hand-drawn-canvas-animation
alesha-pro/tools · Agent Skill
Hand-drawn films in code: the agent draws in ink, pencil, riso or doodles on photos and renders an MP4 — no Blender or video models.
Jump to install ↓Overview
| Publisher | |
|---|---|
| Repository | alesha-pro/tools ↗ |
| Skill folder | hand-drawn-canvas-animation |
| Stars | 616 |
| Forks | 66 |
| Bundled files | 57 |
| License | MIT |
| Last updated | September 25, 2026 |
| Links | Skill source ↗ · Repository ↗ |
Mozgi check
What we found in the skill files. Details come with a subscription.
About this skill
Hand-drawn canvas animation teaches an AI agent (Claude Code, Codex and others) to make short hand-drawn films in code — JavaScript and Canvas 2D. The output is an HTML film player and a finished MP4 with sound. No Blender, WebGL or video models: every frame is drawn with strokes, like real animation.
Becoming — a 60-second example film: five drawing styles, sand animation and a pop-up bookWhat makes it different
Instead of a puppet built from blobs, the skill works like an animator: whole-pose drawings (keys and breakdowns), expressive pressure strokes, intentional holds and an exposure sheet. Motion reads as drawn by hand.
The main study: nine whole-body keys, assisted inbetweens and a replacement blinkFive looks
- Ink — confident brush gestures, black masses and white gaps.
- Riso — overprinted colours, halftone, slight misregistration.
- Screen print — flat opaque shapes, stencil edges, wear.
- Pencil — thin graphite, open contours, form hatching.
- Doodles on photos — a cut-out photo on pastel paper with brush-pen drawings that draw themselves.
Three extra techniques
- Found motion — trace real movement from a video into strokes.
- Sand animation — a backlit sand table, poured, wiped and blown in one take.
- Paper in space — drawn sheets stand up in 3D as a pop-up book.
How to use
Ask for a film with a subject, length and look, e.g.:
20 seconds, the life of a request inside a GPU server, riso look
Requirements
- Node.js 22.12+, Chrome or Chromium and ffmpeg for rendering.
- Python with
numpy scipy scikit-image pillowfor found motion. rembgfor clean photo cut-outs in the doodle look.
By alesha-pro, MIT license. Not for UI animation or slide decks.
Installation
A skill is a folder with a SKILL.md file. Pick your agent and run the commands in a terminal (macOS, Linux or Git Bash on Windows) from your project root.
Copy the prompt into Claude, Codex or another assistant: it will review the skill and install it for you.
Install the “hand-drawn-canvas-animation” skill from mozgi.io (https://mozgi.io/skills/hand-drawn-canvas-animation/) for me. 1. Download the archive: https://mozgi.io/skills/hand-drawn-canvas-animation/download If it fails, take the skill folder from the source: https://github.com/alesha-pro/tools/tree/main/skills/hand-drawn-canvas-animation 2. Read SKILL.md and the other files. Check the skill is safe: no hidden commands, no sending data elsewhere, no scripts doing something other than described. If anything looks suspicious, stop and tell me. 3. Put the “hand-drawn-canvas-animation” folder into your agent's skills folder: .claude/skills/ for Claude Code, .agents/skills/ for Codex, .cursor/skills/ for Cursor, .gemini/skills/ for Gemini CLI. Inside the project if we are in one, otherwise in the home folder (~/.claude/skills/ etc.). 4. Make sure <skills folder>/hand-drawn-canvas-animation/SKILL.md exists. 5. Briefly tell me what this skill does and offer to try it on my task. If you cannot do a step yourself, tell me what to click or type.
git clone --depth 1 --branch main https://github.com/alesha-pro/tools.git /tmp/tools mkdir -p .claude/skills cp -r /tmp/tools/skills/hand-drawn-canvas-animation .claude/skills/hand-drawn-canvas-animation
Restart Claude Code — the skill appears in the list; call it manually with /hand-drawn-canvas-animation. For all projects use ~/.claude/skills.
git clone --depth 1 --branch main https://github.com/alesha-pro/tools.git /tmp/tools mkdir -p .agents/skills cp -r /tmp/tools/skills/hand-drawn-canvas-animation .agents/skills/hand-drawn-canvas-animation
Codex looks for skills in .agents/skills from the current folder up to the repo root; call one with /skills or $hand-drawn-canvas-animation. For all projects use ~/.agents/skills.
git clone --depth 1 --branch main https://github.com/alesha-pro/tools.git /tmp/tools mkdir -p .cursor/skills cp -r /tmp/tools/skills/hand-drawn-canvas-animation .cursor/skills/hand-drawn-canvas-animation
Cursor picks the skill up on start; call it with “/” in Agent chat. Cursor also reads .claude/skills and .agents/skills. For all projects use ~/.cursor/skills.
git clone --depth 1 --branch main https://github.com/alesha-pro/tools.git /tmp/tools mkdir -p .gemini/skills cp -r /tmp/tools/skills/hand-drawn-canvas-animation .gemini/skills/hand-drawn-canvas-animation
Gemini CLI offers to activate the skill when a task matches — you approve it. For all projects use ~/.gemini/skills.
