Video editing API limits: Sume media tool caps in one table

Sume's video editing APIs cap sources at 300 seconds for video filter and video frames and at 1,800 seconds for video trim, audio detach, and video inspect. A Timeline 1.0 render outputs 1 to 1,800 seconds from up to 200 slots, and timeline compose outputs at most 300 seconds. The table below lists each tool's documented length, count, and size caps.

Each row links the docs page it comes from; the caption limits also come from the Sume API reference. All were read on 2026-09-26. Behavior described as "today" is read from Sume's code.

What are the caps for each media tool?

Lengths are in seconds. "Source" is the input file; "output" is what the job produces.

Which rules apply to every media tool?

A few limits are shared rather than per tool:

• Inputs: trim, filter, detach, frames, inspect, and the three Timeline tools read only your workspace's media.sume.com artifacts or assets. Captions take a public HTTPS video URL.
• Waiting: mode: "sync" holds a request for at most 30 seconds. Trim, filter, detach, and the Timeline tools default to async, inspect defaults to sync, and video frames always answers 202.
• Keys: trim, filter, detach, inspect, and the Timeline tools require an Idempotency-Key; video frames requires it on MCP writes, and the docs say to send it on REST too.
• No argv: ffmpeg fields such as codec are refused, as ffmpeg_fields_rejected on trim, filter, detach, frames, and inspect, and as a 400 on the Timeline tools.

Which caps fail at submit and which fail later?

Request-shape caps fail at submit with a 400. Length caps that depend on the file are checked by the worker after it probes the source, so the job fails instead:

• source_duration_exceeded: a trim or detach source longer than 1,800 seconds.
• output_duration_exceeded: a filter source longer than 300 seconds.
• duration_out_of_range: a video frames source longer than 300 seconds.
• frame_time_out_of_range: an at value at or past the probed duration.
• Some limits clamp instead of failing: a trim end past the source warns trim_clamped_to_source, and a compose video.duration past the file warns compose_duration_clamped_to_source.

What are Timeline's finer limits?

Beyond length and slot count, Timeline 1.0 and compose check these values:

• Transitions: at most 1 second, at most 50% of the shorter neighbor, and at least one output frame; more than 8 adjacent fades fails too_many_chained_transitions.
• render.strategy: "single" is refused above 12 slots (render_strategy_unsafe). Edge fades are 0–5 seconds each, and their sum must fit the output.
• soundtrack: fade_out_seconds ≤ 10 and duck_db 0–20. audio.gain_db: −60 to 12. Video coverage may trail the spine by at most 0.5 seconds.
• Compose stack: ratio 0.1–0.9. Compose overlay: width_ratio 0.05–1 and margin_ratio 0–0.45.

What do the caps mean for a long recording?

A 30-minute recording fits inspect, trim, and detach (detach one range of 900 seconds or less at a time), but not filter or frames. Cut it with trim into clips of 900 seconds or less (300 or less before a filter or frames call), then assemble up to 1,800 seconds with Timeline 1.0. The captions page prices a standalone caption job for videos up to 60 seconds, and today the caption worker refuses a longer source with duration_out_of_range. Split a long video into short clips walks through that workflow.