living-ascii — Claude Code skill for image animation
The living-ascii Claude Code skill turns a still image into a looping ASCII video. It runs when asked to animate a picture or make grass, hair, or brush strokes sway.
MIT ★ 68 updated 2026-11-04
npx skills add yusen-dev/living-ascii --skill living-ascii -a claude-codeWhat living-ascii does
The skill turns a single image into an MP4 loop, with moving strands and an ASCII layer whose glyphs follow their direction. Images with many fine, similarly oriented marks work best: grass, fur, hair, wheat, or brushwork. Buildings, faces, and text may smear.
Run npm install, then start the server with npm start and open its preview in a browser. URL parameters control wind direction, lighting, strand density, and frame size. The ASCII export goes to out/NAME-ascii.mp4; an export without the ASCII layer is also available. After rendering, the skill checks the loop join, visible motion, and color.
What it helps with
- turn a still picture into an ASCII animation
- export a looping video from an image
- animate the sway of grass, hair, or brush marks
Example prompts for the agent
- Turn
input/meadow.pnginto a looping ASCII video. - Animate a gentle sway in the hair in
input/portrait.png. - Export
input/wheat.pngas an MP4 with the ASCII layer and as a strand-only video.
How to install living-ascii
1Skills CLI (Vercel)
npx skills add yusen-dev/living-ascii --skill living-ascii -a claude-code2Manually: put the files in place
~/.claude/skills/living-ascii/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/living-ascii/
1Skills CLI (Vercel)
npx skills add yusen-dev/living-ascii --skill living-ascii -a codex2Manually: put the files in place
.agents/skills/living-ascii/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add yusen-dev/living-ascii --skill living-ascii -a cursor2Manually: put the files in place
.cursor/skills/living-ascii/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install yusen-dev/living-ascii living-ascii2Manually: put the files in place
.github/skills/living-ascii/SKILL.md1Skills CLI (Vercel)
npx skills add yusen-dev/living-ascii --skill living-ascii -a gemini-cli2Platform CLI
gemini skills install https://github.com/yusen-dev/living-asciiManually: put the files in place
.devin/skills/living-ascii/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about living-ascii
How do I install living-ascii in Claude Code?
Run npx skills add yusen-dev/living-ascii --skill living-ascii -a claude-code. The skill file is at ~/.claude/skills/living-ascii/SKILL.md.
What does living-ascii do?
It converts one image into an MP4 loop with moving strands and an ASCII layer. It can also export a version that contains only the strands.
What kinds of images work with living-ascii?
Images with many fine marks pointing in a clear direction work best, such as grass, hair, or brushwork. Buildings, faces, and text may smear.
Source text
SKILL.md · MIT
--- name: living-ascii description: Bring a still image to life as a seamless looping ASCII video. The image becomes ~160k swaying strands that move like a school of fish, with an ASCII layer whose characters turn with the strands beneath them; exports an MP4. Use when someone asks to animate a still image, make an image "move" or "breathe", turn a picture into an ASCII animation or ASCII video loop, or make grass, fur, hair, wheat or brush strokes sway. --- # living-ascii Turns one image into a seamless loop: strands that sway like a school of fish, plus an ASCII layer whose stroke glyphs follow each strand's heading. Output is an MP4 (ASCII, and optionally the strands without ASCII). ## Workflow 1. **Install once.** In the repo folder: `npm install`. 2. **Place the image** in `input/`. Read it first and say whether it suits the effect: best are images of many fine strokes with a clear direction (grass, fur, hair, wheat, reeds, painterly brushwork, wind-swept water). Hard-edged subjects (buildings, faces, text) smear; say so before spending a render. 3. **Start the server** with `npm start` (port 5177; set `PORT` to change it). Run it in the background. 4. **Preview stills** in a browser tab you control (never one the user is watching): `http://localhost:5177/?img=input/NAME&t=0.25`, plus `&mode=flow` for the strands alone. Check: - **Wind direction** (`wind`): default `auto` follows the texture. If strands run against the obvious flow, set degrees (0 = right, 90 = up). - **Light** (`light`): set to where the light in the image comes from (130 = upper left, 45 = upper right, 270 = below). - **Strand size**: strands too fat or sparse for the texture → raise `density`. 5. **Export** with `?img=input/NAME&export=ascii,flow` (add any tuning parameters). Keep the tab in front until the title reads `done` (or `failed`). A 12 s loop at 1080 px wide takes about 10 minutes. Results: `out/NAME-ascii.mp4`, `out/NAME-flow.mp4`. 6. **Verify** before handing over (see Checks), then report the file paths and anything that looked weak. ## Parameters (URL) `img` (default: the sample), `mode` ascii|flow, `width` (default 1080), `height` (default: follows the image; with `width` set, the image is cropped to that shape), `focus` x,y 0–1 (which part to keep when cropping, default 0.5,0.5), `wind` auto|degrees, `light` degrees (default 130), `amp` sway (default 1), `density` (default 1), `loop` seconds (default 12), `fps` (default 30), `t` 0–1 freeze, `export` ascii,flow. **Wallpapers.** Set `width` and `height` to the screen's native resolution (e.g. 3840×2160 desktop, 1290×2796 phone) and choose `focus` from stills. Never render small and upscale: glyphs are 5×7 px and blur. A longer `loop` adds no variety: every motion term repeats a whole number of times per loop, so it only slows the motion down. ## Checks Run these on every export: 1. **Seamless loop**: `npm run check -- out/NAME-ascii.mp4` must print `✓ seamless` (last→first change no bigger than an ordinary frame step). 2. **Motion reads at phone size.** Scale two frames 0.5 s apart to 360 px wide and compare: PSNR should be about 25 dB or lower. If it is near 29+, motion is too subtle — raise `amp` (try 1.3–1.5). Users judge this on a phone, where small sway disappears. 3. **Colour matches the image.** Scale the ASCII frame and the source image to ~108 px wide and compare per brightness band: ASCII/source brightness ≈ 1.0–1.1, saturation ≈ the source's. Bright pixels must not be white-washed or neon. Do not "fix" dullness by pushing exposure or stroke saturation; that is what made earlier versions look blown-out and fluorescent. 4. **No black specks**: none should appear; if they do, a shader produced NaN (see Pitfalls). ## Pitfalls already solved — keep them solved - **NaN specks.** MSAA can extrapolate varyings slightly outside 0–1; `pow()` of a negative is NaN, which mipmapping spreads into black squares. Clamp before `pow`. Guard `atan(0, 0)`. - **Loop seam.** Every moving term must repeat a whole number of times per loop. Noise that drifts must use the periodic `pfbm` with an integer period on the drift axis. - **GLSL comments** must not end in a backslash (it continues the line and swallows the next one). `half` is reserved in GLSL — don't use it as a name. - **Brightness by angle.** Stroke glyphs are fitted so each weight lights the same pixel count at every angle; otherwise horizontal strands read darker than diagonal ones. - **Highlights.** Strokes keep their cell's own saturation and only ease toward warm cream at the very top. Whitening them loses the colour; boosting their saturation looks neon. - **Shadows.** Shadows keep thin strokes plus a glow so they read as dark colour, not black. ## Changing the look - Motion: `STRAND_VS` in `src/field.js` (`amp`, the `wave`/`sub`/`lean` mix, `own`). - Light: `sunDir`, `sunPool`, `poolTint` in `src/field.js`. - Glyphs and colour: `WEIGHTS` and the colour solve in `src/ascii.js`; the dot ramp in `src/font5x7.js`.
© yusen-dev · MIT · yusen-dev/living-ascii