Sume media API price list, October 2026: trim to upscale in one table

Most of Sume's media routes bill a flat amount per job: $0.02 for trim, filter and compose, $0.01 for audio detach and Timeline audio, and $0.20 for captions. Timeline render is $0.10 per output minute rounded up, speech-to-text is $0.01 per audio minute, background removal is $0.0225 per image, and video upscale is $0.009 per input second.

The table below was assembled on 2026-10-10 from the Sume docs pages and the public pricing constants they cite. Treat GET /v1/catalog as the live source if a number differs.

The price table

Two routes are not flat. Video frames and video inspect capture their own Modal compute seconds at the Modal list price times 1.25, plus the platform fee, and never more than the hold. Inspect with transcribe: true adds the speech-to-text per-minute rate on top, while a probe-only inspect is free.

Why the ffmpeg routes are flat

Trim, filter, compose, detach and Timeline audio run ffmpeg on Sume's own worker and call no provider model. Compose is flat because video.duration can be absent until the worker probes the video, so Sume does not wait to price it. That makes budgeting easy: count the jobs, multiply by the rate.

The conventional Sume rule for provider-backed routes is the provider list price times 1.25. The STT constant is documented as a per-minute rate at the standard margin on a $0.008 provider list, which gives $0.01.

Two worked totals

Batch one: 100 clips, each trimmed ($0.02), filtered ($0.02) and captioned ($0.20). Per clip that is $0.24, so 100 clips cost 100 x $0.24 = $24.00. The captions are 83 percent of the bill, so captions are where to look first.

Batch two: a 12-minute narrated edit. The render is 12 output minutes x $0.10 = $1.20. Add one Timeline audio concat at $0.01 for the spine and $1.21 covers both. A 30-minute render hits the maximum of $3.00.

A transcription comparison to handle carefully

Microsoft AI described its first streaming transcription model on October 1, with 60 languages, partial results in just over 100 milliseconds, and a price of $0.54 per audio hour through the end of the year (Microsoft AI article, read 2026-10-10). At $0.01 per audio minute, one hour of Sume speech-to-text is 60 x $0.01 = $0.60.

The two numbers describe different shapes. Microsoft's is a streaming model with live partials; Sume's route is a file job that returns text and word timings. The $0.06 gap per hour is also under a promotional window, so it is not a long-run comparison. Compare by workload, not only by the hourly figure.

Guardrails before a big run

Send an Idempotency-Key on every paid write. Read the balance and generation_limits before a bulk wave. Use the free filter check route for filter programs, and Timeline plan, which is unbilled, to read the billable minutes of a render before you commit to it. A failed job releases or refunds its reservation where applicable, per the Generation admission page, and a retry with the same key does not create a second job.

Two reading tips. First, the text-to-speech rate is per character, and spaces and punctuation count, so a script of 2,000 characters costs 2 x $0.0475 = $0.095. Second, image upscale is $0.20 flat per image no matter the input size, while video upscale grows with seconds, so a 10-second clip costs 10 x $0.009 = $0.09 and a 30-second clip $0.27.