Muse - Master Mobile Photography

Inspiration

Ever scrolled through Instagram and thought, "How did they create that look?" Generic tutorials are either too broad or device-specific. We built Muse to bridge inspiration and execution—turning any photo into a step-by-step tutorial tailored to your device and editing app.

What it does

Muse is an AI-powered photography education platform with two core features:

🎓 Learn Mode: Upload any image/video. Muse uses Chrome's on-device AI (Gemini Nano) to analyze style and generate personalized tutorials. Mention "iPhone" or "Samsung" and instructions adapt automatically. Get specific values like "contrast +20" instead of vague tips.
🎨 Style Transfer: Upload a style reference and your photo. Muse extracts color palette, lighting, and editing parameters, then automatically applies them using Canvas API—with instant before/after comparison.

Hybrid AI Approach: Learn Mode uses on-device AI for privacy (photos never leave your device), while Style Transfer leverages cloud AI for complex analysis. Best of both worlds.

How we built it

Tech Stack: React 19, Vite, Tailwind CSS, Chrome LanguageModel API, Canvas API, OpenAI/Gemini/Anthropic

Key Components:

Custom React hooks (useAnalysis, useStyleTransfer) for modular AI orchestration
Intelligent prompt engineering that infers user intent (shooting vs. editing) and adapts responses
Canvas-based image processing pipeline for style transfer
Graceful fallback chain: on-device → cloud → demo mode

Challenge: Chrome's AI APIs are evolving rapidly. We built compatibility layers supporting both LanguageModel and window.ai.assistant with automatic detection.

Challenges we ran into

API Evolution: Kept up with Chrome's transition from window.ai.assistant to LanguageModel, requiring different prompt structures
Multimodal Limitations: On-device AI has uncertain image support, so we designed text-based descriptions for Learn Mode
JSON Schema Enforcement: Models sometimes returned malformed JSON—built robust parsing and error handling
Canvas Performance: Optimized large image processing with resizing and efficient pixel manipulation
Prompt Engineering: Getting AI to distinguish between capture guidance (camera settings) and editing guidance (numeric values)

Accomplishments that we're proud of

✅ First project to successfully integrate both Chrome AI APIs (LanguageModel + window.ai.assistant) with fallbacks
✅ Built a true hybrid AI system—privacy for learning, power for style transfer
✅ Created adaptive, context-aware tutorials that eliminate dropdown menus
✅ Production-quality UI with minimalist design and smooth interactions
✅ Delivered specific, actionable guidance with precise numeric values
✅ Complete style transfer pipeline using client-side Canvas API

What we learned

On-device AI is powerful but limited—excellent for privacy and speed, but complex processing benefits from cloud
Prompt engineering is critical—small changes dramatically affect output quality
Hybrid architectures unlock possibilities—optimize for privacy, speed, and capability simultaneously
Specificity matters—users want actionable guidance, not general tips
Building with cutting-edge APIs requires resilience—expect changes and build abstraction layers

What's next for Muse

Short-term: Video frame analysis, mobile app version, multi-language support
Medium-term: Community features, advanced video editing guidance, expanded software support
Long-term: Global style database, skill certification, personalized AI coach

Vision: Make professional photography education accessible to everyone, turning every photo into a learning opportunity.