agentsclimarketplace

Daftai image to svg

Skill daftAI2026/daftAI-skills/skills/daftai-image-to-svg

Install
npx -y skills add daftAI2026/daftAI-skills --skill daftai-image-to-svg

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Converts transparent or solid-background single-color PNG files and SVG files containing one embedded or local PNG into verified path-only SVG output. Removes pure-color backgrounds automatically. Uses Potrace threshold and parameter scans with alpha error metrics to preserve detailed logos, stamps, engravings, and line art. Use when the user asks to convert images to SVG, vectorize PNG, convert bitmap to true SVG, remove embedded raster data from SVG, 图片转矢量, PNG 转 SVG, 真矢量, or 位图描摹.

SKILL.md

6.6 KB, as published. Nobody here has run it

Image to SVG

Convert transparent or solid-background single-color artwork into a genuine SVG containing paths rather than an embedded bitmap.

Automatic tracing approximates the raster boundary. Only the original vector source can provide mathematically exact curves.

Scope

Supported:

  • Transparent PNG.
  • PNG on a pure-color background, including white, red, green, or other flat colors.
  • SVG containing exactly one Base64 or local PNG.
  • Single-color logos, stamps, engravings, silhouettes, and detailed line art.

Reject:

  • Photos, gradients, textured backgrounds, and multicolor artwork.
  • Remote image URLs.
  • SVG mixing raster artwork with existing paths.
  • Centerline or editable-stroke reconstruction.

For multicolor artwork, recommend VTracer instead of forcing this workflow.

Script Directory

Important: All scripts are located in the scripts/ subdirectory of this skill.

Agent Execution Instructions:

  1. Determine this SKILL.md file's directory path as SKILL_DIR.
  2. Set command working directory to ${SKILL_DIR}.
  3. Execute bun ${SKILL_DIR}/scripts/vectorize.ts.
  4. Replace ${SKILL_DIR} with the absolute path.
ScriptPurpose
scripts/vectorize.tsExtract alpha, scan Potrace candidates, measure error, and deliver the best path-only SVG
scripts/vectorize.test.tsUnit and real-toolchain regression tests

Dependencies

Detect the runtime once per session:

command -v bun
command -v ffmpeg
command -v ffprobe
command -v potrace
command -v rsvg-convert || command -v inkscape

Required:

  • Bun.
  • FFmpeg and FFprobe.
  • Potrace.
  • One SVG renderer: rsvg-convert from librsvg, or Inkscape.

If a dependency is missing, install it directly with the available package manager. Report only when installation fails or no package manager is available.

Common macOS installation:

brew install oven-sh/bun/bun
brew install ffmpeg potrace librsvg

If Homebrew is unavailable, install with the platform package manager. Do not ask before installing missing conversion dependencies.

Workflow

  1. Confirm the input is within scope.

  2. Run the converter:

    bun ${SKILL_DIR}/scripts/vectorize.ts "/absolute/path/to/input.svg"
    
  3. Read <source-slug>.report.json.

  4. Confirm:

    • structure.pathCount >= 1;
    • structure.imageCount === 0;
    • structure.dataImageCount === 0;
    • verified4x === true.
    • backgroundRemoval.applied === true only when the input had a solid background.
  5. Inspect the PNG preview, especially fine lines, holes, hatching, and sharp corners.

  6. Report the SVG, preview, selected threshold, MAE, RMSE, and IoU.

Choose the objective explicitly:

  • Use default mae for the closest alpha appearance.
  • Use iou for overlap, contour match, 轮廓重合率, or IoU requests.
  • Use rmse only when large alpha errors matter more than average error.

Never describe the result as lossless. Say “high-fidelity true vector” and include measured error.

Usage

Specify an output parent directory:

bun ${SKILL_DIR}/scripts/vectorize.ts input.png --output-dir ./output

Override the detected color:

bun ${SKILL_DIR}/scripts/vectorize.ts input.svg --color "#0113a8"

Use solid-background PNG directly; do not preprocess manually:

bun ${SKILL_DIR}/scripts/vectorize.ts "logo-on-white.png"

Preserve all threshold candidates for manual comparison:

bun ${SKILL_DIR}/scripts/vectorize.ts input.svg --keep-candidates

Temporary candidates are deleted by default. Keep them only when tuning, comparing versions, or resolving visual review disputes.

Advanced threshold control:

bun ${SKILL_DIR}/scripts/vectorize.ts input.svg \
  --threshold-min 120 \
  --threshold-max 140 \
  --threshold-step 2

Validated defaults are 124..144, step 2, Potrace alphamax=0.2, and optimization tolerance 0.03.

Optimize for binary outline overlap and scan Potrace smoothing:

bun ${SKILL_DIR}/scripts/vectorize.ts input.svg \
  --objective iou \
  --threshold-min 122 \
  --threshold-max 132 \
  --threshold-step 1 \
  --alphamax 0,0.2 \
  --opttolerance 0,0.03

For detailed hard-edge art, alphamax=0 can raise IoU by avoiding curve smoothing. Do not hard-code color during experiments; inherit wrapper fill or use --color.

Output

Single-file runs write directly into the chosen output directory:

<output-dir>/
├── <source-slug>.svg
├── <source-slug>.preview.png
└── <source-slug>.report.json

Default runs do not create candidates/. With --keep-candidates, also write all measured candidates:

<output-dir>/
├── <source-slug>.svg
├── <source-slug>.preview.png
├── <source-slug>.report.json
└── candidates/
    ├── candidate-t128-a0-o0.svg
    ├── candidate-t128-a0-o0.png
    └── ...

If --output-dir is omitted, create <source-slug>-svg-output/ next to the input. Existing results are never overwritten; repeated runs append a timestamp to the output filename prefix. Do not copy the source file by default; the report records sourcePath.

For multiple input files, create one subdirectory per source under the chosen output parent to avoid collisions:

<output-parent>/
├── first-image/
│   ├── first-image.svg
│   ├── first-image.preview.png
│   └── first-image.report.json
└── second-image/
    ├── second-image.svg
    ├── second-image.preview.png
    └── second-image.report.json

The report records backgroundRemoval when a pure-color background was converted to alpha. Intermediate PGM files remain temporary; candidate SVG/PNG files are preserved only with --keep-candidates.

Validation

Run:

bun test ${SKILL_DIR}/scripts/vectorize.test.ts

For final delivery, also validate the generated SVG:

xmllint --noout "/path/to/result.svg"
rg -o '<path\b|<image\b|data:image/' "/path/to/result.svg" | sort | uniq -c

The final SVG must contain paths and no raster references.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.