Skip to content

GEMINI.md

Latest commit

 

History

History
80 lines (61 loc) · 3.5 KB

GEMINI.md

File metadata and controls

80 lines (61 loc) · 3.5 KB

Gemini Project Context: Draw Things Skill

Project Overview

  • Purpose: Local AI image generation using the Draw Things gRPC backend.
  • Architecture: A Client-Server model.
    • Control Layer (Inside Skill): Pure Node.js implementation using @grpc/grpc-js and pngjs.
    • Infrastructure Layer (External): The Draw Things macOS binary and ~10GB of model weights, managed outside this repository.
  • Core Technologies:
    • Language: Node.js.
    • Communication: gRPC (imageService.proto).
    • Configuration: FlatBuffers binary.
    • Backend: Draw Things (External macOS binary).

External Infrastructure

This skill controls an external server. Do not move or copy these heavy assets into the skill directory.

  • Server Binary: [Path to your Draw Things gRPC server binary, e.g., /path/to/gRPCServerCLI-macOS] (Download from: https://github.com/drawthingsai/draw-things-community/releases)
  • Models Directory: [Path to your Draw Things models, e.g., ~/Library/Containers/com.liuliu.draw-things/Data/Documents/Models]
  • Default Address: localhost:7859 (No TLS, no compression).

Current Status: Pure Node.js Implementation

✅ Pure Node.js: High-performance gRPC implementation using @grpc/grpc-js and pngjs. No external dependencies other than the Draw Things server. ⚠️ Note: Ensure the server binary version is compatible with the GenerateImage method. Refer to IMPLEMENTATION.md for troubleshooting.

Configuration

The Draw Things skill uses a centralized configuration system. Settings are stored in ~/.drawthings-skill/config.json.

1. Initial Setup

Run the setup command to configure your local paths:

# Using global command (if linked)
dt-skill config --server-path "/path/to/gRPCServerCLI-macOS" --models-path "~/Library/Containers/com.liuliu.draw-things/Data/Documents/Models"

# Or using npm
npm run dt -- config --server-path "/path/to/..."

2. View Configuration

dt-skill config --show

3. Server Control

dt-skill server start
dt-skill server status
dt-skill server stop

4. Generate Images

dt-skill gen --prompt "a serene mountain lake at sunrise" --output output.png

Key Commands

Command Description
npm run start-server Starts the Draw Things gRPC server on localhost:7859.
npm run generate -- [options] Generates an image using scripts/generate.js.
npm run health Checks if the gRPC server is responsive.
npm run models Placeholder (model listing depends on local filesystem access).

Project Structure

  • scripts/generate.js: Core logic for gRPC communication, FlatBuffer building, and image saving.
  • scripts/imageService.proto: gRPC service and message definitions.
  • references/: Detailed documentation on gRPC protocols and sampler algorithms.
  • SKILL.md: User-facing documentation and quick start guide.
  • IMPLEMENTATION.md: Technical implementation details (FlatBuffer slots, tensor decoding).

Development Conventions

  • Pure Node.js: All gRPC logic, binary serialization (FlatBuffers), and image decoding are handled in Node.js for maximum performance and portability.
  • Error Handling: Detailed status codes and troubleshooting tables are maintained in the documentation.
  • Model Management: Default models are optimized for speed (e.g., z_image_turbo).
  • Sampler Selection: Recommended samplers include unicpc-trailing and euler-a-trailing.