Model Details
Recraft V3 (Image to Image) takes an existing image plus a text prompt and re-renders it in any of a large library of professional, named design styles. Provide the source image with `image_url`, describe what to change in `prompt`, and pick a look with `style`; `strength` controls how far the result departs from the original (0 stays close, 1 transforms heavily). Its standout strength is the same deep style catalog as the text-to-image variant — photoreal looks, digital illustration, and vector aesthetics — now applied as a restyle/transform of your own image rather than a from-scratch generation. When you pick a `vector_illustration/*` style the result is a true vector (SVG) file rather than a raster image, which makes it well suited to turning a picture into a clean, scalable logo or icon. Default raster styles (`realistic_image`, `digital_illustration`) return a standard webp image.
## Best for - Restyling a photo or illustration into a specific art style (watercolor, line art, pop art, flat vector) while keeping its composition - Converting an existing raster image into a scalable vector (SVG) logo or icon — pick a `vector_illustration/*` style - Applying one consistent, deliberately chosen `style` across a whole set of source images for a coherent brand look - Photoreal cleanups and look changes that stay anchored to the input image via a low-to-mid `strength`
## Choose another model when - You have no source image and want to generate purely from text — use the Recraft V3 text-to-image variant - You need precise, mask-based inpainting or object removal rather than a whole-image restyle — use an inpaint/eraser model - You need motion or animation — use an image-to-video model - You need many variations in a single call — this model returns one image per request
## Tips - The two biggest levers are `style` and `strength`: keep `strength` low (0.2–0.4) to preserve the original's layout and identity, raise it (0.6–0.8) for a stronger stylistic transformation - Browse the style families (`realistic_image`, `digital_illustration`, `vector_illustration`) and their substyles, e.g. `vector_illustration/line_art` for crisp line icons or `digital_illustration/pop_art` for a poster look - Choose any `vector_illustration/*` style when you specifically want an SVG you can scale and recolor; other style families return raster images - Use `negative_prompt` to steer away from unwanted elements when the restyle drifts
## Limitations - Returns a single image per request; there is no batch or variation count - No seed control, so exact runs are not reproducible - Vector styles return SVG and bill at double the standard rate - Fine, precisely spelled text and exact object counts can still be imperfect
To run via the ModelRunner JavaScript client: ```js import { modelrunner } from "@modelrunner/client";
const result = await modelrunner.subscribe("recraft/v3/image-to-image", { input: { image_url: "https://media.modelrunner.ai/qpVDYUNXosDmWAFh-recraft-i2i-probe-source-coffee-cup.png", prompt: "restyle as a warm watercolor illustration", strength: 0.5, style: "digital_illustration/hand_drawn", }, }); ```



