Finishing costs: captions and render as a share of a clip
Captions ($0.20) plus a one-minute render ($0.10) add $0.30 to every finished clip. That is 77% of a Grok clip's bill and 9% of a Seedance 2.5 clip's.
Finishing a single clip with burned-in captions and a timeline render costs $0.30 on Sume. For a cheap clip that is most of the bill: 77% of a Grok Imagine clip with its start image, 38% of an H3 Max clip and 9% of a Seedance 2.5 clip at 720p.
The fixed lines
Video captions are a flat $0.20 per job for a video up to 60 seconds. A Timeline render is $0.10 per output minute, rounded up, so anything under a minute bills one minute. Neither depends on the model that made the clip. That makes finishing a fixed cost per deliverable, and it is why cheap models look less cheap once the deliverable is finished.
Grok Imagine video on Sume needs one input image, so the Grok row includes a low-quality GPT Image 2.5 still at $0.02475.
| Clip model | Clip cost | Finishing | Finished total | Finishing share |
|---|---|---|---|---|
| Grok Imagine video 720p, 5 s, with low-quality start image | $0.0872 | $0.30 | $0.3873 | 77% |
| MiniMax H3 Max 768p, 5 s | $0.5000 | $0.30 | $0.8000 | 38% |
| Seedance 2.5 720p 9:16, 5 s | $2.8890 | $0.30 | $3.1890 | 9% |
What to do with this
Three decisions follow. First, batch the finishing: captions are per job and a render is per output minute, so six clips joined into one 30-second video need one render and one caption job, $0.30 for the lot, instead of $1.80 for six separately finished clips.
Second, do not compare models on clip price alone when the product is a captioned video. The gap between Grok and Seedance 2.5 is $2.80 per clip, but the finished gap is the same, because the finishing cancels out. The share changes; the dollar difference does not.
Third, skip finishing steps you do not need. A clip for a feed that adds its own captions needs no caption job.
Check before a batch
Write the finished deliverable as one line: clip cost, plus captions, plus render. Multiply by count, add a retry reserve on the clip line only, and top up. Finishing steps rarely fail the way generation does, so the reserve belongs on the model line.
Worked example: ten deliverables
Take ten finished five-second H3 Max clips. Finished one by one, each costs $0.80, so ten cost $8.00. Joined into one 50-second video with a single caption job and one render, the clip cost stays $5.00 and finishing drops from $3.00 to $0.30, for $5.30. That is 34% less for the same ten shots.
The catch is that you now have one video, not ten. Choose by what you publish: ten standalone posts need ten finishes, one compilation needs one.
- Ten standalone clips: $5.00 plus $3.00 finishing is $8.00.
- One compiled video: $5.00 plus $0.30 finishing is $5.30.
- Captions cover videos up to 60 seconds, so a compilation longer than that needs a different plan.
- Render is billed per output minute, rounded up.
Sources
Related posts
More in Pricing
- Your first $10 on Sume: model, resolution and length tests
Three experiments on a $10 balance, on model, resolution and clip length, plus captions on the winner cost $8.06 and leave $1.94 for retries.
- 5 minute vertical episode: Seedance 2.5 $173.84 or avatar plus $74
A 5 minute vertical episode: $173.34 in Seedance 2.5 shots or $73.50 in avatar plus clips, plus $0.50 to join. What each route can and cannot show.
- FLUX.2 pro 1920x1080 costs $0.045 on fal: the arithmetic
fal gives $0.045 for a 1920x1080 FLUX.2 pro image and $0.03 for 1024x1024. The rounding rule behind those numbers, and how to compare it with Sume.
- FLUX.2 pro edit takes 9 references, 9MP total: price effect
fal says FLUX.2 pro edit accepts up to 9 reference images, 9MP of input in total, and bills input megapixels too. A worked cost range and a Sume check.
Written by Sume