Upscale a 5-minute clip on Sume: 30-second parts, then join
Sume Video Upscale takes up to 30 seconds per job. A 5-minute clip is ten parts: trim, upscale each at $0.009 per second, join. The math comes to about $3.41.
Sume Video Upscale takes at most 30 seconds per job, so a 5-minute clip is ten parts: trim each one, upscale each, then join them in Timeline with the original audio. At the list rate of $0.009 per video second the upscale alone is $2.70, and the whole pipeline comes to about $3.41 by arithmetic. That figure is not a measured run, and the join step has one condition to check, covered below.
What are the steps and prices?
Sources for trim and Timeline must be media.sume.com files. Upscale takes a public HTTPS video_url.
| Step | Count | Rate | Subtotal |
|---|---|---|---|
| Video trim, 30 s parts | 10 | $0.02 per job | $0.20 |
| Video upscale, 300 s | 300 s | $0.009 per second | $2.70 |
| Audio detach, original track | 1 | $0.01 per job | $0.01 |
| Timeline render, 5 min | 5 | $0.10 per output minute | $0.50 |
| Total | $3.41 |
How do I cut the parts?
Use Video trim with start and duration 30, precision exact. Parts that share an exact boundary join with matching output settings; check the join point yourself, since a cut can still show a small change in tone or motion. Keep every part at the same fps and size by passing the same output block.
What do I check before the join?
Timeline sources must be media.sume.com. Read each upscale result and confirm the URL is on that host before you build the render. If it is not, the file has to be added to your media library first. Run the unbilled Timeline plan call to catch a bad source or an fps mismatch before you pay for the render.
How does the audio stay in sync?
The upscale does not change the duration, so the original soundtrack still lines up. Detach it once from the source with Audio detach, then set audio.duration_seconds on the Timeline render to the full length, up to 1800 seconds.
When is this not worth it?
If only a few seconds matter, trim those seconds first and upscale only them. A single 30 second hero part costs $0.27 to upscale. See the 30-second cap post for that approach.
How do I keep the run safe to retry?
Give every call its own Idempotency-Key, built from the part number, for example upscale-part-03. If a network error hides the response, resend the same call with the same key and Sume returns the original job and does not create a second paid one. Poll each job to a terminal state before you build the Timeline render, and keep a table of part, job id and status.
Sources
Related posts
More in Use cases
- Verify a spoken ad disclosure is in the final mix with STT word times
Detach the final ad's audio, transcribe it with Sume STT, and confirm the required line is spoken and when it starts. Python check, not legal advice.
- Vinted bans edited and stock photos: what AI can still do
Vinted's rules say photos must show the item as it is with no image editing, and ban stock and watermarked pictures. Keep AI out of listings; use it for promos.
- Vocabulary flashcard video: one still and one spoken word per card
Build a vocabulary flashcard video: one image and one TTS word per card, joined with Timeline audio concat, then re-based into render slots from segments[].
- Voice replication blocked in IL, TX, EEA, UK, CH, IN: plan voiceovers
Digital Applied reports Gemini voice replication is unavailable in Illinois, Texas, the EEA, UK, Switzerland and India. A market plan using ready-made voices.
Written by Sume