Read SKILL.md and any companion files before deciding whether to install. Click a file to preview it.
SKILL.md contents
Original author text · read-onlyHand-drawn canvas animation
Produce a film that feels drawn by hand in both its marks and its movement.
For character animation, default to whole-pose drawings, expressive strokes and
intentional exposure. Design the keys and breakdowns before any rig or texture.
A rig may guide construction and contacts; the final drawing must carry the pose.
Canvas 2D is the renderer; SVG path data may be used for authored shapes. No
Blender or video model is needed. WebGL is not a dependency. Deliver an HTML
film with its local modules and an offline MP4. Length and format follow the
brief; a short, finished action is the starting point for a longer film.
Choose the right references
- Read style.md ↗ for the five looks and their quality gates.
- For a film combining looks or techniques, read
mixed-media.md ↗. Plan why the material changes and
what carries through the transition before drawing separate style samples.
- Read redrawn-animation.md ↗ for the default
stroke-led character workflow, whole cels, assisted inbetweens and exposure sheets.
- Read motion.md ↗ before animating characters or a camera.
- Read studio.md ↗ for exposure tracks, stable strokes,
contact IK and materials for optional deformation and construction helpers.
- Read architecture.md ↗ for core APIs and export.
- Read doodle.md ↗ for photo sourcing, anchors and masking;
found-motion.md ↗ for traced movement;
sand.md ↗ or paper3d.md ↗ for those media.
- palettes.md ↗, scenes.md ↗ and
reference-films.md ↗ contain optional palettes,
scene devices and historical references. They are not mandatory film structures.
Workflow
1. Fill brief-template.md ↗. Infer routine choices;
ask only for missing decisions that materially affect the result. Specify the
intended material and acting reference separately. Pick a primary aspect ratio;
alternate formats need composition review, not just a different canvas size.
2. Create one project folder. Copy assets/core.js, assets/studio.js,
assets/cels.js, optional assets/materials.js, scripts/render.mjs and
scripts/package.json into it. For characters, study
examples/sketchbook-bird.html and build an original cast with whole drawings.
Its example imports use ../assets/; change these to local module paths in a
standalone project. assets/film-template.html is a legacy rig demonstration,
useful for runtime wiring, not the default visual model. Install the local
renderer dependencies with npm i --no-audit --no-fund; system Chrome and ffmpeg
are required. Keep reusable engine edits in the skill, not in film copies.
3. Establish the drawing: silhouette, proportions, expressions, necessary views,
line hierarchy, recognizable stroke gestures and a few material samples. Render at delivery size and at
240 px. A material cannot rescue an unclear pose. Use designed curves and
replacement drawings where simple primitives fail.
4. Plan the keys and breakdowns of the hardest action. Mark contacts, gaze,
anticipation, weight transfer and reaction. Use enough controls to express it;
there is no fixed part count or parameter limit. Build a finite exposure sheet
of whole drawings. Use geometric inbetweens only for corresponding forms;
redraw changed overlaps, head angles, compressed silhouettes and expressions.
Establish timing without decorative textures first. Do not add springs to rigid or motor-driven motion.
5. Finish one representative production shot before building the whole film.
Evaluate normal-speed playback, consecutive-frame strips and stills. Correct
anatomy, spacing, contacts and line stability before adding more scenes.
6. Build the remaining shots using a beat sheet: start, duration, what the viewer
notices, main action, camera and sound cue. Cuts, transitions, blueprint views,
montages, anchors and sign-offs are artistic choices. Give each action time to
read. When mixing materials, carry the character's action, position or a
physical object across the change. A palette swap alone does not establish
a new style; several impressive but disconnected shots do not establish a story.
7. Render and inspect. Start with --grid 24, then use --strip 48,12 around
a fast action or contact, and --only 48 for full-size detail. Run the full
renderer for the MP4 and generated score. A contact sheet alone cannot pass
motion QA. If playback is unavailable, explicitly report that limitation.
8. Deliver the source, dependencies/assets, MP4, contact sheet and the remaining
known limitations. Verify duration, dimensions, frame count and successful
decode. Never call a film studio-quality solely because its code or checklist
passed; assess the actual result against the chosen visual reference.
Production invariants
defineFilmdefaults to 24 fps. A scene receives continuoustau;i
remains the legacy 12 Hz index for pulse, boil and older clips. Use exposure
tracks per action. Ones, twos, threes and held drawings are all valid.
- Separate camera timing, drawing exposure and root motion. Check their combined
screen-space motion, especially in a tracking shot. Avoid quantizing twice.
- Randomness is seeded. Make strokes once per drawing, with stable semantic ids.
A held drawing holds its marks too. Seed changes belong to a new drawing, not
every output frame. In a deformation workflow, use stable rest-space marks. Specify
whether a texture belongs to the paper, an object or the screen. A fixed seed
does not prevent swimming if sampling positions or topology change.
