# Photo to Painting

> Free in-browser tool that turns a photo into a painting: oil paint, watercolour, brush strokes, pointillism,
> pencil sketch, coloured pencil and ink & wash, plus neural style-transfer styles. Nothing is uploaded.

URL: https://photo-to-painting.skillsafe.ai/

## What you get
- Seven painterly styles computed on the device; live thumbnails of every style on your own photo.
- Neural styles (fast neural style networks, 6.7 MB each, SHA-256 verified, cached) run with WebGPU or WebAssembly
  when their weights are available on SkillSafe's model host.
- Strength slider (blend with the original photo), brush size, texture, pattern size for neural styles.
- Borrow the colour palette of a public-domain painting or of any reference image (Oklab statistics transfer).
- Before/after divider and a hold-to-see-the-photo button.
- Full-resolution JPEG/PNG export rendered in tiles (up to 16 megapixels).
- Free, no account, no watermark. Photos never leave the browser.

## How it works
- Painterly styles: smoothed Kuwahara filter via summed-area tables, then per-style mark-making. All noise is keyed
  to absolute pixel coordinates so tiles join without seams and the full-size export matches the preview.
- Neural styles: Johnson-style feed-forward transformer networks with instance normalisation, fixed 224 x 224 input;
  the photo is processed in tiles overlapping by 32 px and blended with linear ramp weights.

## Credits and licences
- Neural style weights: ONNX Model Zoo fast_neural_style (Apache-2.0), from PyTorch examples (BSD-3-Clause).
- Runtime: onnxruntime-web (MIT).
- Sample photos: CC0 (Wikimedia Commons). Palette paintings: public domain.