AI
Backgrounds

Designing AI-generated backgrounds from 0→1, with a prompt composer that made generative AI feel approachable.

Company
Zoom
Role
Lead Product Designer
Focus
AI product design
Zoom’s AI background generator in a meeting

6-8 weeks

from first idea to release, with a team of 3 representing design, product and engineering.

The insight

Most-used setting hadn’t changed since launch.

Zoom had just gone AI-first, and backgrounds, one of our most-used features, were the obvious place to start. People still wanted meeting experiences with privacy and professionalism.

The original backgrounds: six defaults and a tiny plus button

The gap

Custom backgrounds hiding in plain sight.

Most people never found the option to add their own. It was hidden behind a small plus icon and still left you to find and size an image yourself.

The add background option, behind a tiny plus button

Discovery

Moving fast, but not blind.

Speed meant keeping research light, so I paired it with early ideation alongside engineering. Every idea started from what we could build.

Competitive scan

Who set the bar for AI images

  • DALL-EText-to-image AI. Set the benchmark for prompt-driven generation.
  • MidjourneyHigh quality and style-forward, with a strong community and defaults.
  • Canva AIAccessible and integrated. Backgrounds inside an existing workflow.

Customer voice

A light touch, on purpose

50+internal surveys
7customer advisory panel members
  • →  Unmoderated survey: directional, not conclusive
  • →  Early signal to catch AI trust concerns before beta
  • →  Advisory panel for qualitative depth

First pass

Teaching people how to prompt.

Scrolling example prompts showed what to ask for. Four images came back, with art styles tucked away.

Early iteration: empty prompt, ideas and styles
Early iteration: four results loaded

Pivots

Real limits made it simpler.

  • Four images became two: faster and cheaper
  • Free typing gained prompt buttons, refreshed every few months
  • A flexible drawer became a fixed one, to fit the tech stack
Before: four results and a style dropdown
After: two results and style buttons

Final design

One prompt, endless looks.

Type an idea or tap a suggestion, restyle it in a click, and it drops straight into the meeting.

Art style options on a generated background
A generated background in a watercolor style
A generated background in a comic style
A generated background in a cinematic style

Mobile

Familiar on mobile, lighter by design.

At launch, mobile kept the backgrounds sheet people already knew and simply pulled in the AI backgrounds they’d made on desktop. No prompt bar on a small screen. Next, I’d let people say it instead of type it, then fix anything misheard before generating.

Phone screen: picking from your AI backgrounds

Shipped

Pick from your AI backgrounds

Made on desktop, ready on your phone.

Phone screen concept: dictating a prompt

Concept

Say it

Dictate a prompt instead of typing on a small keyboard.

Phone screen concept: fixing a misheard word before generating

Concept

Fix it, then generate

See the transcript and tap any misheard word to correct it.

What I took away

Ship fast, but plan how you’ll measure.

I’d have loved to know how many people typed versus tapped, and which styles they tried.