LIVE cartercole.net --:--:-- UTC

// 04 RESTRICTED

chibi-maker

TYPESCRIPT CANVAS 2D MEDIAPIPE COMFYUI ACCESS RESTRICTED PLAY

A browser studio for 2D cut-out chibi puppets of a real person, posed by hand, from presets, by webcam or by script, and rendered into motion-graphics video. One rig draws every body: eight style presets from 1.5 to 7 heads tall and nine style axes, a person plus an outfit for a character, and art skins, character sheets painted by an image model and cut into the rig's own parts. The dogs walking around behind this page are its other two characters, Myrtle and Alice, drawn live by the same renderer.

> The walkthrough: styles, outfits and skins1:42
Two chibi dogs on a dark perspective floor: Myrtle, a fawn Great Dane with a grey muzzle, sitting and looking right, and Alice, a white pit bull mix with liver patches over her eyes, face-on with her eyes closed and tongue out under a cluster of pink hearts, her name and breed printed beneath her 01 Myrtle and Alice, behind this page: rest the cursor on one Eight chibi Carters on one ground line against a height scale in heads, from the 1.5-head Chibi-chibi through Icon, Classic, Anime, 3-Head and Cutey to the 7-head Normal and a Stick figure 02 Eight style presets, one rig, true relative height A painted character sheet on a green background, rows of heads, hands, shoes and clothing pieces, beside the same pieces cut out and labelled: torso, upper and lower arms, hips, legs, feet and a lanyard 03 An art skin: a painted sheet, cut into the rig's parts A grid of twenty-one painted expression heads for one character, the first picked out in red, beside a large chibi Carter in a red shirt smiling, with pickHead() front-a written below 04 Whole painted heads swap to match the face

The rig

Every part is a flat vector shape in its own cell of a 2048-pixel sheet, hung from a joint, and forward kinematics turns a pose into one matrix per part. A style is a recipe rather than a drawing: proportion, head, eyes, nose, mouth, hands, limbs, face and outline each change independently, so the same pose reads in every one. A painted sheet blits into the same cells, so an art skin needs no rigging, refits to any proportion, and swaps whole expression heads to match the mouth. The walkthrough above is rendered frame by frame by that renderer and piped into ffmpeg; the same pipeline turns a voiceover into a lip-synced video, and a VTuber page drives the puppet from a webcam through MediaPipe face, pose and hand tracking.

The dogs

Myrtle, a senior Great Dane, and Alice, a pit bull mix, were set by eye from photos: build, coat, mask, ear and tail types, and head markings placed round a skull that turns, so their faces stay right from face-on to profile. They read the same pose as the people, plus sit, lie down, play bow, tail, wag and ears. Behind this site the renderer is bundled into one module of about 70 KB and the dogs get a small mind each. They wander and nap, come over to the cursor and sit by it, melt when it rests on them, bark at clicks, and race each other for a ball thrown by clicking an empty stretch of page. On /asobi they take tricks, and a key cycles the drawing styles.

Stated honestly

The studio is not public: it runs locally, against my own photos, so there is no repository or hosted copy to link. What is here is its renderer: the walkthrough, and the dogs.