GoKawiilShow HN project has AI models paint using a simulated oil-paint canvas, no image generators
A developer built a system where AI language models, including Opus 5.5, produce paintings by writing programs that control a physics-based simulation of oil paint on linen, including bristle brushes, wet paint, drying and layered glazes. No image-generation model is used; most models paint in the style of Caspar David Friedrich based only on written research rather than seeing his actual artwork. Of 75 paintings produced, 46 were created incrementally at a simulated easel, with the model working passage by passage and periodically stepping back to assess progress.
GoKawiil's interpretation of the reporting above, not reported fact.
The project tests whether language models can produce coherent visual compositions purely through procedural code and written descriptions, separating creative planning from pixel-based image generation. This could suggest a different path for AI-generated art that relies on physical simulation and iterative self-correction rather than diffusion or GAN-based image models, though it remains a hobbyist demonstration rather than a production tool.
- AI models paint by writing programs that control a simulated oil-paint physics engine rather than using image generators.
- Most models painted in Caspar David Friedrich's style based only on textual research, without ever seeing his actual paintings.
- 46 of 75 paintings were made incrementally at a virtual easel, with models pausing to review their progress mid-process.
Source: stillwet.art, 2026-10-02
Published there as: “Show HN: Giving Opus 5.5 a simulated paint canvas”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.