Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
GenClaw: Code-Driven Agentic Image Generation
| Date | Stars |
|---|---|
| 2026-07-31 | 298 |
| 2026-08-03 | 298 |
| 2026-08-06 | 298 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# GenClaw: Code-Driven Agentic Image Generation [](https://huggingface.co/papers/2605.30248) [](https://arxiv.org/abs/2605.30248) [](https://github.com/yejy53/GenClaw) GenClaw explores **code-driven agentic image generation**: instead of only rewriting prompts, an image generation agent uses code as a controllable visual canvas before calling image generation models for final rendering. The core idea is simple: **think, sketch with code, then render**. <p align="center"> <img src="assets/teaser.jpg" width="95%"> </p> ## News - **[2026-05-30]** 🎉 GenClaw released — technical report and project page. - **[2026-06-29]** 🚀 Released a runnable agent implementation, with updated code for **complex scene composition**, **text rendering**, and **world-knowledge-grounded generation**. ## Highlights 🎨 **Code as a Visual Brush.** The agent creates by writing executable visual sketches—SVG, HTML/CSS, Python, lightweight 3D code—turning object count, spatial layout, and text rendering into executable, verifiable, debuggable programs. Image synthesis shifts from implicit diffusion sampling to an explicit, reasoning-friendly process. ✋ **Draw as a Human Artist.** We mirror the human creative loop—conceptualize → sketch → coloring → refine—and make every stage transparent: ideation, reference retrieval, drafting, and incremental rendering are all surfaced as inspectable, editable, revertible artifacts. Generation becomes an iterative collaboration rather than one-shot black-box inference. 🔌 **Agent Harness for Image Generation.** We plug an LLM agent's proven planning, tool-use, and reflection abilities directly into image synthesis, exploring an agent harness for image generation—so that creating images becomes a first-class capability inside the agent's toolbox, not an isolated standalone model. ## Showcase <p align="center"> <img src="assets/showcases.jpg" width="95%"> </p> ## Visual Examples ### Complex Scene Composition <p align="center"> <img src="assets/complex_scene.jpg" width="95%"> </p> ### Text Rendering and Poster Design <p align="center"> <img src="assets/text_rendering.jpg" width="95%"> </p> ### Physical Reasoning <p align="center"> <img src="assets/physical_reasoning.jpg" width="95%"> </p> ### Knowledge-Grounded Generation <p align="center"> <img src="assets/mind_bench.jpg" width="95%"> </p> ## Getting Started A runnable agent implementation lives in this repo. A planner LLM works in a tool-use loop: before each task a `perception` sub-LLM writes descriptive "painter notes" (intent + planning hints — these are hints, not routing decisions), then the planner reads the request, the notes, and the tool cards, opens a `todo_write` plan, and composes a tool pipeline from a set of canonical chains. **Tools**: `t2i` / `i2i` (text/image-to-image), `code_scene_draft` (SVG layout draft), `code_text_draft` (verbatim long-text rendering), `search` (Tavily web/ image search), `reason` (multimodal reasoning), `format_prompt`, `vlm_review`, plus `todo_write` / `tool_search` planning helpers. **Example canonical chains** (from `todo_write`'s card; `format_prompt` can be inserted as needed): spatial/counting `code_scene_draft → i2i`; grounded entity `search → format_prompt → i2i`; scientific derivation `reason → t2i`; world-knowledge long text `search → code_text_draft`. ### 1. Install ```bash git clone https://github.com/yejy53/GenClaw.git cd GenClaw ``` Using `venv` + pip: ```bash python -m venv .venv && source .venv/bin/activate pip install -e . ``` Or using conda: ```bash conda env create -f environment.yml conda activate cc-genclaw ``` Then install the headless browser used by the code-draft renderers: ```bash playwright install chromium # On Linux you may also need: playwright install-d
Excerpt of 6,861 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:5957a9945e9b5771, llm:Repository name and description: "GenClaw: Code-Driven Agentic Image Generation" (Python). Suggests an AI project for image generation using agentic/code-driven approach.
matched fp:5957a9945e9b5771, llm:Repository name and description: "GenClaw: Code-Driven Agentic Image Generation" (Python). Suggests an AI project for image generation using agentic/code-driven approach.
matched fp:5957a9945e9b5771, llm:Repository name and description: "GenClaw: Code-Driven Agentic Image Generation" (Python). Suggests an AI project for image generation using agentic/code-driven approach.