I gave the beta version Krea 3, a try using the free generation allowance.

To be honest, the results didn't impress me.

In any case, even if they release the open weights, I’ll likely continue working specifically on Krea 2.

Krea2 anime UWU

KREA 3 beta

Things are generally fine here, though in my view, Krea 3 looks a bit weaker in terms of composition.

then the next example...

Krea2 anime UWU

KREA 3 beta

Yes, there are a lot of details, and the brushstroke-like execution of the composition is cool; however, it also introduces a "plastic" look, and the overall readability suffers. Besides, you could achieve a similar effect -likely even better- by running an image through Qwen Image 2.1.

This particular piece is 2 megapixels (1536 x 2736), whereas my example is 1 megapixel (768 x 1368).

Krea3 It reminds me a bit of old SD 1.5 fine-tunes, just with better detail or perhaps DALL-E 3.

Krea2 anime UWU

KREA 3 beta

"But your fine-tune creates anime by default - that’s why the result looks like this," you might say. Fair enough. So, I’ve added the "anime" tag to the prompt.

Krea2 anime UWU

KREA 3 beta

The whole prompt doesn't even fit on their website.

Krea2 anime UWU

KREA 3 beta

Well, I didn’t have enough time or enough generated outputs to truly evaluate the model, but I do have some experience that allows me to form an impression based on just a few examples.

Most likely, all the improvements in this version are related to additional supporting features such as regional prompting, editing, and the like.