Top up to run the API
The API, CLI and MCP draw from your balance. Topping up more raises how many jobs run at once.
Metered USD rates for generation over the public API, at the provider’s price with 0% markup, billed per job, second, character, or minute as listed.
Every model in the catalog is billed at the provider’s price, with 0% markup: a workspace without a price book pays what the provider charges, with no margin and no fee on top. The API is paid from the same top-ups as the app, and Sume pays the card fee, so a top-up adds its full amount to your balance.
One-time avatar creation for reusable Sume Avatar 1.0 video workflows.
VEED Fabric 1.0 priced per ceil(audio second). 480p $0.08/s, 720p $0.15/s. Max 300 seconds. Audio must be Sume-hosted and under 10 MB.
One audio generation using Sume Music 1.0.
Sume STT 1.0 transcription priced per started minute of the audio, read from the file; an unreadable file is priced at 10 minutes.
Sume TTS 1.0 priced per transcript character, quoted per 1,000 characters. Spaces and punctuation count. Maximum 20000 characters.
One background-removal generation using Sume RMBG 1.0.
Sume Video Upscale 1.0 priced per second of the source clip, read from the file, at most 600 seconds. The pro tier bills 10 times the per-second rate.
Sume Timeline 1.0 declarative long-form assembly reserved at $0.218266 per ceil(output minute) and captured at the render's own compute at list price, idle time included, never above the reservation. Max 1800 seconds. All media URLs must be Sume-hosted.
Sume Timeline 1.0 audio concat / split reserved at $0.102312 per job and captured at the pass's own compute at list price, idle time included, never above the reservation. One ffmpeg pass over Sume-hosted audio, max 1800 seconds of output. No re-synthesis.
Sume Timeline 1.0 compose (still + video in one frame) reserved at $0.102312 per job and captured at the pass's own compute at list price, idle time included, never above the reservation. One ffmpeg pass over Sume-hosted media, max 300 seconds of output. No provider inference.
Sume Video filter 1.0 (ops[] dim/crop and/or a filters-only filtergraph on one Sume-hosted clip) reserved at $0.102312 per encode job and captured at the pass's own compute at list price, idle time included, never above the reservation; the program check is free. One ffmpeg pass on the worker runtime, max 300 seconds of source. No provider inference.
Sume Video trim 1.0 (a [start, end) range of one Sume-hosted clip → a new MP4) reserved at $0.102312 per job and captured at the pass's own compute at list price, idle time included, never above the reservation. One ffmpeg pass on the worker runtime, max 1800 seconds of source and 900 seconds of output. No provider inference.
Sume Audio detach 1.0 (the audio track of one Sume-hosted video → a durable wav or mp3) reserved at $0.102312 per job and captured at the pass's own compute at list price, idle time included, never above the reservation. One ffmpeg pass on the worker runtime, max 1800 seconds of source and 900 seconds of output. No provider inference.
Sume-managed image generation. Exact bands vary by request.
Sume-managed video generation. Exact bands vary by request.
The provider's price, 0% markup. Per-model rates in the catalog.
The provider's price, 0% markup. Per-model rates in the catalog.
Full rates for managed models and routers: GET /v1/catalog, or the billing docs. sume.com/docs/billing
The API, CLI and MCP draw from your balance. Topping up more raises how many jobs run at once.