Story
When I am not sure what I want the AI to make, I want to see a few clearly different versions at once, so I can point at the one that feels right instead of describing it.
Wanjiru, organising a street party in Nairobi, writing a note for her neighbours' group on her phone.
- 01 Setting
Wanjiru asks for a note about Saturday's street party.
- 02 Trigger
She is not sure how it should sound - short, warm, or full of details.
- 03 Without it
She gets 1 note, tries again 4 times, and loses track of the one she liked.
- 04 With it
3 versions arrive together, each labelled by how it is different.
- 05 Outcome
She picks Warm and posts it. The other 2 stay 1 tap away.
Live demo
This demo follows a script, so it plays the same every time - there is no real AI behind it. Change a setting to see the same moment play out differently.
What happens, step by step
| State | The person sees | The AI is | When | Screen reader |
|---|---|---|---|---|
| 1 | The ask. | Starts work. | 0 s | - |
| 2 | "Writing 3 versions", with all 3 filling in at once. | Writes 3 versions. | 0.5-3 s | “Generating a response” |
| 3 | "3 versions - pick 1 to use." Each has a label, the text and "Use this". "3 more" sits below. | Idle. | when all are done | “3 versions ready” |
| 4 | The chosen version lifts. The others fade. | Idle. | on tap | - |
| 5 | The pick becomes the answer. "2 other versions" opens the rest again. | Idle. | after the tap | “Version 2 chosen” |
On each device
- phone
- Versions stack in 1 column, 2 lines each, with the label on top. 3 fit on 1 screen. 6 do not, and people stop comparing.
- tablet
- 2 or 3 side by side. Images sit in a 2 x 2 grid.
- desktop
- 3 or 4 side by side, so the eye moves across and compares. Hover shows the full text of a long version.
When to use it
- Taste matters more than correctness - wording, names, images, layouts.
- People find it easier to point at what they like than to describe it.
- 1 version at a time would mean many retries and lost good ones.
- There is 1 right answer - a fact, a sum, a date. Several versions only cause doubt.
- People want 1 more try, not a set to compare. Use Regenerate, edit and branches instead.
- People already know the change they want - shorter, more formal. Use Adjust length and tone instead.
The parts of the pattern
- ALabelSays how this version is different: "Short", "Warm", "With details".
- BThe set2 to 4 versions, side by side or stacked, all the same size.
- CUse thisPicks 1. The others fold away but are kept.
- DMoreA new set from the same ask, when none of these fits.
Do and Avoid
Short
Street party Sat, 2 pm. Bring a dish!
Warm
Hello neighbours! Games, music and food…
With details
Road closes 1 pm, please move cars…
Hello neighbours! Street party on Saturday…
✓ Using “Warm” · › 2 other versions
Hello neighbours! Street party on Saturday…
The other versions were removed.
Writing and accessibility
- Label each version by how it differs: "Short", "Warm", "With details" - not "Option 1".
- Say the count and the job at the top: "3 versions - pick 1 to use."
- Name the buttons for what they do: "Use this", "3 more".
- Mark up the set as a list. Each version's name starts with its label and place: "Warm, 2 of 3".
- Announce "3 versions ready" once, when all are done - not each version as it finishes.
- "Use this" is a real button named with the label: "Use the Warm version".
- After a pick, move focus to the kept version. "2 other versions" is a button that says if it is open or closed.
Seen in real products
- MidjourneyImage generator · Nov 2026From the image grid, V1-V4 buttons make variations of 1 image. "Vary (Subtle)" makes small changes; "Vary (Strong)" changes more but keeps the main idea.Midjourney Docs: Variations ↗
- Adobe FireflyImage generator · Nov 2026Firefly Image 3 and Image 4 make 4 image variations per prompt; Image 4 Ultra and Image 5 make 1. "Generate more" makes a new set from the same prompt and settings.Adobe Help: Create images from text prompts (updated 4 Nov 2026) ↗
- Image PlaygroundiOS, iPadOS and macOS · Jun 2026Apple's guidelines say Image Playground can make several images of a person, so someone can pick the one they prefer.Apple HIG: Generative AI (updated 8 Jun 2026) ↗
Each example comes from the product’s help pages, release notes or teardowns, with the month it was checked. Products change every month - tell us when 1 is out of date.
Check yourself
Seen this pattern done better, or a product that changed? Press Suggest next to any heading, or tell us about the whole page.