quiverai

// Generate and vectorize SVG graphics via the QuiverAI API (Arrow model). Use when the user asks to create logos, icons, or illustrations as SVG, convert raster images (PNG/JPEG/WebP) to SVG, or generate vector graphics from text prompts.

$ git log --oneline --stat

stars:1,933

forks:367

updated:March 4, 2026

SKILL.mdreadonly

SKILL.md Frontmatter

namequiverai

descriptionGenerate and vectorize SVG graphics via the QuiverAI API (Arrow model). Use when the user asks to create logos, icons, or illustrations as SVG, convert raster images (PNG/JPEG/WebP) to SVG, or generate vector graphics from text prompts.

metadata[object Object]

QuiverAI — AI Vector Graphics

QuiverAI generates production-ready SVGs from text prompts or raster images.

Site: https://quiver.ai
Docs: https://docs.quiver.ai
API base: https://api.quiver.ai/v1
Model: arrow-preview
Auth: Bearer token via QUIVERAI_API_KEY
Billing: 1 credit per request (regardless of n).

Setup

Get an API key at https://app.quiver.ai/settings/api-keys (create account at https://quiver.ai/start first).

Text to SVG

Generate SVGs from a text description.

Endpoint: POST /v1/svgs/generations

curl -X POST https://api.quiver.ai/v1/svgs/generations \
  -H "Authorization: Bearer $QUIVERAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "arrow-preview",
    "prompt": "A minimalist monogram logo using the letter Q",
    "n": 1,
    "stream": false
  }'

Node.js SDK (npm install @quiverai/sdk):

import { QuiverAI } from "@quiverai/sdk";
const client = new QuiverAI({ bearerAuth: process.env.QUIVERAI_API_KEY });

const result = await client.createSVGs.generateSVG({
  model: "arrow-preview",
  prompt: "A minimalist monogram logo using the letter Q",
});
// result.data[0].svg contains the SVG markup

Parameters

Param	Type	Default	Description
`model`	string	—	Required. Use `arrow-preview`.
`prompt`	string	—	Required. Describes the desired SVG.
`instructions`	string	—	Additional style guidance (e.g. "flat monochrome, rounded corners").
`references`	array	—	Up to 4 reference images (`{ url }` or `{ base64 }`).
`n`	int	1	Number of outputs (1–16).
`temperature`	float	1	Sampling temperature (0–2). Lower = more deterministic.
`top_p`	float	1	Nucleus sampling (0–1).
`max_output_tokens`	int	—	Upper bound for output tokens (max 131072).
`stream`	bool	false	SSE streaming (events: `reasoning`, `draft`, `content`).

Response

{
  "id": "resp_01J...",
  "created": 1704067200,
  "data": [{ "svg": "<svg ...>...</svg>", "mime_type": "image/svg+xml" }],
  "usage": { "total_tokens": 1640, "input_tokens": 1200, "output_tokens": 440 }
}

Image to SVG (Vectorize)

Convert a raster image (PNG/JPEG/WebP) into SVG.

Endpoint: POST /v1/svgs/vectorizations

curl -X POST https://api.quiver.ai/v1/svgs/vectorizations \
  -H "Authorization: Bearer $QUIVERAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "arrow-preview",
    "stream": false,
    "image": { "url": "https://example.com/logo.png" }
  }'

SDK:

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

Additional parameters (beyond Text-to-SVG shared ones)

Param	Type	Default	Description
`image`	object	—	Required. `{ url: "..." }` or `{ base64: "..." }`.
`auto_crop`	bool	false	Crop to dominant subject before vectorization.
`target_size`	int	—	Square resize target in px (128–4096) before inference.

Response format is identical to Text-to-SVG.

Error codes

Status	Code	Meaning
400	`invalid_request`	Malformed body or missing fields.
401	`unauthorized`	Bad or missing API key.
402	`insufficient_credits`	Out of credits.
429	`rate_limit_exceeded`	Too many requests; back off and retry.

Tips

Save SVG output to a .svg file for immediate use.
Use instructions to control style without changing the prompt.
For logos, try low temperature (0.3–0.5) for cleaner, more consistent results.
Use references to provide visual examples the model should match.
For vectorization, enable auto_crop: true when the source image has excess whitespace.