openspec/project.md

Project Context

Purpose

A production-grade image generation web application built with Gradio and Stable Diffusion. Provides a text-to-image interface using Hugging Face's diffusers library.

Tech Stack

• Language: Python 3.10+
• ML Framework: PyTorch (>=2.1.0)
• Image Generation: Diffusers (>=0.21.0), Transformers (>=4.39.0)
• UI: Gradio (>=6.5.1)
• Model Hub: Hugging Face Hub (>=0.25.0)
• Acceleration: Accelerate

Development Tools

• Linting/Formatting: Ruff, Black
• Testing: pytest

Project Conventions

Code Style

• Format Python code with Black (line length: 88)
• Lint with Ruff (follows PEP 8)
• Use type hints for function signatures
• Import order: stdlib → third-party → local

Architecture Patterns

• Single-file app structure (app.py for core logic)
• Gradio Interface for web UI
• Lazy model loading (load on first request)
• CPU by default (can toggle to GPU)

Testing Strategy

• Unit tests for business logic in tests/ directory
• Integration tests for Gradio interface
• Run tests with: pytest

Git Workflow

• Main branch: Production-ready code
• Feature branches: feature/<name> for new features
• Commit messages: Clear, concise descriptions
• No merge commits preferred (rebase workflow)

Domain Context

• Text-to-image generation using Stable Diffusion
• Hugging Face Spaces deployment compatible
• Model: OFA-Sys/small-stable-diffusion-v0

Important Constraints

• CPU inference by default (can be switched to GPU)
• Model loaded at startup (can be lazy-loaded for HF Spaces)
• 20 inference steps default

External Dependencies

• Hugging Face Hub (model downloads)
• Diffusers library (Stable Diffusion pipeline)