Agent Skills
› Forward-Future/scarlet
› motion-graphics
motion-graphics
GitHub指导生成视频动态图形(如下三分之一标题、字幕等),确保风格统一、可读性强且时机恰当,遵循项目规范与视觉安全区域。
Trigger Scenarios
需要为视频添加标题或说明性文字
需要制作下三分之一栏或结尾卡片
Install
npx skills add Forward-Future/scarlet --skill motion-graphics -g -y
SKILL.md
Frontmatter
{
"name": "motion-graphics",
"category": "Picture and graphics",
"description": "Make motion graphics over the picture (lower thirds, titles, callouts, end cards) that read well on this footage, in the project's style."
}
Motion graphics
A graphic is there to be read, then to get out of the way. Make each one legible on the footage it sits on, in the project's look, and on screen only as long as it earns.
The look
- When the project has a style guide, follow it as written: its fonts, colors, sizes, placement and motion. Where it is silent, keep to the spirit of what it does say rather than adding a style of your own.
- With no guide, take the look from what the editor has already approved in this project, so every graphic in a video reads as one family. A first graphic with nothing to go on stays plain: one clean sans-serif, white or near-black, one accent at most.
- When a saved template fits what was asked, use it rather than drawing something new, and keep its look. Name a field's value as the editor would say it.
Legibility on the actual picture
- Check the graphic over the frames it will sit on, at the moment it is fully in. Text should hold its contrast against the brightest and busiest part of the picture behind it; add a backing panel or a soft shade when the footage won't give it that.
- Keep clear of faces, eyes and mouths, of text already burned into the picture, and of anything the viewer is meant to be looking at. A lower third sits low and to one side, inside the frame's safe margins, never over the speaker's chin.
- Size for the screen the video will be watched on: names large enough to read at a glance, secondary lines smaller but never squinting size.
Timing
- Bring a graphic in on a beat: as a person is introduced, as a point is made, on a cut. Not mid-word, and not in the same instant as another change on screen.
- Hold it long enough to read twice at a relaxed pace, about three seconds for a name and role, longer for more words, then take it off before it overstays. Leave it off the last moments before a cut unless it is meant to carry across.
- Motion is quick and quiet: a short ease in and out, a few tenths of a second. Nothing that bounces, spins or competes with the speaker.
Restraint
One graphic at a time, and only where it helps the viewer: who is speaking, where we are, a figure worth seeing. When in doubt, leave it out and say so to the editor.
graphic_render, template_list, graphic_get and timeline_propose cover what this needs.
Version History
- b6bdb58 Current 2026-11-18 20:45


