LESTER
How it works

Your template is the design

Lester never invents a caption style. It reads the one you made and repeats it, which is the whole reason it exists.

STEP 01

Build one caption

Make a single text layer and animate it however you like: a Text Animator, a Range Selector, and an Offset property with keyframes on it. That is the only requirement.

STEP 02

Select it and press Generate

Lester checks the layer first and tells you plainly if something is missing. It will not guess, and it will not produce anything from a template it does not fully understand.

STEP 03

Read the timeline

Captions arrive as layers beside your template, each a copy carrying your styling and your keyframes. Your template switches itself off. One Ctrl+Z undoes the whole run.


What it reads

Everything on the layer

Captions are copies, not re-creations, so anything you put on the template comes along with them.

Type
Font, size, colour, stroke, tracking, leading, alignment, all paragraph settings.
Animation
The Text Animator and its Range Selector, with Offset keyframes re-timed to the real words. Easing, influence and interpolation are preserved rather than approximated.
Everything else
Effects, expressions, parenting, blend modes, masks, transform values.
Timing
Word-level timestamps from the transcription, snapped to frames, with each caption held long enough to actually be read.

Transcription

Runs on your machine

Local, not cloud

A speech model runs on your own computer. Your audio and footage never leave it, there is no per-minute cost, and no account to sign into.

  • Installer builds a private Python environment
  • Model downloads once, then works offline
  • Swap to a bigger model if words come back wrong
  • Add names and jargon as a vocabulary hint

When it struggles

Speech over music is the hard case, and Lester handles it rather than handing you two broken captions.

  • Picks the voiceover track, not the music bed
  • Decodes again without the silence filter if the result looks short
  • Falls back to the rendered mix only as a last resort
  • Names the words it was least sure about, with timecodes