Frame AI

October 2025 FastAPI · Python · Gemini Flash · nano-banana · SQLite
Live Demo Source Code

Frame AI is an AI-powered photography assistant that bridges the gap between taking photos and knowing how to improve them. It analyzes images using Gemini 2.5 Flash to evaluate photography principles, then generates three distinct enhanced variations using Gemini 2.5 Flash Image (nano-banana).

Demo

Frame AI in action: analyzing a photo and generating enhanced variations

Results

Original photo
Original
Enhanced variation 1
Variation 1
Enhanced variation 2
Variation 2
Enhanced variation 3
Variation 3

Key Features

  • Intelligent Analysis: Detailed compositional feedback with numerical scores (exposure, composition, lighting, overall)
  • Three Creative Variations: Technical perfection, atmospheric reinterpretation, and conceptual narrative
  • Smart Caching: Content-based hashing (SHA-256) for deduplication and cost optimization

Technical Highlights

  • FastAPI backend with SQLite for persistence and Server-Sent Events for streaming analysis
  • Content-based hashing (SHA-256) enables reliable deduplication while remaining privacy-friendly
  • Separate LLM call generates editing prompts from analysis before passing to nano-banana—produces more precise edits than passing raw analysis directly
  • Parallel image generation using asyncio.gather for three simultaneous variations

Read the full blog post →

← Back to Projects