The largest hold one Sume call can place: a wallet sizing table
Maximum catalog band per Sume endpoint, from 13 cents for music to $56.25 for a 300-second Fabric clip. Use the table to size a wallet for concurrent jobs.
How much can a single Sume call reserve?
The catalog's maximum_usd_cents is the most one call can reserve on that endpoint. The highest is VEED Fabric at $56.25, then Video Router at $42.86 and Video 1.0 at $17.42. Most utility endpoints top out below $3.
This helps size a wallet, because each submit reserves its own estimate and the reserves stack across jobs in flight.
What are the maximum holds by endpoint?
Figures are from the catalog.
| Endpoint | Maximum | Driver |
|---|---|---|
| VEED Fabric 1.0 | $56.25 | 300 s at 720p |
| Video Router | $42.86 | Longest, priciest model |
| Video 1.0 | $17.42 | Auto ceiling |
| Image 1.0 / Router | $7.36 | 4 images, max quality, 16 refs |
| Kling 3.0 motion control | $4.73 | 30 s |
| Face swap | $8.25 | 15 s beta maximum |
| Timeline render | $3.00 | 1800 s output |
| H3 Max lip-sync | $3.00 | 14.8 s audio, top resolution |
| Video upscale | $0.27 | 30 s input |
| TTS | $0.95 | 20,000 characters |
| STT | $0.10 | 10 minutes |
| Music | $0.13 | Fixed per generation |
Why not just use the maximum as the price?
Because it describes the worst case. A 10-second Fabric clip reserves $1.875, not $56.25. For your own planning, compute the price from the rate and your input, then use the maximum only to understand the cap on any one reserve.
Avatar videos are different: the catalog bands cover the 4-second floor, so size those from the script length using the per-length table.
How do I check what I have free?
GET /v1/balance returns your balance in USD. Top-ups from $10 to $1,000 are made in the dashboard only, as the credits docs describe, and the API only reads balance and usage. See balance checks in CI for a script.
What is a safe floor for a small team?
Add the reserves of the jobs you expect to run at the same time. Plans set concurrency: Pro allows 4, Startup 8 and Scale 20, and Free allows 1. Four 30-second Plus avatar jobs at $7.35 each need $29.40 at once, for example. Keep spare balance for retries, since a failed job is refunded but a retry reserves again.
What does a worked wallet check look like?
A studio runs on Startup with concurrency 8. Its heaviest jobs are 60-second Plus avatar clips with a product at $15.48 each. Eight in flight reserve $123.84 at once. Add a margin for retries and a $150 balance is a sensible floor.
If the same studio also queues 15-second face swaps at the catalog estimate, include those reserves in the total. The sum of what is in flight is what matters, not any single maximum.
Check GET /v1/balance before a large batch. A top-up is a dashboard action, so build the check into the step before you submit.
Sources
Related posts
More in Pricing
- Luma modify video price: $0.60 vs $1.75 for a 720p 5-second clip
Luma prices Modify Video per million pixels: $0.00544 on ray-flash-2 and $0.01582 on ray-2. Worked example, and how Sume quotes a clip.
- Luma Ray 3.2 API price per second: 540p, 720p and 1080p
Luma lists Ray 3.2 text and image to video at $0.15, $0.30 and $1.20 per 5-second clip at 540p, 720p and 1080p. The per-second math, and 10-second clips.
- Luma Ray 3.2 HDR price: 2x SDR, and 3x with EXR, worked out
Luma's API page prices HDR at twice SDR and HDR plus EXR at three times. On a 5-second 1080p Ray 3.2 clip that is $1.20, $2.40 and $3.60. Sume's output.
- Minimum balance to submit a 10-second AI video job
A 10-second video job needs its estimate in your wallet at submit: $0.125 on Grok, $1.25 on Wan 720p, $3.75 on Omni Flash 4K, or a 402 comes back.
Written by Sume