Recraft: Recraft V4 Styles Pro

Recraft: Recraft V4 Styles Pro is an image generation model from Recraft. It sits in the image category and is built around style consistency: the provider describes it as a style-consistent image generation model, and every request requires at least one style reference image. If you need generated images that match a look you already have, rather than a generic rendering, this is the model Recraft positions for that job.

The model accepts text and image inputs and returns images. Its context window is 65,536 tokens, with a maximum output of 58,982. The image parameters are where most of the control lives. aspect_ratio is an enum covering 1:1, 2:1, 1:2, 3:2, 2:3, 4:3, 3:4, 5:4, 4:5, 16:9, 9:16 and auto. n is a range from 1 to 6, so a single request can return up to six images. input_references is a range from 1 to 10, which is how you supply the style reference images the model requires. Note that the supported_parameters list is empty, so the image parameters above are the practical controls rather than the standard sampling parameters you may be used to.

Those capabilities point to a fairly specific set of use cases. Because at least one reference image is mandatory and up to ten can be supplied, the model suits work where a visual style must carry across many generated assets: campaign imagery that has to match an existing art direction, illustration sets for a product or publication, character or scene variations that stay on model, and batch generation where you want several options per prompt via n. The wide aspect ratio enum makes it usable for both square social formats and tall or wide layouts without post cropping. Text input drives the content of the image, while the references drive how it looks.

You call it through api14's OpenAI-compatible API using the model id "recraft/recraft-v4-styles-pro". Because the interface is OpenAI-compatible, existing clients and SDKs generally work with a change of model id and base URL. In the request you pass your prompt as text, include your style reference images as inputs, and set the image parameters you need: aspect_ratio, n between 1 and 6, and input_references between 1 and 10. Remember that at least one reference is required, so a request with text only will not satisfy the model's input requirements.

On limits, the facts here do not describe unsupported features or failure modes, so the main thing to plan around is the reference requirement: if you want pure text to image generation with no style anchor, a model without a mandatory reference input will fit better. Similarly, if you depend on tuning standard parameters, this model exposes none, so check that the image parameters cover what you need before committing.