AI Photo Editing Agent
A full-stack tutorial project that builds an AI agent-driven photo editor: a user types a natural-language instruction, the agent plans a sequence of edits, and 21 image-editing tools carry them out on a browser-based layered canvas.
This project pairs a Python and FastAPI backend using LangChain and LangGraph with a React and Konva-based canvas editor. A user can generate images from a text prompt, then edit them by describing what they want changed in plain language; the agent analyzes the request, plans the editing steps, and calls from a shared registry of 21 tools covering background removal, color and tone adjustments, crop and transform operations, AI background replacement, outpainting and super-resolution, with the same tools also available as manual toolbar buttons.
It integrates the Segment Anything Model for point-and-click or brush-based region selection, supports local inpainting and region-specific text-guided replacement, and can split an image into semantic layers, such as subject, background and OCR text, that can be independently transformed and are stored as persistent JSONB layer data. For complex requests the agent produces a structured, dependency-ordered multi-step plan that the server validates before execution, and a user can confirm, retry a single step, or cancel the remaining steps. The repository is packaged as a paid tutorial course with accompanying written material.