GenClaw: Code-Driven Agentic Image Generation

Junyan Ye, Jun He, Zilong Huang, Dongzhi Jiang, Xuan Yang, Rui Chen, Weijia Li

arXiv:2605.30248 · 2026-05-29 공개 · arXiv · PDF

llm text-to-image three-js controllable-generation html visual-synthesis sketch-to-image code-driven

Abstract

Image generation models have evolved from text-conditioned pixel synthesis toward multimodal agents endowed with visual comprehension and tool invocation capabilities. Yet, existing agents remain at the mercy of underlying black-box image models. Their workflow is trapped in a repetitive cycle of prompt rewriting for generation refinement, leaving them with no mechanism to directly manipulate the canvas. In essence, the potential of LLMs to serve as a genuine "brush" for precise visual construction remains largely untapped. In this paper, we propose GenClaw, a code-driven agentic image generation paradigm that empowers the agent to create like a human artist: first conceptualizing, then sketching, and finally coloring. Specifically, the agent first constructs the conceptual knowledge and context through search and reasoning. It then utilizes code (e.g., SVG, HTML, Three.js) to render executable visual sketches. Finally, it employs an image generation model to supplement textures, materials, and photorealism. In this workflow, code serves as a controllable intermediate canvas bridging linguistic reasoning and pixel synthesis, seamlessly integrating programmatic logic with the visual expressiveness of generative models. By transforming image generation from a black-box paradigm into a staged process akin to authentic human creation, GenClaw offers a step toward for highly controllable and interpretable visual generation systems.

한국어 요약

📋 한 줄 요약

**[Agentic Image Generation / Code 기반]** GenClaw가 이미지 생성을 conceptualize→sketch→colorize 3 단계로 변환, agent가 SVG·HTML·Three.js로 executable 시각 sketch 렌더링 후 image model로 texture·material·photorealism 보완 — code를 controllable intermediate canvas로 사용.

🎯 핵심 기여도

💡 핵심 아이디어

Agentic 이미지 생성의 controllability·interpretability 한계는 prompt rewriting cycle에 갇혀 있어서이며, 인간 artist의 conceptualize→sketch→colorize 프로세스를 모방해 code(SVG·HTML·Three.js)를 controllable intermediate canvas로 활용하면 언어 추론과 픽셀 합성을 명시적으로 bridge할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Image gen agent의 black-box·prompt cycle 한계 해소 패러다임, code intermediate canvas의 controllability·interpretability 결합, 인간 artist 워크플로 모방의 직관적 paradigm. **한계**: SVG·HTML·Three.js code 생성 품질이 underlying LLM에 의존, 매우 사진적 자연 장면에서 sketch 단계 충실성 한계, 정량 비교·SOTA 위치는 abstract에서 불명확.

🚀 실용적 활용