You upload a reference image. You type a prompt. The output matches the style of the reference without you writing a single word about style. That is Krea 2, and it changes how you think about AI image generation.
Krea launched Krea 2 on May 12, 2026. It is their first image model built entirely from scratch. Not fine-tuned. Not adapted. A new foundation model designed with one idea at its core: style transfer should be native, not a workaround.
What Krea 2 Actually Does
Krea 2 generates images from text prompts, same as every other model. The difference is what happens when you bring a reference image.
You feed Krea 2 a style reference, any image that has the visual feel you want, and it generates output that carries that style into the result. Watercolour? Oil paint? 90s film grain? Neon editorial? You do not describe the style in words. You show it one image and Krea 2 absorbs the aesthetic: the palette, the texture, the lighting mood, the compositional tension.
This is not img2img where you blend two images. This is not “in the style of” prompt engineering. The model treats style as a first-class input. You give it a reference and it builds the new image inside that visual language.
Why This Matters
Every AI image tool gives you text-to-image. Most let you upload a reference image. Almost none treat that reference as a proper style input. You end up writing long style prompts: “cinematic lighting, warm colour palette, grainy film texture, moody shadows.” Krea 2 skips all of that.
The benchmark numbers back it up. Krea 2 sits 0.14 points behind GPT Image 2 on the Contra Labs human-preference leaderboard. That is close enough that most people will not see a quality gap. The real advantage is not raw quality. It is control.
For anyone making brand content, social media assets, or client work where visual consistency matters, native style transfer is the difference between “this looks sort of right” and “this looks exactly like our brand.” You stop fighting the model to match a mood board. The model reads the mood board.
How to Use It
Here is the workflow:
- Go to krea.ai and select the Krea 2 model (it shows as “Krea 2” or “K2” in the model selector).
- Write your prompt as normal. Describe what you want in the image, not the style. “A person standing at a window in a city apartment at dusk” is enough.
- Upload your style reference image. This can be a photo, an illustration, a film still, a brand asset, anything with a look you want to replicate.
- Adjust the style influence slider if you want more or less of the reference coming through.
- Generate. The output carries the subject from your prompt and the visual DNA from your reference.
That is it. No style keywords. No negative prompts about what you do not want. One image in, one prompt, and the model does the aesthetic heavy lifting.
What to Watch For
Krea 2 is still new. A few things to keep in mind:
- The model comes in multiple quality tiers. Krea 2 Medium is the default. Krea 2 High (or Pro) gives better detail at the cost of speed. Pick based on whether you need speed or fidelity.
- Style reference works best when the reference has a strong, coherent visual identity. A busy collage of mixed styles will confuse the model. One clear aesthetic per reference.
- You can combine multiple references by adding them to a moodboard in Krea’s interface. The model blends the styles. This is useful when you want something between two visual worlds.
- Text generation inside images is still inconsistent. If your output needs legible text, generate the image in Krea 2 and add text in a design tool afterward.
What This Unlocks
Krea 2 removes the biggest friction in AI image generation for anyone doing repeated creative work. If you produce content that needs to look consistent, brand-aligned, or visually coherent across a batch, style reference as a native input means you stop writing style prompts and start showing style examples.
The model is free to try on krea.ai with a limited number of generations. Paid plans unlock higher generation counts and the full-quality tiers.
For small teams producing social content, product visuals, or editorial imagery at volume, this is the first model that makes style consistency genuinely easy. Not perfect. Easy. And in creative work, easy consistency is the thing that saves hours.
Want more like this? Join MOKU Club for free. Weekly resources, early access to new guides, and occasional templates you can actually use. Join below.
