achimala/dream-loop

What it solves

Dream-loop is an agent skill designed to create games, apps, or visual scenes with high-quality aesthetics. It solves the problem of AI agents struggling to maintain a consistent visual target when building complex visual projects from scratch.

How it works

The system creates a closed-loop feedback cycle to guide the AI's development process:

1. Dreaming: The AI generates a high-quality target screenshot using image generation tools.
2. Building: The AI develops the project with that target image as a visual goal.
3. Critiquing: A separate AI critic compares the live screenshot of the current build to the target image and provides feedback.
4. Iterating: The AI loops back to the building phase until the critic is satisfied.
5. Refining: Optionally, the AI can generate a new, even better target image based on the current state of the project.

Who it’s for

Developers and AI agent users who want to create visually impressive 3D scenes, browser-based graphics demos (such as those using Three.js), or apps with a specific art style.

Highlights

• Closed-loop visual feedback: Uses a target image to guide the AI's building process.
• Vision-driven development: Leverages vision input and image generation to ensure visual fidelity.
• Blender integration: Supports custom 3D modeling via Blender MCP or scripting interfaces for improved results.
• Iterative refinement: Allows for continuous improvement of both the final product and the target visual goal.