Clips: trim then caption; do not caption the VOD

Trim first. A Twitch VOD is a source file. A clip is one range. Captioning the VOD is how you pay a transcription engine to typeset four hours of queue pop, death chat, and silence, then discover the joke lived at 2:13:40.

Trim a video is cut_video with one [start, end) range in seconds from the start of the source. It returns just that portion. It is not a re-edit, not a reframe, and not a caption pass. Add captions to a video is the next named job: VEED transcribes spoken audio and burns styled, timed text. Speech in, subtitles out. The clip has to exist before that job has anything honest to hear.

The VOD is the source, not the post

Channels that live on clips already know the beat: one death, one read, one clutch. The rest of the recording is inventory. Feeding the whole file to captions treats inventory as a deliverable. Overlapping Discord, game SFX, and a second mic degrade the transcript. Four hours of that is a long, noisy input, not "thorough."

cut_video keeps one video. Timestamps are seconds from the start. If you need to drop a middle section and keep both ends, that is cut a section out of a video. Isolate the moments first. A generate invents a scene; it does not seek 2:13:40. Ask estimate_cost on a long source if you want the number first. Editor previews and final export is the formula family: charge count, not a per-second generate.

One range, then the burn

Order is load-bearing:

1. Mark in and out on the VOD.
2. Run the trim. Get a new file.
3. Caption that file.

Add captions transcribes in 165 language codes. It does not translate. BASIC presets (simple, plain, beans, corpo, and the rest — 21 looks) run at the standard 1× rate. DYNAMIC presets (glass, whisper, glide, fusion, terminal, handwritten, backdrop — 9 looks) cost 2×. Preview a style on the first seconds with preview_caption_style before you commit a long render. Full-captioning a DYNAMIC look you have not seen is how double-rate becomes a surprise.

This tool typesets what someone says. A silent huddle, a facecam with no mic, or a clip that is only game audio will give the transcriber nothing useful. If the hook is a line you wrote ("HE FLIPPED THE THROW"), that is a text overlay — authored copy, not a transcript. Do not run VEED captions hoping the headline appears.

Transcription hates four hours of dead air

Mute autoplay is why clips need words on screen. That is an argument for captioning the clip. A VOD is a backlog with no hook until you cut. Trim until the speech you care about is most of the file. Then burn.

add_dynamic_captions is retired. VEED is the only captioning engine live in the app. Do not hunt an in-house style on a four-hour file.

Order is the product

Would you post this range with no captions?

If no, because the words are the joke — trim to the joke, then caption.

If the words were never spoken — overlay, after the trim.

If you have not picked a range — you do not have a clip yet. Captioning will not find it.

FAQ

Can I caption the VOD and trim later?

You can. You will have paid to transcribe everything you then throw away, including DYNAMIC 2× if you picked a premium look. Trim first. Caption the keep.

Does trim work if I only want to drop the last ten minutes?

Yes. Pass a single segment covering the range to keep. Anything after the end is dropped. Same for dead air at the start.

Will captions translate the stream into English?

No. Set the language code to what is actually spoken. Translation is a different editing task.

What if I need three moments from one VOD as one short?

That is still not "caption the VOD." Isolate or stitch the ranges first, then caption the short. cut_video is one source at a time; several source videos need a compose path, not a caption pass on each archive.