How to put a caption on every mention of a name in a long video
Ask Vizard Agent to find every spoken instance in the word-level transcript and put the overlay on each one. The catch in an edited video is that the timings come from the source, so each mention has to be mapped onto the cut timeline. A gag that misses three out of eleven reads as a mistake.
What do you type into Vizard Agent?
Say "every". A request that reads as an example, such as "put FANGIRL up when she says the name", invites a demonstration on the first few mentions, whereas naming the requirement as exhaustive makes the completeness of the list the deliverable. Give Vizard Agent the variant spellings too, because those are the misses you would never notice.
Prompt
What does Vizard Agent actually do?
Mapping is the whole job. Word timings come from the transcript of the source, and the finished video has had sections cut out of it, so a mention at 9:14 in the recording is somewhere else entirely in the export. Vizard Agent translates each position through the edit rather than trusting the transcript's numbers. In the session behind this article it:
- Transcribed both source videos of a fifteen-minute reaction edit with word timings and found every spoken instance of the name, in each of its spellings.
- Mapped each one onto the edited timeline rather than the source timecode, then added the timed overlays there.
- Read the failed filter command when the caption expression would not parse, and corrected the quoting in the overlay expression.
- Fixed the filter string a second time where the syntax was still wrong, rather than dropping the awkward instances.
- Verified the duration and inspected every caption moment individually, confirming one at each of the eleven mentions.
- Reviewed the overlays on the uploaded copy and checked the caption timing through the opening and the outro.
When does this not work well?
Some words cannot be found reliably. A name that the transcript hears differently each time, one that overlaps with a common word, or speech under loud music will produce misses, and Vizard Agent will tell you which instances it is unsure about rather than quietly skipping them. Overlapping speakers blur the timings too.
Common questions
Can it count them for me? Yes. Vizard Agent reports how many mentions it found and where.
What if the name is misheard? Tell Vizard Agent the variants and it matches those too.
Will the captions survive a re-edit? Tell Vizard Agent about the re-edit and the positions are mapped again.