source-video-prompt — Claude Code prompt: what it does, how to install
The source-video-prompt prompt for Claude Code structures video generation requests for converting motion clips into code. It specifies rules for flat shapes, timing ranges, and linked transitions.
MIT ★ 361 updated 2026-10-20
/plugin marketplace add lovisdotio/motion2code
/plugin install motion2code@motion2codeWhat source-video-prompt does
The guidelines restrict generated clips to solid flat colors without shadows, gradients, or camera terms like pan and zoom. Shapes transition through morphs, slides, and merges that map directly to SVG elements.
Additional instructions cover color palettes, looping behavior, and exact timing per second. The reference setup was tested with MiniMax H3 Max on fal.
What it helps with
- write flat motion clip prompts without camera moves
- define time ranges for animation beats
- link sequential shape transitions
- set up color palettes and loops for code export
Example prompts for the agent
- Run the source-video-prompt for motion design
- Generate flat motion clips with source-video-prompt
- Write a video prompt using source-video-prompt rules
How to install source-video-prompt
1Plugin marketplace
/plugin marketplace add lovisdotio/motion2code/plugin install motion2code@motion2code2Manually: put the files in place
.claude/commands/source-video-prompt.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about source-video-prompt
How to install source-video-prompt in Claude Code?
Run /plugin marketplace add lovisdotio/motion2code, then /plugin install motion2code@motion2code, and access .claude/commands/source-video-prompt.md.
What does source-video-prompt do?
It provides prompt instructions for generating flat motion clips that are easy to rebuild faithfully in code.
What is required to run source-video-prompt?
The prompt is designed for generating video sources via MiniMax H3 Max on fal before writing code.
Source text
source-video-prompt.md · MIT
# Writing source prompts that convert well
Any flat motion clip can be rebuilt, but some videos are much easier to rebuild faithfully. When you generate the
source yourself (the gallery used MiniMax H3 Max text-to-video on fal), these habits help:
- **Flat only.** "Every shape is one solid flat color: no shadows, no shading, no highlights, no gradients, no
textures." Each extra effect is something the code must fake or drop.
- **No camera language.** Do not write "3D", "camera", "zoom", "pan" or "cut". Describe what the shapes do: they
slide, grow, split, merge, morph, wipe. Shapes map to SVG; camera moves map to guesses.
- **Linked beats.** "The last shape of each moment becomes the first shape of the next." Chains of morphs and
wipes are easy to time from the frames and read as one continuous film.
- **Second by second.** Give a time range per beat ("0.0-0.8 s: a white almond eye shape opens"). The shot list
then almost writes itself, and you can check the video against your own timing.
- **Exact words, few of them.** "Every word is spelled exactly as written and no other words appear." Short
uppercase words in a bold sans-serif match Google Fonts well.
- **Loop.** "At the end everything returns exactly to the opening frame, so the film loops seamlessly." The video
will not always obey; the code loops anyway, but a source that loops needs no invented return.
- **Small palette.** Name 3 to 4 colors. Palettes sample cleanly and stay stable across the clip.
- **Interfaces.** "A single continuous screen recording of a real, working website, no browser chrome, small
round cursor, running text shown as rows of thin flat bars, the only readable words are ..." gives a replay
that can also become a real site.
Example (the source of `gallery/eye`, abridged):
> Award-winning flat 2D kinetic typography and motion design: bold sans-serif words and only basic geometric
> shapes, solid flat colors, no gradients, no shadows, no textures. Twelve shots in ten seconds, and every
> single transition is linked. Palette: black, white and electric blue. 0.0-0.8 s: a white almond eye shape
> opens. 0.8-1.7 s: the pupil is a blue circle. 1.7-2.5 s: the pupil becomes the O of "LOOK". ...
> 9.2-10.0 s: the eye closes into a line. Snappy easing with anticipation, overshoot and squash. Every word is
> spelled exactly as written and no other words appear.
© Lovis Odin · MIT · lovisdotio/motion2code