Krea 2 is now available through a public API. The model, which Krea built entirely from scratch, is the company's first in-house foundation image model and its clearest statement yet that it wants to own the full creative stack, not just wrap other people's models. The API launch puts it directly in the hands of developers who want to build style-aware image generation into their own products.

A different bet on what image generation should be

AI image generators have reached a strange point. Many of them are powerful, fast, and visually impressive, but their outputs can start to feel familiar: the same polished lighting, the same cinematic glow, the same "AI image" texture that looks good at first glance but becomes easy to recognize. Krea's answer to that problem is Krea 2.

Krea 2 is their first foundation image model, trained completely from scratch to give users full control over the look, feel, and creative direction of every image they create. What differentiates it at the training level is data curation and objective function: Krea trained it heavily on aesthetically curated visual content, including editorial photography, fine art, design, and concept work, and tuned the reward signals to prioritize visual harmony, coherent lighting, and style consistency over raw prompt adherence. The result is a model that interprets prompts through an aesthetic lens rather than a literal one.

Krea 2 Medium has already debuted at #6 on the Artificial Analysis Text-to-Image Leaderboard, trailing only models from OpenAI, Google, and NVIDIA. Notably, Krea 1 was developed in collaboration with Black Forest Labs, making this the first model family Krea trained entirely on its own.

What the model actually does well

Most image models handle complex prompts well, but tend to fall short on style, defaulting to something sharp, polished, and "safe." That narrows the creative range you can get out of them. With Krea 2, the team put as much effort into how the model understands style as how it understands prompts. The result is a model that can render a wide range of aesthetics: expressive, raw, niche, experimental, without falling into AI looks.

The model ships with four headline capabilities: