Image to SVG

Convert image inputs into production-ready SVG outputs.

Convert an image input into SVG via POST /v1/svgs/vectorizations.

Examples below use arrow-2. To discover model IDs available to your organization, call GET /v1/models. Arrow 2 handles most vectorization work; Arrow 2 Telos is better for detailed images where color, alignment, and fine structure need more fidelity.

Set stream: true to receive reasoning, draft, and content Server-Sent Events while the SVG is being produced.

Example

import { QuiverAI } from "@quiverai/sdk";

const client = new QuiverAI({
  bearerAuth: process.env["QUIVERAI_API_KEY"],
});

const result = await client.vectorizeSVG.vectorizeSVG({
  vectorizeSVGRequest: {
    model: "arrow-2",
    autoCrop: true,
    image: {
      url: "https://example.com/logo.png",
    },
  },
});

curl --request POST \
  --url https://api.quiver.ai/v1/svgs/vectorizations \
  --header 'Authorization: Bearer <QUIVERAI_API_KEY>' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "arrow-2",
    "auto_crop": true,
    "image": {
      "url": "https://example.com/logo.png"
    }
  }'

Preparing the image

The App Prompt guide explains how vectorization differs from generating a new design with a reference image and what to inspect in the result.

Cropping the input image tightly to the subject usually improves output quality. Use auto_crop as a fallback when you can’t crop manually.

Image inputs can be provided as { url: "..." }, { base64: "..." }, or URL string shorthand. Direct base64 image payloads can be PNG, JPEG, WebP, GIF, or SVG. Decoded image inputs must be no larger than 12,582,912 bytes, 4096 x 4096 pixels, or 16,777,216 total pixels.

Image URLs must use HTTP or HTTPS, resolve to a public network target, and return an image content type. The API follows up to 3 redirects and applies the same decoded image limits after fetching.

For logos, icons, and most source images, arrow-2 is the right starting point. For complex illustrations, technical drawings, and other dense inputs where smaller details need to stay stable, use arrow-2-telos.

Parameters

Parameter	Type	Default	Description
`model`	string	–	Required. Model identifier (for example, `arrow-2` or `arrow-2-telos`).
`image`	object \| string	–	Required. Input image as `{ url: "..." }`, `{ base64: "..." }`, or URL string shorthand.
`auto_crop`	boolean	`false`	Automatically crop the image to the dominant subject before vectorization.
`target_size`	integer	–	Square resize target in pixels (128 to 4096).
`stream`	boolean	`false`	When `true`, returns a Server-Sent Events stream with progressive rendering phases (`reasoning`, `draft`, `content`).
`temperature`	number	`1` for Arrow 2	Controls randomness (0 to 2). Lower values produce more deterministic output; higher values increase variety.
`top_p`	number	`1` for Arrow 2	Nucleus sampling (greater than 0 and at most 1). Lower values make output more focused.
`presence_penalty`	number	`0` for Arrow 2	Penalizes tokens already present in prior output (-2 to 2). Positive values encourage the model to explore new patterns.
`max_output_tokens`	integer	–	Upper bound for output token count (1 to 65536).

For native SVG vectorization, Arrow 2 supports temperature, top_p, and presence_penalty. Arrow 2 Telos does not support any of them. Omit all three keys when using arrow-2-telos; an explicit default value or null is rejected with HTTP 400. Check supported_sampling_parameters in GET /v1/models for the selected model. Responses requests have their own sampling rules.