Video Summarizer

The Summarizer transcribes your video and produces a readable summary you can skim, share with an editor, or turn into show notes and titles. It is the written counterpart to clipping: same Whisper pass (or a reused transcription), then an LLM writes a recap instead of ranking highlight windows.

Long VODs are hard to hand off. A producer should not need to watch four hours to write a description. Summarizer stores the result as a library file next to the source so you can open it from the file detail page, copy it into a newsletter, or use it to decide whether a VOD is worth clipping at all.

How it works

Select a VOD or download (10 GB cap). Credits are charged up front. The worker pulls the source from B2 or the local source cache, transcribes if needed, truncates very long transcripts (about 30k characters) so the LLM stays in context, and retries the summary call a few times on failure. The JSON summary is uploaded and linked from the source file’s library page.

If a compatible transcription already exists — same source, segment duration, and language — it is reused, which is why summarizing after a clip job is often faster than summarizing first.

Useful for

• Catching up on a VOD before clipping
• Drafting descriptions, titles, and newsletter blurbs
• Agency workflows where producers need context fast
• Deciding which recordings are worth a full clip pass

Who it is for

Streamers who want show notes without rewatching. Editors who need a map of the VOD before they cut. Teams that share a library and cannot all sit through every recording.

Summaries are stored as library files alongside the source. Pair with the Clip Generator when you are ready to cut highlights, or the Trimmer if you already know the window.

FAQ

Which files can I summarize?

VODs and downloads. Clips and shorts are usually short enough to watch.

Does summarizing clip the video?

No. You get a text/JSON summary file. Clipping is a separate job and a separate credit charge.

What language does it use?

The same Whisper language setting as your clip-generator config when a transcription is reused; otherwise the job transcribes according to the request and model defaults.