to 10, but each model has its own cap. Read it from the catalog; use 4:5 on Nano Banana or 1088 x 1360 on GPT for feed ads.
Written by Sume