Krea: K3 Edit

Krea: K3 Edit is an image editing model from Krea. It sits in the image category and is built specifically for image-to-image work rather than generation from text alone. If you already know Krea as a provider, this model is the editing counterpart to their image generation line, and it is identified on api14 by the model id krea/k3-edit.

The model accepts text and image inputs and returns an image. You supply a text prompt together with one to four reference images, and it produces a single edited image that follows your instructions. The context window is 65,536 tokens, and the maximum output is 58,982 tokens. Output resolution is chosen from two options, 1K or 2K, and the aspect ratio is selected from a fixed set: 1:1, 4:3, 3:2, 16:9, 4:5, 2:3 or 9:16. A seed option is also exposed as a boolean, so you can control whether a seed is used in the request.

Those capabilities point to a fairly specific set of jobs. Because the model takes references rather than starting from nothing, it fits workflows where an existing image needs to be modified according to a written instruction: adjusting or restyling a photo, applying a consistent look across a small set of reference images, or producing a new composition that stays anchored to the material you provide. The one to four reference range means you can pass a single image for a straightforward edit or several images when the result needs to draw on more than one source. The resolution and aspect ratio options let you match the output to a target format, whether that is a square social asset, a wide banner or a vertical layout.

Calling it through api14 works the same way as any other model on the platform, since the API is OpenAI-compatible. You send your request to the standard endpoint and set the model field to krea/k3-edit, passing your prompt as text and your reference images as image inputs. The model does not list any supported sampling parameters, so there is no temperature, top_p or similar tuning to configure. What you do control is the image side of the request: resolution (1K or 2K), aspect ratio from the seven listed values, one to four input references, and the seed boolean.

One thing to keep in mind is that the output is always an image, so this model is not a fit if you need text back from your prompt. It is also an editing model, which means it expects at least one reference image; if you want an image produced purely from a text description, a text-to-image model would be the better choice. Within those bounds, krea/k3-edit is a reasonable pick when your task is to transform images you already have.