Recraft: Recraft V3

Recraft V3 is an image generation model from Recraft, a provider that specialises in visual generation tooling. It sits in the image category on api14, which means it is built to produce pictures rather than to hold conversations or write text. If you are building a product that needs generated imagery, this is the kind of model you would evaluate alongside other image models on the platform.

The model accepts text and image inputs and returns image output. That combination matters: you can describe what you want in a prompt, and you can also supply a single input reference image to steer the result. Output is generated at roughly 1K resolution. The context window is 65536 tokens, with a maximum output of 58982 tokens. Aspect ratio is configurable across 1:1, 4:3, 3:4, 16:9, 9:16 and auto, so you can match the shape of the image to the slot it will occupy in your interface rather than cropping afterwards. You can request between 1 and 6 images in a single call, and you can attach up to one input reference.

Those capabilities point to a fairly specific set of jobs. Marketing and product teams can generate hero images, social assets and thumbnails at whatever aspect ratio each channel needs, requesting several variations at once and picking the best. Because a reference image can be supplied, the model also suits workflows where you want a new composition that stays close to an existing visual, such as restyling a product shot or producing alternates of an approved concept. The auto aspect ratio option is useful when you would rather let the model choose a sensible framing from the prompt alone.

Calling it through api14 works the same way as any other model on the gateway, using the OpenAI-compatible API with the model id "recraft/recraft-v3". The model itself exposes no general supported parameters, so generation behaviour is controlled through the image configuration instead. That configuration covers aspect_ratio, which takes one of the six enum values; n, which ranges from 1 to 6; and input_references, which ranges from 0 to 1. If you already have an OpenAI client wired up, pointing it at api14 and setting the model id is the main change required.

One thing to weigh is that this is purely an image model. It does not return text, so it will not help with reasoning, summarisation, classification or any task where the output needs to be language. If your pipeline needs both a written response and an image, you will want a text model alongside it. The single reference image allowance is also worth noting: if your workflow depends on blending several reference images, this model will not do that.