# Building a look-first AI image generator instead of another blank prompt box

> Source: <https://dev.to/eudy/building-a-look-first-ai-image-generator-instead-of-another-blank-prompt-box-3deo>
> Published: 2026-08-09 16:22:44+00:00

Most AI image tools still begin the same way:

Here is a blank prompt box. Now describe exactly what you want.

That is powerful if you already know how to write prompts. But for many users, especially when the goal is visual style, the hardest part is not generation. It is deciding what to ask for.

I have been building [BlingBling Up](https://blingblingup.app/), an AI image generator focused on a different workflow:

Instead of starting from text, the product starts from taste.

Prompt-first tools are flexible, but they also create a lot of friction.

If someone wants a glam, fashion, editorial, or soft-glam image, they may not know whether to write:

Even when users know the rough direction, they often struggle to translate taste into prompt language.

So I wanted to test the opposite flow:

Let users pick the direction visually, then let the system carry the prompt detail.

The public gallery works like a collection of visual starting points. Each Look contains an image direction and prompt context. The user can choose one, add their own photos, and generate a result from there.

The current version focuses on a few style pages:

I intentionally kept the category set small. It is tempting to create dozens of SEO pages and style categories immediately, but that usually creates thin pages and a worse product experience.

For now, I would rather have fewer pages that are easier to understand.

The first screen is not a marketing landing page with abstract claims. It shows the actual workflow:

``` php
Pick a Look -> Add a photo -> Create
```

The gallery appears directly below the hero area, so users can start browsing immediately.

Early versions of AI products often expose too much:

For this product, those details are secondary. The user is trying to find a look they like.

So the discovery experience is intentionally simple:

The creation flow supports saved image profiles and one-time photo uploads.

For uploads, the rule is simple:

That keeps the workflow lightweight while still giving the generation system enough visual reference.

Instead of exposing model names in the main product flow, the UI uses quality levels:

That is easier for users to reason about than provider-specific model language.

The project is built as a web app with Next.js and deployed on Cloudflare.

The app has a few moving parts:

One thing I had to be careful about was balancing SEO with performance.

For example, a gallery full of large images is useful for discovery, but it can make the homepage slow. The better compromise is:

That way the product does not sacrifice user experience just to chase indexing.

Positioning is the hardest part.

There are several possible ways to describe the product:

The broad keyword has more search demand, but the narrower positioning is more memorable.

Right now I am trying to make the first screen communicate this:

Choose a polished Look, add your photo, and create a beautiful AI image without starting from a blank prompt.

That is the core idea.

If you try it, I would love feedback on three things:

You can check it out here:

I am still early, but the main lesson so far is that AI product design is often less about adding more controls and more about choosing the right starting point.

For this product, the starting point is not a prompt.

It is a Look.