| name | image-to-pindou |
| description | Convert photos or illustrations into recognizable, printable MARD fuse-bead patterns. Use for AI-assisted pixel-art redrawing, local conversion, feature-preserving pixelation, background removal, bead color matching, editable grids, and purchase lists. |
Image to Pindou
Run the bundled CLI from this directory. Prefer an AI-redrawn pixel source when resemblance matters; use the local path when the user requests offline or deterministic conversion.
Workflow
- Identify the subject, output directory, target size, background treatment, color cap, and 2–6 identity-defining details. Finish when every detail that would make the result recognizable is listed.
- Choose a source path:
- Best quality: redraw the image as crisp square-pixel art with the built-in image-editing tool, then run the CLI with
--ai off; or use --ai openai|gemini when its API key is already configured.
- Fully local: run the original image with
--ai off.
- Use
cartoon for flat illustrations, bead for general buildability, faithful for photographic shading, or all only for local comparisons. Pass each critical detail with a separate --preserve when using CLI AI.
- Inspect
ai-draft.png when present, then sampled-preview.png, pixel-preview.png, and pattern.png. Finish only when the subject is recognizable at preview size, critical details survive, the background is correct, and no isolated wrong-color speckles remain.
- If a check fails, change one cause and rerun: revise the redraw for missing identity, increase size for lost geometry, lower the color cap for noisy shading, or switch
bead/cartoon for weak contours. Reinspect all four stages after each rerun.
- Verify that the
.pindou.json project, bom.csv, and rendered pattern agree on dimensions, colors, and bead count before reporting.
For built-in image editing, request deliberate pixel art at the target grid size, strong one-cell outlines, flat coherent color regions, a plain white outer background when removing it, and preservation of the listed details. Request no visible grid, bead circles, labels, text, gradients, blur, or added objects. The CLI enforces the exact grid locally.
CLI
npm install
node scripts/generate.mjs photo.jpg \
--style cartoon --size 50 --max-colors 18 --background remove \
--preserve "asymmetrical eyes" --preserve "hat shape" \
--out outputs/cat
Run node scripts/generate.mjs --help for limits and optional flags. Never silently make repeated paid AI calls; show the draft or report why a rerun is needed first.
Read references/principles.md when explaining or changing the conversion method.
Treat the .pindou.json file as the editable source of truth and mention that it can be imported into Pin Dou for further editing. pattern.svg is the scalable construction drawing; bom.csv is the purchase list. When --style all is used, each style receives its own output directory.
Report the grid dimensions, color count, bead count, output path, and any background-removal refusal.
