Shortest AI avatar video: 4 seconds, from $0.74 on Sume Standard

The shortest Sume avatar video is 4 seconds, and on the Standard tier that costs $0.736 at the published $0.184 per second, shown in the catalog as an estimate of 74 cents. Plus is $0.98 for the same 4 seconds and Max is $2.20. Avatar video length is accepted from 4 up to 60 seconds, and Plus is the default quality when you leave the field out.

Creating the avatar itself is a separate flat charge of $0.95 per avatar, paid once. After that, each talking video is priced per second on its tier. This post lists the rates and works out clip lengths, so you can budget a batch before submitting it.

What are the per-second rates?

The rates below come from Sume's catalog and pricing code. Adding a product image raises each tier's rate by a small amount, and the catalog's maximum estimate for a clip includes that product rate.

What do longer clips cost?

Multiply the length in seconds by the tier rate. These figures use whole seconds and the no-product rate. They are straight multiplication of the published numbers, not catalog estimates, and the amount reserved at submit comes from your real script length.

Why is 4 seconds the floor?

The docs state that the estimated duration must land between 4 and 60 seconds inclusive, and that longer scripts must be shortened or split. The catalog basis says the same thing in billing terms: the estimate is the tier rate times 4 billable seconds, the shortest chunk-planned clip.

Multi-segment videos with speech and silence beats follow the same window. Total planned duration must land inside 4 to 60 seconds, and a silence beat needs an explicit duration.

A worked batch makes the spread concrete. Ten 15 second Standard clips come to $27.60 at the no-product rate, the same ten on Plus come to $36.75, and on Max to $82.50. Adding the one-time $0.95 avatar creation brings the Standard batch to $28.55, as long as all ten clips use the same avatar. Switching those ten clips to the product rate on Standard adds a flat $0.01 per second, or $1.50 across the batch.

Captions are optional and are burned into the final MP4 after generation. The docs say inline captions are rejected when the estimated duration is above 60 seconds, which is the same ceiling the video itself has, so a script that fits the window also fits the caption rule.

• Minimum planned length: 4 seconds.
• Maximum planned length: 60 seconds.
• Default quality: plus.
• Avatar creation: $0.95 flat, once per avatar.
• Product image: higher per-second rate on every tier.

What happens if my balance is too low?

Sume reserves the estimate when you submit. If the balance cannot cover it, the API returns 402 insufficient_credits and no job starts. When a job fails, the reservation is released rather than captured. Read the balance first, then compare it with the figure for your length and tier.

curl https://api.sume.com/v1/balance \
  -H "Authorization: Bearer $SUME_API_KEY"

Which tier should I pick for a short clip?

For a 4 second test, the spread between Standard and Max is under $1.50, so test on Standard and move up when the script is final. For longer ad or demo cuts, the 60 second row shows how fast the gap widens, from $11.04 to $33.00.

A practical plan is to write the script first, estimate the spoken length, pick a length bucket from the table, and only then choose the tier. If the first take misses the window, shorten the script rather than the tier, because every extra second costs the full per-second rate. Keep the balance check in the loop so a batch stops cleanly on a 402 instead of failing halfway.

Two last points. The estimate in the catalog is a floor and ceiling for a request, with the 4 second figure as the minimum, so the number reserved for a longer script will be higher than the 74 cents on the headline. And the avatar creation charge is separate from every video, so a second video on an existing avatar does not pay it again. Read the catalog entry for the tier you plan to use before you commit a large batch, because it carries the rate card version and the basis for each estimate.