Recraft: Recraft V4.1

Recraft: Recraft V4.1 is an image generation model from Recraft. Recraft develops generative image tooling, and this model sits in that line as a version tuned for high aesthetics. It is an image output model, meaning that whatever you send in, what comes back is an image rather than text. The model id on api14 is recraft/recraft-v4.1, and it belongs to the image category.

On the input side, Recraft V4.1 accepts text and images, so you can drive it with a written prompt, with a reference image, or with both together. Its output is an image, produced at roughly 1K resolution. The model carries a context length of 65,536 tokens and a maximum output of 58,982 tokens. Aspect ratio is a first class concern for this model: it supports 1:1, 4:3, 3:4, 16:9, 9:16, and auto. You can also ask for more than one image in a single request, since the n parameter accepts a range from 1 to 6. Reference input is similarly parameterised, with input_references accepting 0 or 1, so one reference image at most per request.

Those capabilities point to a fairly specific set of jobs. Prompt driven generation covers the usual ground: artwork, illustrations, concept imagery and product or marketing visuals where the aesthetic quality is the point. Because the model takes image input as well, it also works when you want the output to follow something you already have, for example generating a variation or a new composition from a single reference while keeping the look consistent. The aspect ratio choices matter here: 1:1 and 4:3 suit social and editorial layouts, while 16:9 and 9:16 map onto wide banners, presentations and vertical story formats. Requesting several images at once (n up to 6) is useful when you want to sample options before picking one, though note the single reference image limit if your workflow assumes supplying multiple.

Calling it through api14 uses the same OpenAI compatible surface as the rest of the catalogue. Send a request to the chat completions endpoint with the model id recraft/recraft-v4.1, and read the image back from the response. The supported parameters list is empty, so there are no generic sampling parameters such as temperature or top_p to tune. Instead, this model exposes image specific parameters: aspect_ratio, which takes one of the enumerated values (1:1, 4:3, 3:4, 16:9, 9:16, auto); n, an integer between 1 and 6; and input_references, which accepts between 0 and 1 reference image. Beyond those, prompt content is what shapes the result.

There are no published benchmarks or scores here, so treat Recraft V4.1 as an aesthetics oriented generator rather than something you can compare on leaderboards. If your task is text generation, classification, or anything that expects a text response, this is not the model for it, since its only output modality is images. If you need to condition on several reference images at once, the input_references maximum of 1 will constrain you. Within image generation, it is a reasonable default when visual quality and control over aspect ratio and batch size are the priorities.