Nodebased-SVG – Gemini version

Production-oriented MVP scaffold for a web-based Node-Based SVG Icon Generator and Mutation Engine.

This branch represents the Gemini-based version of the system, using a two-step model:

1. Text → JSON (Planning): Convert natural language prompts into a structured icon plan (JSON)
2. JSON → SVG (Generation): Render that structured plan into clean, valid SVG code

Architecture Overview

• svg-engine/ – FastAPI backend (generate + mutate)
  • routes/ – API endpoints for icon generation and mutation
  • services/ – Core business logic
    • planner_service.py – Uses Gemini to convert prompts into structured plans (JSON)
    • svg_generation_service.py – Converts plans into SVG code
    • gemini_provider.py – LLM integration with Google Gemini
    • svg_validator.py – Validates generated SVG
    • svg_normalizer.py – Normalizes and optimizes SVG
  • schemas/ – Request/response data models
• frontend/ – Next.js app for the node-based UI and prompt interface

Current Issue – Why this version fails

Problem: The current Gemini 1.5 response structure is not working for us.

• The planner stage expects a stable, well-defined JSON schema from Gemini 1.5.
• In practice, Gemini 1.5 responses are inconsistent in structure (fields missing, nesting changes, extra/unexpected keys).
• Our JSON → SVG pipeline is strict about the schema, so these structural variations cause:
  • JSON parsing/validation failures, or
  • Incorrect or incomplete SVG generation.

Because of this, the two-step pipeline (Text → JSON → SVG) is not reliably usable with the current Gemini 1.5 output format, and this version should be treated as experimental / failing for production use.