- Each frame must reproduce after seeking in the same pinned runtime. Sand uses
fixed-step simulation and bounded checkpoints. Font and browser differences
can change pixels between machines; provide fonts or use path lettering when
reproducibility matters. Allow fonts and photos to load before export.
- Preserve physical contacts until their planned release. IK clamps unreachable
targets; check the returned error and fix the pose rather than stretching bones.
- Record sources/licences for external media. Use only assets permitted for the
intended delivery. Optional SVG is path geometry, not automatic rigging.
Quality gates
Drawing: clear silhouette; consistent proportions across views; controlled
outer/inner line weights; purposeful pressure, taper and gaps; no accidental
tangencies or broken joints. Avoid assembled oval limbs with a texture laid over
them. Clean curves and aligned fills remain valid for the selected medium, but
the default sketch should retain visible drawing gestures. Wobble,
misregistration and boil require a purpose.
Motion: whole poses that feel redrawn, meaningful keys, readable breakdowns, deliberate spacing, convincing
weight and planted contacts. Inspect overshoot, camera tracking, topology changes
and fast movement at normal speed and frame by frame. Use a designed smear when
needed; legacy smear() makes a ghost trail, not a new drawing.
Material: ink hatching follows form; graphite respects pressure and tone;
riso plates register coherently; screen ink may be solid; doodles interact with
the photo. Avoid excessive grain, changing raster phase and unintended shimmer.
For mixed-media work, inspect the transitions as actions, not only a style board.
If weather drives the scene, its force should affect both setting and character.
If paper unfolds, show a compact starting state, the hinges and an attached base.
Delivery: no page errors, accidental blank frames, clipped action, stale
frames or unfinished final lettering. If a closing card is requested, let it
finish and hold long enough to read. Check the encoded video, not only PNGs.
Working examples and commands
examples/becoming-phoenix/phoenix.html ↗
is a complete 60-second film with original music. It combines all five looks,
sand and paper projection through one story. Study inkScene for a moving
storm, ensurePrintedPage for an actual print becoming a page, and
unfoldPhoenix for staggered wing and tail opening. Its branding is specific
to the example. Use the techniques that serve the new brief, not its shot list.
examples/sketchbook-bird.html is the primary pencil/ink drawing study: nine
whole-body keys, assisted inbetweens, a replacement blink and an exposure sheet.
examples/weight-study.html is a secondary seven-second contact/weight study rendered in
five looks with --look ink|pencil|riso|screen|doodle. It demonstrates mechanisms,
not the aesthetic target for redrawn characters or proof of feature-film quality. examples/material-studies.html adds sand-displacement and page-curl close-ups. Earlier
examples remain useful for particular scene devices; see their references above.
node scripts/render.mjs examples/sketchbook-bird.html --look pencil --grid 18 --out /tmp/pencil-preview
node scripts/render.mjs examples/sketchbook-bird.html --strip 58,24 --out /tmp/motion-preview
node scripts/render.mjs examples/sketchbook-bird.html --width 1920 --out /tmp/drawn-film
node scripts/verify.mjs
The renderer honours the film's aspect ratio unless --ar is supplied. Each full
render stages its own frames, then replaces that film's outputs after successful
encoding. Old PNGs cannot extend a shorter revision. --grid, --strip and
--only are preview modes and do not replace the full MP4.
For Remotion integration use timebase conversion:
drawFrame(Math.min(__NDRAW - 1, Math.floor(frame * __fps / compositionFps))).
With film and composition both at 24 fps this is simply drawFrame(frame).
Frequently asked questions
What does the hand-drawn-canvas-animation skill do?
Hand-drawn films in code: the agent draws in ink, pencil, riso or doodles on photos and renders an MP4 — no Blender or video models.
How do I install hand-drawn-canvas-animation?
Open the “Installation” block on this page and pick your agent — Claude Code, Codex, Cursor or Gemini CLI — for ready-made commands that copy the skill folder to the right place. Or copy the prompt from the “Via assistant” tab and let the assistant review and install the skill.
Which AI models work with hand-drawn-canvas-animation?
Skills are plain Markdown instructions (the Agent Skills format), so hand-drawn-canvas-animation works with any agent that reads SKILL.md: Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and more. The model loads the instructions only when a task matches the skill's description.
Is hand-drawn-canvas-animation free?
The skill is published openly by alesha-pro under the MIT license. The skill itself costs nothing — you only pay your AI provider for the tokens you use.
Is it safe to install skills?
A skill is a set of instructions your agent will follow, so read SKILL.md and its companion files before installing. Scripts in the skill folder only run if your agent runs them — keep an eye on what it does and do not grant unnecessary permissions.
How is a skill different from an MCP server or a plugin?
An MCP server or plugin gives the model new tools — code that runs and returns a result. A skill gives it knowledge and process: how to approach a task, which steps to follow and what good output looks like. It needs no server or API key.