Built inside ANKLE OS (selected for the Claude for Startups cohort by Anthropic). Instead of generating pixels from diffusion noise, ASC synthesizes new studies directly from 42,638 real vector strokes extracted from 111 of my physical ink and Procreate drawings — guided by a live CPU reinforcement-learning model (Epoch v187 · 1,389 human preference decisions) with SHA-256 provenance for every mark.
verify_study_on_disk (checks PNG dimensions, >94% open paper ratio, stroke_id membership in master_stroke_bank.json, and donor SHA-256 hashes; does not judge aesthetic merit).
Every mark in ASC originates from these hand-drawn ink sketchbook scans and Procreate works by Alex Headlam. Click any source drawing to open its modal and see the studies built from its strokes.
Side-by-side multimodal grounding: Source Input Artifact (unconstrained aspect ratio) × 409K SURF Corpus → Compiled 768×987 .dada Object with L0–L3 Semantic Spine and lossless PNG iTXt provenance.