Why We Didn't Make Prompting the Core UX of an AI Couple Photo Generator

Recommended tags: ai ux product image-generation

A lot of AI products default to the same interaction: give people a big prompt box and let them figure it out.

We deliberately did not do that for our couple photo product.

The reason is simple. People do not come to an AI couple photo tool because they want to experiment with wording. They come because they want to turn two solo photos into one believable shared photo. That is a very different job.

In this workflow, prompt text can describe style, but it does not reliably solve the hardest parts:

• whether both people still look like themselves
• whether two mismatched source photos can be merged naturally
• whether pose, spacing, and eye line feel like a real photo
• whether the result looks like a memory instead of a collage

If prompt becomes the main control, users end up doing failure recovery for the system. The photo looks wrong, so they type more. Identity is off, so they type more. The scene feels fake, so they type more again. That may look flexible on paper, but in practice it just pushes model uncertainty onto the user.

We found that structured choices work better for this kind of task. Let people choose a scene, a style, and a composition direction first. If they want, they can add one extra preference later. That keeps the interaction expressive without making every user behave like a prompt engineer.

For products like this, the real question is not “How much control can we expose?” It is “How much uncertainty can we remove before the user even has to think about it?”

That is the product direction we are taking with AI Couple Photo: prompt can exist, but it should be a secondary control, not the main interface.