Recraft V3
Recraft V3 is Recraft's second-generation image model. It adds 20+ specialized style presets, native vector output, long multi-word text generation with precise positioning, and improved anatomical accuracy over Recraft V2.
import { experimental_generateImage as generateImage } from 'ai';
const result = await generateImage({ model: 'recraft/recraft-v3', prompt: 'A red balloon on a wooden table.'});Frequently Asked Questions
What style presets does Recraft V3 include?
Over 20 specialized styles spanning realistic portraits, digital illustrations, pixel art, vector graphics, and more. Each preset applies a consistent aesthetic without style descriptions in the prompt.
Does Recraft V3 support vector output?
Yes. Recraft V3 generates scalable vector graphics directly. Use it for logos, icons, and print assets where resolution independence matters.
How does text generation differ between V2 and V3?
V3 handles long multi-word text with precise positioning in the composition. V2 renders text accurately, but V3 extends this to longer copy and gives finer control over placement.
What does anatomically correct output mean in practice?
More natural human proportions, realistic hands and faces, and correct body poses. You'll spend less time retouching the anatomical errors common in generated images.
Is Recraft V3 a multimodal LLM?
No. Recraft V3 is a dedicated image generation model. It doesn't have a conversational context or text completion interface. Use
generateImagefrom the AI SDK.How do I use Recraft V3 on AI Gateway?
Pass
recraft/recraft-v3togenerateImagefrom the AI SDK. AI Gateway routes the request to available providers automatically.What does Recraft V3 cost?
This page lists the current rates. Multiple providers can serve Recraft V3, so AI Gateway surfaces live pricing rather than a single fixed figure.
When should I choose V3 over V4?
Choose V3 when style presets and vector output are central to your workflow and you don't need V4's photorealism improvements. If natural skin textures, clean lighting, and fewer synthetic artifacts matter more, evaluate V4.