video-caption-subtitle-planning
Installation
SKILL.md
Captions and On-Screen Text
Two rules, and the first one is absolute.
Never render text inside the generation. Generative models produce garbled, misspelled and warping text. Even when a single frame looks right, it deforms across motion. Overlay text in an editor.
Second: design the frame for the text before you generate it. Text over a busy area is unreadable regardless of the font, and by the time you're in the editor the frame is fixed.
Why not in the render
| Rendered text | Overlaid text | |
|---|---|---|
| Legibility | Garbled, often misspelled | Perfect |
| Stability in motion | Warps and morphs | Fixed |
| Editable | Re-render required | Instant |
| Per-variant changes | New generation each | One project, many exports |
| Localisation | New generation per language | Swap the text layer |
| Accessibility | Not machine-readable | Real caption files |