ImagineArt Streak
A 4-day retention loop that turns daily free-credit users into paid subscribers.
Consumer AI
CAPTIONS
imagine.art/video/captionsUpload a video, pick a style, and add captions to it. It can translate them into other languages too.
- Year
- 2026
- Role
- Lead Product Designer
- Scope
- Research, design, tracking
Visitors
2,073
first eight weeks
Made a video
72%
of everyone who landed
Downloaded it
52%
start to finish
Time live
8 wks
300 to 400 visitors a week
01/Context
It had to be the easiest one.
ImagineArt already had video tools. Captions was one more tool that people search for by name, so it had to feel simpler than everything else out there. I gave it one rule: as few steps as possible.
The hard part is accuracy. Nobody notices a hundred right words, but everyone notices one wrong name. And names are the words the AI gets wrong most. How we fixed that became the biggest decision in the project.
02/Approach
Test it first. Then stay out of the way.
Before we drew any screens, I hooked up another company's caption tool, Veed, and used it from start to finish. You only feel the annoying parts when you use a thing yourself. That test showed where people slow down: choosing settings, and fixing wrong words.
The tool also had to be where people already are, so nobody has to go looking for it. Captions sits in the same menu as every other video tool, and on the toolbar of any video you already made. I also planned where it should go in Ads Studio, because ads usually play with the sound off. That part is planned, not built yet.
03/Styles
Pick a look by looking.
Every style is a real picture with the captions already on it, because nobody picks a caption style from a list of font names. The panel shows nine styles, and See All shows the rest.
The styles themselves came from working sessions with our creative team. They make videos like this every day, so they know which looks people really use. Each word lights up as it is spoken. That is the fun part of this caption style, so it is on from the start.
04 / The big decision
No big text editor.
Back to the one wrong word. The normal fix is a big text editor with a timeline. That takes months to build, and it makes the user check every word, when they came with a finished video and no time to spare. I asked for something much smaller: let people swap the word.
The mistakes are almost always the same kind. A brand name, a product name, a special term. So you type the right word, tell it which word to replace, and run it again. It took a small part of the time to build, and there is nothing to learn.
The obvious fix and the right fix are not always the same size.
05/Outcome
Most visitors make a video. Half of them download it.
Mixpanel · ImagineArt · /video/captions · 20 Jul to 15 Sep 2026
Visitors
2,073
in eight weeks
Made a video
72%
1,486 people
Downloaded it
52%
1,070 people
Bought credits
2%
46 people
Visitors went from 6 in the first week to a steady 300 to 400 a week. 97% of them use a laptop, which ended a long debate about mobile.
Money is the open question. Captions brings people in and gets them to a finished video. The next job is to turn that video into a purchase, and to add proper tracking so we measure each step instead of guessing.
