AI Art Coach

Inspiration

Learning how to draw is fundamentally a process, not a static outcome. Traditional tutorials, like books, videos, or step-by-step images, lack real-time feedback, while human instruction is expensive and inaccessible.

We wanted to explore whether Gemini 3 could act as a live art instructor: one that not only explains concepts, but watches, listens, adapts, and responds as a human teacher would. The inspiration came from classic drawing books and lessons that progressively build sketches from gesture to finish, and the question: what if this process could be interactive, adaptive, and live?

What it does

SketchCoach is a real-time AI art coaching studio that teaches users how to draw through progressive reveal lessons, live critique, and adaptive guidance.

Users select a subject or enter a custom prompt, then follow a structured lesson that moves from gesture → structure → volume → refinement → finish. In Studio Mode, users can draw, upload images, or ask questions while receiving live voice coaching, visual examples, and prioritized feedback. At the end of each session, SketchCoach generates a summary of learnings, images, and chat history for reflection.

How we built it

SketchCoach is powered by multiple Gemini 3 capabilities working together:

Gemini 3 Pro generates structured lesson plans and enforces progressive drawing constraints.
Gemini 3 Image Models create cumulative, step-by-step sketch visuals that build on prior steps.
Gemini Live API enables real-time voice coaching and conversational feedback.
Multimodal reasoning allows the coach to analyze uploaded sketches and respond contextually. Google AI Studio was used end-to-end for rapid iteration, prompt tuning, and orchestration.

Challenges we ran into

The biggest challenge was visual consistency across step-by-step images. Early steps require loose construction lines, while later steps must feel cohesive and polished, without stylistic drift. Solving this required strict prompt constraints, cumulative image conditioning, and separating “goal preview” generation from instructional steps.

Another challenge was balancing verbosity in live coaching, keeping feedback helpful, but not overwhelming, during drawing sessions.

Accomplishments that we're proud of

Built a real-time, multimodal art tutor instead of a static chatbot
Achieved consistent progressive sketch generation across multiple subjects
Successfully integrated live voice coaching with visual feedback
Designed a learning flow that mirrors how professional art instructors teach

What we learned

Gemini 3 excels when treated as an active instructor, not a passive responder. Structuring prompts, enforcing constraints, and orchestrating tools allowed the model to maintain pedagogical consistency across time, modalities, and user actions. We also learned that high-quality creative tools require process awareness, not just output quality.