Riso still
SkillMediaLets your agent create and refine risograph-style artwork, posters and illustrations as PNG files.
Available today. Use it from your connected AI after setup.
No other account needed.
Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.
Then ask your AI: use the Riso still skill
About this skill
Create or improve procedural risograph still art, illustrations, posters, print series and images in this repo. Develops reference-led composition, perspective, pose and ink plates; delivers a native-resolution PNG and self-contained Canvas 2D source. Use for standalone still requests; use riso-film
What this skill tells your AI
The instructions your AI receives, as published by sevenevesai/riso-windowseat in .claude/skills/riso-still/SKILL.md and read by ahel’s review.
Deliver prints/<name>/index.html and a native PNG in out/. Same medium as the films: Canvas
2D only, no libraries, fonts, images or network calls. External references are observation
sources, never embedded assets. If the user explicitly wants generated raster imagery, that is a
different workflow; do not quietly substitute it. Commands run in tools/.
Develop the picture
Read .claude/rules/riso-plates.md, docs/visual-development.md and docs/drawing.md (its
composition and frame-budget sections apply in full); add docs/scene-space.md when perspective or
attached details matter and docs/characters.md for figures. Skip film timelines, transitions and
sound. examples.md maps the shipped series (prints/cabinet, prints/sceneries,
prints/workings); open the matching source before inventing a routine it already has.
Extract subject, feeling, format and exclusions. If the user wants ideas, offer different scenes or viewpoints; if they want art, choose and draw. Default to 1080 square only when no format was given; record other sizes before composing.
- An open request ("a collection", "detailed sceneries", "your strengths") is permission to
art-direct a series of three to five prints under one principle stated up front: one
construction style held across different subjects (
prints/cabinet), or one technical strength per print (prints/sceneries: constructed perspective with fog, lights with reflection, strata with layered haze). - For an unfamiliar subject, look at real references and note proportions, gesture, construction and material; measure in the photo's own pixel frame where it matters (wing outline, vein and eyespot positions). A mood adjective or an unopened image link is not enough.
- For ambitious work, compare small compositions with different viewpoints and value groupings, then develop the strongest. A local edit needs less setup.
- Solve support, perspective, occlusion and identity before detail. Draw figure landmarks and contact points before clothing or texture: the hand reaches the tool, the tool meets the work. Build details on shared surfaces so ellipses, slats and legs agree on orientation and scale. Grow natural subjects from their own geometry (a logarithmic spiral, Vogel's floret model, a distance transform from a cavity wall), so every mark sits where growth put it.
- A film's 0.25 s reading time does not govern a still: sustained detail is allowed, but marks should explain form, material or the moment. Concentrate detail at the focus and leave quiet areas. Random hatching and misregistration do not rescue weak drawing.
Build and inspect
node new-riso.mjs --kind still --out ../prints/<name>/index.html
node verify.mjs ../prints/<name>/index.html --times 0,1,2
node still.mjs ../prints/<name>/index.html --at 0 --out ../out/<name>.png
The starter is blank paper with the print and craft kits and no animation loop. Draw in its art
block. A single picture uses duration: 1 and a seek(t) that draws the same picture; a series
uses integer print indices (floor(t), as prints/workings does), verified at every index and
exported one PNG per index. No MP4, score or artificial motion.
For another native size, change canvas, paper, plate dimensions, projection and screen pitch
before baking, and audit square assumptions like W,W. Larger prints must rerasterize geometry
and screens at that size; CSS/DPR enlargement and PNG upscaling do not make a higher-resolution
original. Press-ready separations need a separate print-production brief.
- 2160 square (
OUT = 2160,K = OUT / W) served both Cabinet and Sceneries.PITCH = 4.6 * K(9.2 device px) is the same picture enlarged, right for landscape masses; small dense subjects took a finer3.4 * K(6.8 px). Registration scales withK. - Fine engraved line needs coverage 1 to print solid; the kit's threshold tile leaves the interstices open even at full coverage (right for film tone), which breaks thin line into dots.
- A feature under a few screen cells (florets, scales, birds) is modelled by shape and solid
overprint, not dot size. A small fixed dither on mid-tones breaks remaining dot-size steps in a
gentle ramp; leave paper and solids untouched. At 2160, centre the screen's dots on pixels
(
docs/drawing.md, tone that prints) or ramps contour into rings. - Add debug views by query string before polishing: construction lines alone compared against the reference diagram (it caught over-curved septa in a shell section), and one ink at a time for plate isolation.
Inspect the whole picture at viewing size, a value/silhouette view, and 1:1 crops of difficult
anatomy, attachments and screen. Fix perspective, tangencies, missing support or an unclear
subject before polishing grain; docs/quality-bar.md lists the review cases. still.mjs exports
the canvas backing store and checks repeatability; it does not judge the art. Confirm dimensions
and open the PNG.
Keep a short PRINT.md beside substantial work: request, the series principle, references per
print, viewpoint, geometry and light decisions, native size and pitch, inspected output and each
print's remaining weaknesses. Keep the previous version when revising.
Signals
- GitHub stars
- 247
- Forks
- 29
- Last commit
- Sep 2026
Advanced
- Item type
- skill
- Key
riso-still- Source
- github.com/sevenevesai/riso-windowseat