mcpField
G
generate-svg

generate-svg

github.com/b-open-ioCreative & media1254 installsUpdated 5w ago
Repository

About generate-svg

A skill that uses Quiver AI's Arrow `arrow-preview` to create SVG graphics from text or vectorize images. It is for people creating graphics, logos, icons, and illustrations, with options for style guidance and output settings.

Key features

  1. 01Generate SVGs from text
  2. 02Vectorize images into SVG
  3. 03Accept style guidance and references
  4. 04Set cropping and target size

Use cases

  1. 01Create an icon or logo from a description
  2. 02Convert an existing bitmap to SVG
  3. 03Generate multiple SVG design variants

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/b-open-io/gemskills --skill generate-svg

After installing, type/generate-svgin Claude Code to confirm it loaded.

Bundled files.clawnet/unsigned-skill.jsonscripts/generate.ts

Details

From the project README

Generate SVG

Generate SVG graphics using Arrow arrow-preview by Quiver AI. Supports both text-to-SVG generation and image-to-SVG vectorization.

When to Use

Use this skill when the user asks to:

  • Create SVG graphics, logos, icons, or illustrations from a text description
  • Vectorize / convert an existing image to SVG

API Key Required

Set QUIVERAI_API_KEY or QUIVER_API_KEY in your environment. Get a key at https://quiver.ai.

Usage

bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts "prompt" [options]
Text-to-SVG Options
FlagTypeDefaultRangeDescription
--instructions <text>string——Additional style or formatting guidance
--references <url>string—up to 4Reference image URL or base64 (repeat flag for multiple)
--count <n>integer11–16Number of SVGs to generate
--temperature <n>float10–2Sampling temperature (lower = more deterministic)
--top-p <n>float10–1Nucleus sampling probability
--presence-penalty <n>float0-2–2Penalty for tokens already in prior output
--max-tokens <n>integer—1–131072Max output tokens (use 131072 for highest detail)
--output <path>stringoutput.svg—Output file path
Image-to-SVG (Vectorize) Options
FlagTypeDefaultRangeDescription
--vectorizebooleanfalse—Enable vectorization mode
--image <url|base64>string——Source image (required in vectorize mode)
--auto-cropbooleanfalse—Auto-crop to dominant subject before vectorization
--target-size <n>integer—128–4096Square resize target in pixels before inference
--count, --temperature, --top-p, --presence-penalty, --max-tokens, --output———Same as text-to-SVG
Examples
# Simple SVG generation
bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts "minimalist mountain logo"

# High-detail render (max tokens)
bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts "photorealistic bee" --max-tokens 131072

# With style instructions
bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts "geometric pattern" --instructions "Use only blue and green colors, bold strokes"

# With reference images
bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts "logo in this style" \
  --references https://example.com/ref1.png \
  --references https://example.com/ref2.png

# Save to specific file
bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts "company logo" --output logo.svg

# Vectorize an image URL to SVG
bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts --vectorize --image https://example.com/photo.jpg --auto-crop

# Vectorize with high-res target
bun run --cwd ${CLAUDE_PLUGIN_ROOT} ${CLAUDE_PLUGIN_ROOT}/skills/generate-svg/scripts/generate.ts --vectorize --image https://example.com/photo.jpg --target-size 2048 --max-tokens 131072

Model

Uses Arrow arrow-preview (Quiver AI) for SVG generation. Override with SVG_MODEL env var.

Cost: 1 credit per request (regardless of --count). Rate limit: 20 requests per 60 seconds.

Last verified: March 2026.

Comments

Markdown supported