All notable changes to this Actor are documented here.
A substantial reliability and speed release, with finer control over what you extract and
clearer answers when a video can't be transcribed.
- Transcript cleaning levels — choose
none for the text exactly as published, mild to
remove filler sounds, or aggressive to also strip conversational fluff.
- Optional timestamps — turn the per-segment timestamp list off for a lighter result when
you only need the full transcript text.
- Transcript word count and covered duration on every result, returned whether or not you
keep the timestamp list.
- Transcript availability reason — when a video has no transcript you now get an
explanation (captions disabled, video unavailable, and so on) instead of an empty field.
- Resume support — an interrupted or restarted run skips videos it already delivered, so
you are never billed twice for the same video.
- Concurrency and batch-size settings — tune how many videos are processed in parallel and
how often results are delivered.
- A comments-focused data view, alongside the full, transcript and channel views.
- Documented output — every field in the results is now described with its type and an
example, so the output is self-explanatory in the Console and over the API.
- Result provenance — each record carries its source and originating URL, so rows stay
traceable after you merge datasets.
- Faster runs, particularly when a video or a network hop is slow: requests that stall are
now abandoned quickly instead of holding up the rest of the batch, and independent work runs
in parallel rather than one step after another.
- Videos without a transcript are delivered free. They previously produced nothing at all;
you now receive the full metadata for them at no charge, with a field stating why they were
not billed.
- Results arrive in batches as they complete, rather than only at the end of a run.
- Runs that reach a spending limit now deliver everything already accounted for before
stopping.
- Runs that produce no results explain why — every URL invalid, duplicated, or already
processed — instead of finishing with an empty dataset.
- Paid-plan runs now reliably receive the faster processing path.
- The Standby API description now matches the service exactly: request options, response
fields and streamed events.
- Transcript cleaning and timestamp settings had no effect; both now work as described.
- Word count and duration were advertised in the output but never populated.
- Requesting full metadata through the Standby API could return only the transcript.
- A comment-extraction failure no longer discards the video's transcript and metadata — the
video is still delivered, with the comment error recorded on the record.
- Several settings the Actor accepted were missing from the input form and could only be sent
over the API.
- Long or repeatedly failing videos could stall a run far longer than intended.
- Optional comment extraction alongside transcripts and metadata, including replies.
- Comment sorting and per-video limits for comments and replies.
- Output fields for comment availability, counts and comment data.
- Comments and metadata are gathered together to reduce total run time.
- Clearer handling of videos with comments disabled.
- Transcript extraction with timestamps, plus full video metadata: title, description,
publish date, views, likes and thumbnail.
- Channel details including name, subscriber count and verification status.
- Pre-configured data views for different use cases.
- Batch processing for multiple videos in a single run.
- Improved transcript text cleaning.
- Better handling of videos without transcripts.
- More reliable metadata extraction.
