Architect's year-end project review video: six slides, $0.35 on Sume
A 60-second client review of one project costs $0.35 on Sume: four concept images, one voiceover, one render. Twelve projects come to $4.20.

An architecture studio can produce a 60-second year-end review of one project for $0.35 on Sume: four concept images at $0.05, a 900-character voiceover for $0.05, and a one-minute render for $0.10. A practice with twelve active projects spends $4.20 (read 2026-10-03).
December is when many clients ask where their project stands and when budgets for next year get set, so a short narrated recap sent with the status letter is useful. The video is a summary, not a substitute for drawings, and every image generated from a sketch should be labeled as an illustration.
Where each dollar goes
Concept images: use the Image Router's seedream-4-5-edit with your sketch or massing photo as the reference. The list rate is $0.04, and Sume bills $0.05 after its margin. Ask for aspect_ratio 16:9 so the slides match the screen, and describe materials and light rather than details you want preserved exactly. Treat the output as mood, since an image model will invent windows and structure.
Voiceover: 900 characters of narration is $0.04275 at $0.0475 per 1,000 characters, which bills as $0.05. That is about a minute spoken at a measured pace, but check the real length after generation and give Timeline the actual duration. Render: six slides of 10 seconds each, with fade transitions, is one render minute at $0.10. If the narration runs to 61 seconds the render bills as two minutes, so trim the script rather than the slides.
| Item | Rate | Quantity | Cost |
|---|---|---|---|
| Concept images, Seedream 4.5 edit | $0.05 per image | 4 | $0.20 |
| Voiceover, 900 characters | $0.0475 per 1,000 characters, billed to the cent | 1 | $0.05 |
| Timeline render, up to 60 s | $0.10 per output minute | 1 | $0.10 |
| Total per project | $0.35 |
Keep the facts out of the pixels
Dates, budgets, square footage and permit status belong in your slide tool, not in a prompt. Export those slides as images, import them, and use generated images only for the mood slides. A figure rendered by an image model can be wrong in a way nobody notices until a client quotes it.
For the final slide, a clear line stating what is approved and what is still open protects you more than any visual does. If scope changes in January, you regenerate one slide and the voiceover, not the whole production: $0.05 plus $0.05 plus $0.10 is $0.20 (read 2026-10-03). For a stronger sketch-to-render workflow, the existing post on reference edits from a sketch covers prompts and ratios.
Budget and checks before you start
Budget the whole practice before the first run. Twelve projects at the per-project total is the number above, and a second pass after client comments roughly doubles the voice and render lines but not the images, if you keep the same four concepts. Write one script template with the same five headings for every project, so the narrator voice, the slide order and the length stay consistent across the portfolio.
Finally, decide where the video lives. A link in the year-end letter works better than an attachment, and the file you send should be the render output rather than a screen recording, so the quality is the same for every client.
Sources
Related posts
More in Use cases
- Does Ask YouTube search include Shorts? What to make findable
YouTube's interactive search compiles long-form videos and Shorts and was Premium-only in the US at launch. What a Short's spoken text can and cannot do.
- Assembly or process video from a 3D render with Seedance 2.5
ByteDance shows Seedance 2.5 turning a clay render into an assembly sequence. How to plan ordered steps, references and 4-30 s clips for a process video.
- Audience Network in-stream: 5-30 seconds, sound required, keep audio
Meta lists Audience Network in-stream at 5-30 seconds, the one placement where sound is required. Keep audio when you trim; do not reuse a silent cut.
- One voice in 23 languages: a test matrix to run before you commit
MAI-Voice-2.1 claims one voice with a native accent in 23 languages. Build a 23-line audition matrix and cost it before you promise it to a market.
Written by Sume