Old family photos are emotionally valuable, so an AI restoration tool has to do more than produce a sharp image. It has to make people feel in control of a sensitive, irreplaceable file. While building Clarivnt, an online old-photo restoration tool, I found that the product experience around the model matters just as much as the model output.
1. Make the comparison the center of the interface
A restored photo is difficult to judge in isolation. Users need to see what changed and confirm that the people still look recognizable.
Clarivnt uses a large split before-and-after view as the main visual element. The original stays visible, the restored version sits beside it, and a draggable divider makes the difference easy to inspect. This is more useful than placing two disconnected thumbnails on the page because users can compare the same facial features and damaged areas directly.
The broader lesson applies to many AI products: show the input and output in one continuous context whenever users need to evaluate a transformation.
2. Explain the workflow before asking for a file
The upload area states the supported formats—JPG, PNG, and WEBP—and the 10 MB limit before a file is selected. Nearby copy explains what the tool can improve: fading, scratches, blur, lost color, and soft facial details.
This removes small uncertainties that often cause abandonment. A user should not have to discover requirements after an upload fails.
3. Put privacy language next to the action
A family photograph is not a generic test image. Privacy information belongs near the upload control, not only in a policy page. Clarivnt tells users that photos are not placed in a public gallery and that the original remains untouched.
That sentence is short, but it answers two important questions at the moment they arise: “Will other people see this?” and “Will I lose my original?”
4. Keep the application layer predictable
The web app uses Next.js 16, React 19, TypeScript, and Tailwind CSS. The goal of the stack is not novelty; it is to keep the interaction responsive and maintainable while supporting authentication, restoration credits, localized pages, and the upload/result flow.
For this kind of product, the interface has several states that should remain explicit:
- no image selected
- upload validation
- restoration in progress
- completed result
- comparison and download
- recoverable errors
Treating those as product states rather than scattered loading flags makes the experience easier to reason about and test.
5. Give people a low-risk first result
New accounts receive two free restoration credits. That is enough to test the workflow on a real photo without forcing a purchase decision before users understand the result quality.
For AI tools, a useful free experience is often more convincing than a long feature list. The output itself becomes the product demonstration.
What I would carry into the next AI tool
The strongest lessons were simple:
- Make changes inspectable.
- State file requirements before upload.
- Put privacy reassurance beside sensitive actions.
- Model the full workflow, including failures.
- Let users experience a real result before asking them to pay.
You can try the live workflow at clarivnt.com. I would be interested to hear which part of an AI image workflow most affects your trust: the comparison UI, the privacy policy, or the consistency of the result.
Top comments (0)