MEET & DRAW

A quiet drawing-practice PWA where AI creates the references and you keep drawing by hand.

Comment

Inspiration

Generative AI is often presented as a way to make the final image for us. I wanted to explore the opposite direction: can AI help a person spend more time observing and drawing by hand?

I began practicing portrait and figure drawing because, in an AI-heavy creative environment, looking carefully and making marks still feels important. The practice works, but searching for references interrupts it. MEET & DRAW grew from the idea that an app could quietly prepare the practice environment without taking over the creative act.

What it does

MEET & DRAW presents one fictional AI-generated person at a time. The artist chooses Portrait, Figure, or Mix; selects a pace from 30 seconds to 10 minutes or No limit; chooses 5, 10, 15, or 20 references; and starts drawing.

The session includes automatic switching, Pause, Next, and Exclude. An excluded reference is removed from that browser, does not count toward completed drawings or practice time, and is replaced when another reference is available. A local 30-day guard prioritizes people the artist has not recently drawn.

When the session ends, Session Review shows every person encountered in order. Each image has a short ID such as portrait-031. The artist can write that ID in a sketchbook, then later search the Reference Library using the full ID or a fragment such as 031. References can also be marked Favorite, viewed in detail, or downloaded as the original WebP.

The current library contains 70 references: 42 Portraits and 28 Figures.

How we built it

The app is built with React, TypeScript, and Vite. vite-plugin-pwa generates the Manifest and Service Worker, and the static demo is hosted on GitHub Pages. A versioned localStorage model keeps settings, image preferences, encounters, and practice-session records without requiring an account or backend.

The custom responsive grid supports iPhone, iPad portrait, iPad Split View, and Mac. Its quiet white interface keeps the person and drawing time visually dominant.

How GPT-5.6 and Codex were used

TAROUT led the project’s purpose, drawing philosophy, feature priorities, reference-image standards, and final approvals.

GPT-5.6 helped define the Build Week MVP, drawing flow, Exclude and Favorite rules, local data boundaries, and standards for the fictional reference library. It also helped refine the international English language used throughout the experience.

Codex carried those decisions into implementation and verification. It built the React and TypeScript PWA, responsive layouts, timing and session logic, local history, Session Review, Download, Image ID search, and the 70-reference metadata pipeline. It also ran automated tests, production builds, responsive and offline checks, browser verification, documentation, and the GitHub Pages deployment.

The result was a collaborative workflow: human direction and quality judgment, GPT-5.6-assisted reasoning, and Codex-driven implementation and testing.

Challenges we ran into

The hardest state transition was Exclude. A rejected image needed to disappear immediately and not count as practice, while still remaining in the session’s encounter history. Favorite and Excluded also had to remain mutually exclusive.

We also worked carefully on the reference library. The goal was not generic beauty or advertising photography, but readable anatomy, varied ages and facial structures, dynamic expressions and gestures, realistic asymmetry, and a neutral presentation suitable for drawing.

Accomplishments that we're proud of

The Build Week app has 70 verified references, three practice modes, session memory, Image ID search, Favorite, Exclude, original-image Download, and offline PWA behavior.

Twenty-five automated tests pass, and all 70 public image URLs were verified. The app also passed iPhone and iPad device checks, including Add to Home Screen.

Most importantly, the product keeps AI in a supporting role. The user still does the difficult and valuable part: looking, deciding, and drawing.

What we learned

A small piece of durable identity can connect digital practice to a physical sketchbook. Writing portrait-031 beside a drawing turns the Image ID into a path back to the person and the time spent observing them.

What's next for MEET & DRAW

Future directions include user-created nicknames, Person Profiles, a careful Meet Again experience, cross-device continuity, and demand-led library growth with explicit human review.

Built With

Share this project:

Updates