GenClaw: Code-Driven Agentic Image Generation

Researchers introduced GenClaw, a staged image generation framework for agentic, code-driven visual creation.

An LLM gathers context and writes executable visual sketches in code, using formats like SVG, HTML, and Three.js.

A generative image model then turns those sketches into textured, photorealistic outputs.

The approach aims to make image generation more controllable, interpretable, and easier to direct.