source_too_long_for_reference_ingest: clips over 300 s

source_too_long_for_reference_ingest means the clip you sent to POST /v1/reference-ingest is longer than 300 seconds. The reference-ingest docs say the call reads one clip of 300 s or less, and name the fallback: use video_inspect (and video_segment where it is listed).

This post covers what the limit applies to, how to check a clip before you submit it, and what you lose and keep when you fall back.

What does the 300-second cap apply to?

Reference ingest is the deterministic preflight a reference clip goes through before an agent plans a remix, a brief or a face swap. It reads one media.sume.com clip the workspace already owns, and it answers a ReferenceVideoManifest with shots, on-screen text lines, audio facts and a labeled strip. The cap is on source duration, and it is a refusal at admit rather than a truncation: you do not get a manifest for the first five minutes.

The same 300 s ceiling appears in hypit Understand, which answers source_too_long_for_hypit_understand for the same reason. Neither surface re-encodes the source.

How do I check the length before I submit?

Probe first. Video inspect with frames: false is probe only and unbilled, so it costs nothing to ask for the duration. Its own cap is 1800 s, so it can read clips that reference ingest cannot.

curl -X POST https://api.sume.com/v1/video-inspect \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: probe-001" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_demo/long.mp4","frames":false}'

What do I use for a clip over 300 seconds?

The docs give a short table of what to reach for. None of these produce the full manifest, so treat the result as a thinner read.

What if I only need the opening of a long video?

Most hook analysis only needs the first 30 to 60 seconds. If you own the file, a production trim gives you a new MP4 under the cap, and you can ingest that. Video trim is the documented route for a new cut.

Be honest about the trade: a trimmed clip is a different artifact, so shot times in its manifest are relative to the trim, not to the original. Keep the offset next to the artifact id.

What is a sensible workflow for a long reference?

Treat the long file in two passes. First, probe it and sample stills with video inspect to find the stretch that matters, usually the opening, a product demo or the call to action. Second, cut that stretch to a new clip under 300 seconds and run reference ingest on the cut, so you still get shots, on-screen text and audio facts for the part you plan to remake.

Keep three things next to the artifact id so the result stays interpretable:

• The start offset of the cut in the original file.
• The probe duration, so you can tell later that the manifest covers part of the source.
• The idempotency key you used, since every write here requires one and a retry with the same key does not pay twice.

What does this not do?

Sume does not split a long video into 300 s windows for you, and reference ingest does not accept YouTube or other off-host URLs: the source must be a media.sume.com artifact, asset or chat attachment. Where a flag gates a route, the docs say so: these surfaces are listed where the matching environment flag allows them (development auto-on, production opt-in), so check tools_list or the OpenAPI before you build on one.