WorkRésuméContact
All work
Meta AI · 2024

Imagine Yourself

Meta AI's tuning-free personalized image generation, see yourself in any scene from a single reference photo, no per-person fine-tuning required. I built the prototypes we used to test the models in research sessions, and the promo that introduced it.

Official promo, via Meta on Facebook.
Imagine Me face-scan flow in Instagram
In-product face-scan flow, via Meta Newsroom.
Imagine Yourself generated images with prompts, 1
Imagine Yourself generated images with prompts, 2
Imagine Yourself generated images with prompts, 3
Imagine Yourself generated images with prompts, 4

Generated results with prompts, via Meta Newsroom.

RolePrototyping · Research
Year2024
TeamMeta AI
FocusGenAI · Personalization

Overview

Imagine Yourself lets people put themselves into AI-generated images, reimagined as an astronaut, in a painting, anywhere a prompt can describe, from just a single photo. Unlike earlier approaches, it's tuning-free: there's no lengthy per-person training step, so the experience can feel immediate.

I built the interactive prototypes the team used to test the underlying models with real people in research sessions, and the Origami-built promo that introduced the feature. The prototypes turned a research capability into something you could actually hold, prompt and react to.

Expand your world

The feature promo introducing Imagine Yourself, and Mark Zuckerberg sharing it.

Feature promo, via Meta on Facebook.
Mark Zuckerberg on Instagram.

Shared by Mark

Mark Zuckerberg sharing Imagine Yourself on Threads.

Via @zuck on Threads.

What the prototype had to prove

One photo in

A single reference image is enough, no fine-tuning, no waiting to train a model on you.

Prompt to place

Describe a scene and see yourself in it, testing how identity, likeness and prompt intent hold together.

Feels immediate

Pacing, loading and reveal designed so a heavy generative process feels responsive and alive.

Honest to the model

Wired to real outputs, so what participants reacted to in research was the actual model behaviour.

Prototyping to test a model, not just a UI

When the thing you're evaluating is a generative model, the prototype has to be real enough that people react to the model, not to a mockup around it. I built the experience so research participants could prompt it, wait through the generation, and respond to genuine results, so the team could learn how the model felt in someone's hands.

Those sessions fed directly back into how the feature was shaped before it reached Meta AI.

There's more inside

The research prototype and session material are protected. Enter the access code to watch the walkthrough.

Incorrect code, try again.
Protected · Unlocked

The research prototype

A full, interactive walkthrough of the experience, the model we put in front of participants to test how the generation felt, prompt by prompt.

Full prototype walkthrough, used to test the AI models in research sessions.

Want the full story?

I'll walk you through it.