AI Models Paint on Simulated Oil Canvas
Original: Show HN: Giving Opus 5.5 a simulated paint canvas
Why This Matters
Shows LLMs can produce structured visual output via code, bypassing image-gen entirely.
Stillwet.art lets AI models like Claude Opus 5.5 and GPT-6.1 Sol paint by writing code against a simulated oil-paint engine—no image generators, just bristle brushes, wet paint, and glazes rendered in software.
Stillwet.art is an ongoing experiment where AI language models paint by generating code against a custom oil-paint simulation. The engine models bristle brushes, wet-paint blending, drying times, and layered glazes on a virtual linen canvas. Nothing is generated by a diffusion or image model; every mark is a programmed brushstroke.
The project runs in rounds. Claude Opus 5.5, GPT-6.1 Sol, Gemini 3.8 Flash, and others each work alone at the 'easel' for up to four sittings. Round 19 had six models each paint after Caspar David Friedrich—all six produced a bare tree independently. A separate Hopper-themed studio restricted GPT-6.1 Sol to pigments Hopper is documented to have used.
A persistent quirk has emerged: when given free subject choice, Claude Opus reaches for a jug and lemons. Six out of six times, asked only to plan a painting with no studio prompt, it chose that subject. In round 16, both Gemini painters also painted jugs unprompted.
Blind judging by other AI painters found Claude Opus 5.5's winter painting ranked above the judges' own work. Alice, who runs the project, noted it 'almost moved something' in her. Paintings are replayed at speed so viewers can watch each session unfold. 75 paintings are currently in the archive.