ArtFlow: AI Drawing Guide
An AI-powered web application that instantly generates custom, step-by-step drawing tutorials from any uploaded image, leveraging Gemini 1.5 Pro's multimodal reasoning and a unique, self-correcting AI Supervisor for error-resilience.
Overview
ArtFlow is an AI-powered web app that instantly turns any uploaded image into a custom, step-by-step drawing tutorial—eliminating the 'blank canvas' anxiety that holds back beginners and children. It uses Gemini 1.5 Pro's multimodal reasoning to deconstruct an image into a cumulative drawing path (shapes → outlines → details) and Imagen to visualize each step, while Firebase handles data and photo storage. Its standout innovation is a self-correcting AI Supervisor loop that validates each generated step against the original goal, restarting on foundational errors and patching minor ones on the fly. As creator and AI developer, Rohan designed the full AI workflow, built the Supervisor, integrated Gemini and Imagen, and set up Firebase Firestore and Storage.
The problem
The crucial real-world problem of personalized art education and eliminating 'blank canvas' anxiety for beginners and children by providing custom, achievable drawing paths.
Key features
- Custom, step-by-step drawing tutorials
- AI Supervisor self-correction loop
- Multimodal image analysis using Gemini 1.5 Pro
- Visual step generation using Imagen
- Firebase data storage and photo handling
What I did
- Designed and implemented the full AI-powered drawing guide workflow
- Utilized Gemini 1.5 Pro for multimodal reasoning and planning the cumulative drawing path (shapes → outlines → details)
- Developed the innovative self-correcting AI Supervisor loop
- Integrated Imagen for visual step generation
- Set up Firebase Firestore and Storage
AI under the hood
ArtFlow's intelligence centers on Google Gemini 1.5 Pro, whose multimodal reasoning deconstructs an uploaded image into a structured, cumulative drawing plan (shapes → outlines → details). Google Imagen then renders a visual for each instruction, keeping the guide and text in lockstep. The defining technique is a self-correcting AI Supervisor: an automated validation loop that compares each generated step against the original goal, triggering a full RESTART on foundational errors and dynamically PATCHING future steps for minor drift—delivering error-resilient tutorials at scale. Firebase Firestore and Storage persist the generated plans and user photos.