Recraft: Recraft V4.1 Utility

Recraft V4.1 Utility is a general-purpose image generation model from Recraft, listed on api14 under the model id recraft/recraft-v4.1-utility in the image category. It is positioned as a utility model, which in practice means it handles everyday generation work rather than a narrow specialty, and it comes from a provider that focuses on image generation tooling.

The model accepts text and image inputs and produces image output. That means you can prompt it with text alone, or pair a prompt with a single reference image. Output is generated at roughly 1K resolution, and you can choose from several aspect ratios: 1:1, 4:3, 3:4, 16:9, 9:16, or auto. A single request can return between 1 and 6 images, and the model supports up to 1 input reference image. The context window is 65,536 tokens, with a maximum output of 58,982 tokens. If you need more references per request or higher output resolutions, you would need to look at other image models on api14.

Because it takes both text and images as input, the model fits workflows where you want to generate variants from an existing image as well as from a written prompt. Typical uses include producing social or web assets in the aspect ratio a placement requires, generating several candidates at once with n set between 1 and 6 and picking the best, and using a reference image to steer style or composition while a text prompt describes the content. The aspect ratio options map directly onto common formats: square, landscape and portrait at 4:3 and 3:4, and widescreen or vertical at 16:9 and 9:16.

You call it through api14's OpenAI-compatible API using the model id recraft/recraft-v4.1-utility, so existing OpenAI-style client code works with a change of base URL and model string. The model itself does not accept any of the standard sampling parameters, which are not supported. What it does expose is a set of image parameters: aspect_ratio, an enum over 1:1, 4:3, 3:4, 16:9, 9:16 and auto; n, an integer from 1 to 6 for the number of images per request; and input_references, an integer from 0 to 1 for the number of reference images. Pass these in the request alongside your prompt.

Since the model has no supported sampling parameters and caps input references at one, it suits straightforward generation and light reference-guided work rather than fine-grained parameter tuning. If you need multiple reference images in a single call or other control parameters, another image model on api14 may be a better fit.