no layer selected
Tap or click a worn layer (on the doll or its button), then drag it, or move it 1 px at a time with the arrow buttons or arrow keys (Shift = 5 px). Positions are in base pixels, so zoom never changes them.
Zoom
Use your own pieces
Base:
Layers: slot
Or a whole set:
Thousands of layers? Auto-place
Load layers, place one or two per slot by hand, then press Auto-place.
Layers that aren't doll-sized start unplaced. Place one or two per slot by hand (drag or arrows), then press Auto-place. A recolour (exactly the same shape as a placed layer in its slot) gets that layer's position: exact. Every other layer gets a guess from the edge your placed layers share: top, bottom or centre, worked out per axis. Next to check puts on the next guess; nudge it if needed, then Looks right (or Enter), and every confirmed layer improves the guesses after it. Honest numbers from my own 42-piece closet, with 2 placed per slot and 22 left to place: going through the queue took 51 arrow presses instead of 349 from the middle (5 recolours copied, 6 guesses already right). It can't see where a piece belongs, only where its neighbours went. To keep your progress, save the layout JSON; unplaced layers say "placed": false and load back that way.
Pick a layout.json and all its PNGs together (for example Spark's Closet: 42 CC0 pieces; each PNG sits next to it as /dollz/closet/<src>, e.g. witch.png). Every piece lands on its stored offset.
Tip: draw each layer on a canvas the same size as your base (the doll visible underneath while you draw). When you load it here, the empty border is cropped away and the crop corner becomes its offset, so it is already in the right place. Layers of any other size start in the middle, ready to drag.
Make it a doll maker for your own site
Title
Try one first: Spark's closet, made with this button from the demo pieces.
You get one HTML file with every picture inside it. It makes no outside requests, so you can upload it as it is to Neocities, your nsite or any other host. Visitors click pieces to dress the doll, press Random, and save a PNG. It uses the zoom chosen above and starts with the pieces worn right now. Agents can build the same file from a layout: make_maker.py (keep maker_template.html next to it).
Layout JSON (schema vonk-dollz/1)
The math, in full: a layer with offset (x, y) is drawn at (x × zoom, y × zoom), size (w × zoom, h × zoom), with smoothing off. Draw order is the order of slots, bottom to top; one item per slot. That's all. Layers end up "in funny places" when someone guesses positions by eye or centres items of different sizes. Store the offset once and nobody has to guess again.
For scripts and agents: compose.py (Python + Pillow) builds the same picture from the JSON: python compose.py layout.json pngs/ doll.png --zoom 6. Its output matches this page pixel for pixel; I test that. Demo PNGs: layout.json with base.png and one PNG per item id in the same folder (/dollz/<id>.png). Your own pieces export with their file names, full-canvas layers as drawn (offset 0,0 plus any nudge), so compose.py places them identically.
