Agency avatar roster for 25 clients: creation cost, first month
Twenty-five client avatars cost $23.75 to create on Sume at $0.95 each. A first month of four 30-second clips per client is $552.00 on standard.
An agency that gives each of 25 clients its own avatar pays $23.75 once, at $0.95 per avatar on Sume. A first month of four 30-second clips per client is 100 clips and 3,000 seconds, which is $552.00 on the standard tier and $735.00 on plus at list rates with no product image. The creation cost is small next to the video spend, so the real planning is the queue.
Prices are from Sume API pricing, avatar creation from Create new avatar, and limits from Generation admission, read 2026-10-03.
What does the roster cost?
Each avatar is one job from a prompt, a profile or a reference image, and the returned handle is then reused on every video request. Video is priced per second by tier, and quality defaults to plus when omitted, so send quality: "standard" to get the standard column. Each clip must be 4 to 60 seconds. The table adds the one-time creation charge to the first month.
| Item | Quantity | standard | plus |
|---|---|---|---|
| Avatar creation | 25 avatars | $23.75 | $23.75 |
| Clips, no product image | 100 x 30 s | $552.00 | $735.00 |
| Month one total | $575.75 | $758.75 |
How do you keep clients apart?
Give each client its own handle and keep a table of client, handle and the job ids you receive. The handle may be sent with a leading @, but Sume stores it normalised without it, so store the bare form. Clips reference the avatar with the top-level avatar_handle, and each request is one job, which makes per-client cost easy to attribute: sum your own job records by client.
Derive each Idempotency-Key from client, month and clip number, for example acme-2026-10-clip3-v1, so a rerun of your script does not pay twice. A different body on the same key is a 409 idempotency_conflict, which is useful as a guard against accidentally reusing a key for a rewritten script.
- Create avatars before you quote the month, because creation is its own job and must finish before video.
- Never mix one client's avatar into another client's handle.
Getting 100 clips through the queue
Accepted capacity is 24 on Pro and 48 on Startup, so 100 jobs need waves unless the workspace is on Scale or Enterprise at 120. Submit a client or a week at a time, poll with backoff, and read generation_limits.queue_capacity_remaining before the next wave. A 429 queue_full is a signal to wait, and the same idempotency key makes the retry safe.
If a client revises one script, re-render only that clip, not the batch. At 30 seconds on the standard tier that is $5.52.
Plan the invoice around what Sume does and does not meter. Sume charges per avatar created and per second of video rendered at the tier you pick, so your client price is whatever you add on top. A flat per-clip fee that covers a standard-tier 30-second cost of $5.52 with room for review time is easier to explain than a per-second markup, and the plus tier at $7.35 per 30-second clip can be offered as an upgrade.
Sources
Related posts
More in Use cases
- AI accent wall wallpaper preview from a room photo and a pattern
Preview wallpaper on one wall: a bedroom photo, a photo of the wallpaper sample, a mask on that wall, and a Sume edit that keeps furniture in place.
- AI airport sign photos: set arrows and words in code, not the model
Image models still bend arrows and misspell wayfinding text. Generate a sign with no text, then set the exact arrows and words in Pillow. Runnable script.
- AI app icon via API: generate 1:1, add a transparent option, resize
App icon art from the Sume image API: 1:1 text-free art, ChatGPT Image 2.5 for a transparent background, and a Pillow loop that exports the sizes you need.
- AI mockup of your app screenshot on a phone in a hand
Put an app screenshot on a hand-held phone photo with one Sume image edit: photo first, screenshot second, and a check that the UI text is intact.
Written by Sume