INSIDE THIS AI SKILL
The craft behind its decisions.
These are the principles and procedures the skill applies. You can understand the reasoning without having to perform every production task yourself.
A caption that cannot be read in the time available is not a caption.
Reading speed is the constraint almost nobody sets, and it is the one that decides whether the audience follows the scene or gives up on it.
01 / Captions
The constraints that actually bind
- Reading speed
- Characters per second the viewer can actually absorb. It differs by script and by language, and a limit set for one track does not transfer to another.
- Line length and line count
- A vertical frame is narrow. Two short lines almost always beat one long one, and three lines is usually a sign the line needs cutting rather than wrapping.
- Safe area
- The app puts a handle, a caption of its own, buttons and a progress bar over your frame. Captions have to live where none of that does.
- Timing to the cut
- A caption that survives a cut belongs to the previous shot in the viewer's head. Break on the cut unless there is a reason not to.
- Script rendering
- Devanagari, Tamil and Telugu have different height and spacing behaviour from Latin. A layout that works in one will clip in another.
02 / Captions
Producing them per track
- 01Set a reading speed per language
Before writing any captions. It is the constraint everything else obeys, and retrofitting it means recutting every line.
- 02Caption from the approved script, not from the audio
Transcribing generated audio bakes in any drift from the approved dialogue instead of catching it.
- 03Cut lines for sense, not for length
Break where the sentence breaks. A line split mid-phrase costs the viewer a beat they do not have.
- 04Time to the cut
In and out on shot boundaries wherever the dialogue allows it.
- 05Check inside the actual interface
Not in a player. Over a mock of the platform, at phone size, with the buttons present.
- 06Have the native reviewer read them
Per track, as part of their gate. A caption is a translation decision as much as a timing one.
03 / Captions
Disclosure sits here too
Questions about this skill
Will the AI Skill work out caption timing and placement?
The Captions Skill helps your AI Director prepare readable phrases, timing and placement around faces, products and platform controls. Provide the dialogue, format and agreed language scope. The skill supplies the craft guidance; each delivered language track still follows its review requirements.
Burned-in captions or a subtitle file?
It depends on the destination, and the delivery specification in the brief should say. Burned-in guarantees appearance and forfeits accessibility features and per-market swapping; a file preserves both and surrenders control of the styling. Many platforms want both.
Can captions be generated from the audio?
They can, and for a production with an approved script it is the wrong direction. Captioning from the approved script keeps the text and the dialogue aligned and surfaces any drift as a continuity issue rather than preserving it.
Do all language tracks share caption timing?
They share the picture, so the cuts are the same, but line breaks and timing are per track because line length differs by language. This is one of the places where treating a language as a track rather than a conversion earns its keep.
What reading speed should I use?
Set it per language against your actual audience rather than copying a broadcast standard built for a different screen and a different viewing posture. The thing that matters is that it is set before the captions are written.
Are captions checked automatically?
Accuracy, timing and reading speed are checked in the same pass as visual continuity, along with whether a required disclosure is present. A native reviewer still reads them, because a caption that is technically correct can still be a bad translation.