Rohan LopesRohan LopesSenior Integration & AI Architect
HomeWorkBlogAboutContact

Rohan Lopes

Senior Integration & AI Architect. Building integrations and AI agents that do real work.

HomeWorkBlogAboutContactBlog archive

© 2026 Rohan Lopes · Built with Next.js and Claude Code · Privacy

    All posts
    AI
    Gemini
    Hackathon
    Firebase

    From Image to Masterpiece: How We Built ArtFlow, the AI Drawing Guide, at the Google AI Hackathon

    Our Google DeepMind hackathon entry: ArtFlow turns any photo into a personalized, step-by-step drawing tutorial using Gemini 3 Pro.

    December 13, 2025 · 2 min read

    We recently participated in the Google DeepMind - Vibe Code with Gemini 3 Pro in AI Studio Hackathon, and we're thrilled to share the details of our submission: ArtFlow - AI Drawing Guide.

    Our goal was simple yet profound: to eliminate "blank canvas" anxiety and make personalized art education accessible to everyone. ArtFlow transforms any uploaded photo into an instant, custom, step-by-step drawing tutorial, acting as a true personalized teacher.

    💡 What ArtFlow Is

    ArtFlow is a novel web application that solves the crucial real-world problem of personalized art education. It instantly generates custom, step-by-step drawing tutorials from any uploaded image, tailoring the curriculum for beginners and children.

    🖼️ Custom Tutorials

    Instantly generates a unique, multi-step guide from any uploaded image, moving from observation to creation.

    🧠 Gemini 3 Pro Reasoning

    Utilizes multimodal reasoning to deconstruct the image, identifying style, colors, and the ideal drawing path.

    🛡️ AI Supervisor (Innovation)

    A unique self-correcting loop that validates and patches the tutorial steps for error-resilience.

    ✨ Imagen Visuals

    Each step is visualized clearly by the Imagen model, ensuring the guide is easy to follow.

    ⚙️ How It Works: The Core Workflow

    ArtFlow utilizes a robust, multi-step process powered by Google’s Gemini and Imagen models.

    1. Planning and Reasoning Task Gemini 3 Pro’s multimodal reasoning is used to deconstruct an image, identifying style, colors, and a cumulative drawing path (e.g., shapes → outlines → details).

    2. The AI Supervisor: Self-Correction Loop This unique innovation validates the tutorial during execution. It compares generated images against the original goal.

    If foundational errors occur, it triggers a RESTART; for minor issues, it dynamically PATCHES future steps to correct the course.

    3. Visual Generation Each step is visualized by Imagen, ensuring the guide matches the text instructions perfectly.

    4. Data and Storage The application integrates Firebase Firestore and Firebase Storage for handling data and uploaded photos.

    🛠️ Key Technologies Used

    Google Gemini 3 Pro · Google Imagen · Firebase Firestore · Firebase Storage · Google AI Studio


    Why It Matters

    ArtFlow eliminates "blank canvas" anxiety. By combining advanced reasoning with self-correction, we provide a scalable, error-resilient art tutor that empowers anyone to move from observation to creation, regardless of their initial skill level.

    Authors: Rohan Lopes, Anju Tuscano

    Competition Track: Overall Track

    View Official Writeup

    Project

    ArtFlow: AI Drawing Guide

    All postsOriginally published on blog.rohanlopes.com