image-to-svg

Image to SVG

Vectorize an image with QuiverAI POST /v1/svgs/vectorizations, then write the returned SVG markup to a file.

Always fetch the current docs first. Never pipe through head, tail, sed -n, awk, | less, or any command that truncates:

curl -sL https://docs.quiver.ai/api/models/image-to-svg
curl -sL https://docs.quiver.ai/api-reference/vectorize-svg/vectorizesvg

API key lives in ~/.config/opencode/skills/image-to-svg/.env as QUIVERAI_API_KEY. Load it. Never print the key. Never commit .env. If the value is empty, stop and ask the user to paste a key from https://platform.quiver.ai/api-keys

set -a
source ~/.config/opencode/skills/image-to-svg/.env
set +a
test -n "$QUIVERAI_API_KEY"

Convert

Default model is arrow-1.1 (logos, icons, most images). Use arrow-1.1-max for dense illustrations or technical drawings.

Keep stream false. The JSON body has the full SVG. Do not use SSE.

Crop the subject tightly when you can. Set auto_crop: true only when the image has extra empty space you cannot crop yourself.

Public URL

OUT=/tmp/vectorized.svg

curl -sS --fail-with-body -X POST 'https://api.quiver.ai/v1/svgs/vectorizations' \
  -H "Authorization: Bearer $QUIVERAI_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "arrow-1.1",
    "auto_crop": true,
    "image": { "url": "https://example.com/logo.png" }
  }' \
  | jq -r '.data[0].svg' > "$OUT"

test -s "$OUT"

Local file (PNG, JPEG, WebP, GIF, SVG). Send raw base64, not a data URI. Decoded input must stay under 12,582,912 bytes, 4096x4096, and 16,777,216 pixels.

SRC=/path/to/image.png
OUT=/tmp/vectorized.svg
B64=$(base64 -i "$SRC" | tr -d '\n')

jq -n --arg b64 "$B64" '{
  model: "arrow-1.1",
  auto_crop: true,
  image: { base64: $b64 }
}' | curl -sS --fail-with-body -X POST 'https://api.quiver.ai/v1/svgs/vectorizations' \
  -H "Authorization: Bearer $QUIVERAI_API_KEY" \
  -H 'Content-Type: application/json' \
  -d @- \
  | jq -r '.data[0].svg' > "$OUT"

test -s "$OUT"

Save next to the source when the user did not name an output path: /path/to/image.svg from /path/to/image.png. Otherwise use the path they gave. Always confirm the file exists and starts with <svg.

Logos for websites

After vectorization, rewrite logos and wordmarks so they drop into sites without extra CSS hacks:

1. Transparent background. Delete full-canvas white/black <rect> fills, fill on <svg>, and any path that is only the backdrop. The SVG must have no opaque background.
2. currentColor for the mark. Replace the logo fill and stroke (#000, #000000, black, and the dominant ink color) with currentColor. Sites can then set color in CSS for light and dark mode.
3. Do this when the input is a logo, icon, or wordmark, or when the user asks for website/asset use. Skip it for full-color illustrations where hue is part of the art.

Do the rewrite on the saved SVG before you upload it. Do not leave hardcoded black or white on a logo meant for a site.

Response

Success JSON (non-stream):

{
  "id": "resp_...",
  "created": 1704067200,
  "data": [{ "svg": "<svg ...>", "mime_type": "image/svg+xml" }],
  "credits": 1
}

Write only .data[0].svg to the file. Do not write the wrapper JSON.

A sk_test_... key returns a mock. Reject it for real work if the SVG has data-quiver-sandbox="true" or the response header is x-quiver-environment: test.

On HTTP errors, show status, code, message, and request_id. For 429, wait Retry-After seconds then retry once.

Discord

After a successful save, upload the SVG with kimaki upload-to-discord --session <current-session> "$OUT" so the user can open the file. Do not paste the full SVG markup into chat.