Krea: Krea 2 Medium

Krea: Krea 2 Medium is an image generation model from Krea. It sits in the image category and is described by its provider as a balanced, cost-efficient option and a practical starting point for a broad range of use cases. If you are choosing an image model for a new project, this one is positioned as a general-purpose default rather than a specialist tool.

The model accepts text and image inputs and produces image output. Its context window is 65,536 tokens, with a maximum output of 58,982 tokens. Generation is controlled through image parameters rather than the usual sampling controls: resolution is an enum with a single value, 1K, and aspect ratio is an enum covering 1:1, 4:3, 3:2, 16:9, 4:5, 2:3 and 9:16. You can supply between zero and one input references, and a seed is available as a boolean option. The provider notes that extensive post-training supports stable, consistent generations.

Those capabilities point to a fairly wide set of practical uses. Text to image work covers the obvious cases: concept art, marketing and social visuals, illustrations, and rapid exploration of visual ideas. Because the model also takes image input with up to one reference, it suits reference-guided generation, where you want an output that follows the style, subject or composition of a supplied image. The fixed 1K resolution and the range of aspect ratios make it straightforward to target common formats, from square posts to wide 16:9 frames and vertical 9:16 stories, without extra resizing steps. The seed option helps when you need to reproduce or iterate on a particular result.

You call it through api14's OpenAI-compatible API using the model id "krea/krea-2-medium". Because the endpoint follows the OpenAI format, you can point existing OpenAI client libraries at api14 and change the model id rather than rewriting your integration. Note that the model lists no supported parameters in the usual sense; the controls it exposes are the image parameters described above, namely resolution, aspect ratio, input references and seed. Plan your request construction around those fields.

It is worth checking whether this model fits before committing. Resolution is limited to 1K, so work that needs higher native output would require upscaling or a different model. Input references are capped at one, so tasks that depend on combining several reference images are not a good match. If your workload is text generation rather than image generation, this model is not the right choice at all, since its only output modality is images.