WARNING: This is an experimental project, use at your own risk.
AI-PP3 is an advanced command-line tool that combines multimodal AI analysis with RawTherapee's processing engine to create optimized development profiles (.pp3 files) for RAW photography. Designed for both automated workflows and creative exploration.
See example documentation for practical usage scenarios.
- Key Features
- Compatibility
- Installation
- AI Configuration
- Basic Usage
- Advanced Features
- CLI Options
- Roadmap
- FAQ
- License
- 🖼️ AI-driven analysis of RAW files (DNG/NEF/CR2/ARW)
- ⚡ Batch PP3 creation with consistent processing parameters
- 📝 Customizable development settings through natural language prompts
- 🔀 Multi-model support (OpenAI, Anthropic, Google, Local)
- 🎚️ Fine-grained control over PP3 sections (Exposure, Color, Detail)
- 🔍 Interactive preview generation with quality controls
- RAW Files: All RawTherapee-supported formats including:
- Common: CR2/CR3, NEF, ARW, RAF, DNG
- Specialized: IIQ, PEF, RW2, ORF
- Output Formats:
- JPEG (8-bit)
- TIFF (8/16-bit with ZLIB/NONE compression)
- PNG (8/16-bit)
- Node.js ≥18
- RawTherapee ≥5.8 (CLI required)
- API Key for cloud AI or local GPU for self-hosted models
# Global installation npm install -g ai-pp3 # Verify installation ai-pp3 --version# Environment variables (.env file) OPENAI_API_KEY=your_key # Default provider ANTHROPIC_API_KEY=your_key # Claude models GOOGLE_GENERATIVE_AI_API_KEY=your_key # Gemini# Cloud models ai-pp3 input.dng --provider anthropic --model claude-3-opus-20240229 # Local models ai-pp3 input.dng --provider openai-compatible --model llama3:8b-instruct-q5_K_M# Basic processing with defaults ai-pp3 input.dng -o output.jpg # PP3-only mode with custom prompt ai-pp3 input.dng --pp3-only -p "You are a RawTherapee processing profile (pp3) optimization ARTIST. Your mission is to creatively transform the attached pp3 file with artistic vision. A JPEG preview is provided - use it as a starting point for your artistic interpretation. ARTISTIC MANDATE: - Prioritize artistic expression and unique visual style - Create a distinctive mood or atmosphere in the image - Experiment with color relationships and tonal contrasts Key Rules: 1. Only modify existing parameter values 2. Keep original section order and parameter order 3. Make bold, creative enhancements 4. Only include parameters you want to change Common Parameter Value Ranges: %%PARAMETER_RANGES%% Output Format: ANALYSIS: - Artistic opportunities and potential visual directions PLAN: - Creative vision and mood you're aiming to create EXECUTION: \`\`\` [Exposure] Clip=0.15 Compensation=-0.7 \`\`\` Example with another section: \`\`\` [ColorToning] Redlow=20 \`\`\` [Additional changes following these rules] - Only include parameters you want to change - Never change section headers - Include only the sections that need changes Current pp3 to transform: " # Multi-section processing ai-pp3 input.dng --sections Exposure,ColorToning # Base profile refinement ai-pp3 input.dng --base existing.pp3 --preview-quality 85# Parallel processing (GNU Parallel) ls *.DNG | parallel -j8 ai-pp3 {} -o {.}.jpg # TIFF output with compression find . -name '*.NEF' -exec ai-pp3 {} --tiff --compression z \;# Multi-model comparison ai-pp3 input.dng \ --provider openai --model gpt-4-vision-preview \ --base neutral.pp3 --keep-previewThe --generations option enables AI-powered comparison of multiple PP3 profiles:
# Generate 3 different approaches and let AI select the best ai-pp3 input.dng --generations 3 # Advanced multi-generation with custom settings ai-pp3 input.dng \ --generations 5 \ --preset creative \ --provider anthropic \ --model claude-3-opus-20240229 \ --verboseHow it works:
- Generates multiple PP3 profiles using the same base settings
- Processes the RAW image with each profile
- Uses AI vision models to evaluate and compare all results
- Automatically selects the best processed image
- Saves the winning PP3 profile and final image
- Cleans up intermediate files (unless
--verboseis used)
-o, --output <path>: Output file path (default:input.pp3orinput_processed.[format])--pp3-only: Generate PP3 file without processing image-p, --prompt <text>: Natural language prompt for AI analysis--preset <name>: Preset style to use (aggressive,creative,balanced,technical)
--provider <name>: AI service provider (openai,anthropic,google,openrouter,openai-compatible)--model <name>: Model name (default:gpt-4-vision-preview)
--tiff: Export as TIFF format--png: Export as PNG format--compression <type>: TIFF compression (zornone)--bit-depth <n>: Bit depth (8or16)--quality <n>: JPEG quality (1-100)
--base <path>: Base PP3 file for incremental improvements--sections <list>: Comma-separated PP3 sections to process (e.g.Exposure,ColorToning)--preview-quality <n>: Preview quality (1-100, JPEG only, default: 85)--preview-format <format>: Preview image format (jpegorpng, default:jpeg)--max-retries <n>: Maximum number of retries for AI API calls (default: 2)--generations <n>: Generate multiple PP3 profiles and use AI to select the best one (1-10, default: 1)-v, --verbose: Enable detailed processing logs-k, --keep-preview: Retain preview file after processing
# Basic conversion with quality control ai-pp3 input.dng --tiff --compression z --bit-depth 16 # Multi-model comparison ai-pp3 input.dng --provider anthropic --model claude-3-opus-20240229 # Section-limited processing ai-pp3 input.dng --sections Exposure,Detail --pp3-only # Use PNG preview for AI analysis (better for some models) ai-pp3 input.dng --preview-format png --keep-preview # Multi-generation mode: Generate 3 different PP3 profiles and let AI pick the best ai-pp3 input.dng --generations 3 --verbose # Multi-generation with custom settings ai-pp3 input.dng --generations 5 --preset creative --provider anthropic --model claude-3-opus-20240229- ART (.arp) profile compatibility
For detailed questions and answers, see our FAQ documentation.
![@google-labs-jules[bot]](https://avatars.githubusercontent.com/in/842251?s=64&v=4){kind=small}
