Krea Ai: Your visual direction, exactly as you imagined it

Krea 3 is Krea AI's image generation and editing model. Krea 2 generates from text prompts and up to 10 style references, and only the look of your references carries over. Krea 3 edits and combines up to 4 reference images, so the content carries over: merge a face, an outfit and a backdrop into one image, restage a product in a new scene or material, and keep the same character across follow-up shots.

Krea 3 also outputs at up to 2K with custom dimensions, compared to 1K in Krea 2. If you're generating from scratch in a specific style, Krea 2 is still the right pick. If you're editing or compositing, use Krea 3.

Upload up to 4 reference images, describe the change in plain text and generate. Krea 3 reads each reference as content, not just style, so it can place a product in a new scene, swap a material or change one named area while the camera, framing and layout stay as they were.

For the most predictable results, give each reference one job by number ("person from image 1, gown from image 2"), name the exact target ("the wall behind the staircase") rather than asking for a general redesign, and say what must stay ("preserve camera, framing, lighting and furniture positions").

Krea 2 has two quality settings. Medium is faster, with stable, consistent outputs. It’s particularly strong on illustration, anime, painting, and expressive styles, and the right starting point for most use cases. 

Large is a bigger setting with rawer, more textured outputs. It’s built for photorealism and visual treatments like motion blur, grain, and low dynamic range. 

Both settings support the same inputs and features. The difference is in the character and quality of what they produce.

Krea 2 Turbo is an accelerated generation mode that produces images in roughly two seconds. It’s built for rapid iteration: testing directions, swapping references, and exploring layout options without waiting on render times.

Turbo supports the full text-to-image workflow, including style references: upload up to 10 reference images and run generations at Turbo speed to quickly test how your visual direction translates into output. Once the direction is locked, switch to Standard mode for a more refined result.

You can use up to 10 reference images in a single generation. Krea 2 reads the visual style of each (lighting, color, feel), and applies it to your output. So if your prompt is “a girl riding a bike through a wheat field” and you add a Picasso painting as a reference, that’s what you’ll get — your scene, rendered in that visual style. The content of your references doesn’t carry over, only the look. It’s more useful for working from moodboards or pulling a specific aesthetic than trying to describe a style in words.

Krea 2 generates at up to 1K across seven aspect ratios: 1:1, 16:9, 9:16, 4:3, 3:2, 2:3 and 4:5. Krea 3 generates and edits at up to 2K (2048×2048) with custom width and height, snapped to multiples of 16, and outputs PNG or JPEG. Both download in formats suitable for digital use, social media and most production workflows. For print-scale finishing, run either model's output through Artlist's upscaling after generation.

No. Krea 2 is built with professional controls, but the generation itself is straightforward. Simple prompts produce well-composed outputs without requiring complex prompt engineering. The Creativity setting handles how much the model fills in on its own, so you can keep it tightly anchored to your prompt or let it interpret more freely. Reference images do more work than long descriptions, and Turbo mode makes it fast to experiment and see what lands.

For specific details on AI training use, refer to Artlist’s Privacy Policy. Your text prompts, uploaded reference images, and generated outputs are accessible and can be deleted at any time through your Artlist account.