We have now had enough usage on the Mianki AI platform to draw some meaningful conclusions about how creative teams actually use AI image generation — as opposed to how they say they will use it. The gap between the two is instructive.

Here is what the data showed us.

The 10-image training threshold is real

We consistently see a quality cliff at ten images. Styles trained on fewer than ten images produce outputs that are recognisably in the right direction but miss the subtle details that make a brand style feel precise. Styles trained on ten to fifteen images — particularly when those images are carefully selected to represent the full range of the brand's visual identity — produce outputs that even experienced designers struggle to distinguish from the originals.

The lesson: the training selection matters as much as the training quantity. Ten images that show the same scene from slightly different angles are worth less than ten images that show the breadth of the brand's visual world.

Teams are using it for things they didn't expect

When we ask teams what they planned to use Mianki AI for before they started, the answers cluster around campaign imagery and social content. When we ask what they actually use it most for, a different picture emerges.

Pitch deck imagery. Internal presentation materials. Email header visuals. Blog post illustrations. These are the high-volume, low-glamour visual needs that eat enormous amounts of designer time and that nobody thinks to mention when building a business case for an AI tool. Teams are using Mianki AI to solve these problems at scale — and the time savings here are, in aggregate, larger than the savings on campaign production.

The prompt learning curve is real but short

New users consistently report that their first generation sessions produce results that are good but not quite right. By the third session, the quality of prompts has improved dramatically. By the fifth, most users are producing outputs they describe as production-ready on the first generation.

The learning curve is not about learning to use the tool. It is about learning to describe what you want with precision. The teams that invest time in developing a prompt library — a set of tested, refined prompts for each type of image they regularly need — see dramatically better results than those who write prompts from scratch each time.

Quality beats quantity every time

The generation sessions that produce the most usable outputs are not the longest ones. They are the most focused ones. Teams that generate fifty images targeting a specific scene with refined prompt language consistently get better results than teams that run five hundred prompts broadly exploring what the tool can do.

The data suggests that the most effective Mianki AI users approach each session like a focused brief rather than an open exploration. They know what they need. They describe it precisely. They generate, review, and refine quickly. The creative discipline that makes a great designer makes a great Mianki AI user.