# Image-to-Sketch Generation

The image-to-sketch generation module provides functionality to convert raster images into sketch-style outputs, including both raster edge maps and vector sketches.

**Source**: `tests/integration/test_image2sketch.py`


## Available Methods

The module currently supports the following methods:

```{eval-rst}
.. autosummary::
   :toctree: generated

   sketchkit.image2sketch.HEDModel
   sketchkit.image2sketch.PhotoSketchModel
   sketchkit.image2sketch.SwiftSketchModel
   sketchkit.image2sketch.AniLinesModel
```

### HED

The method implements [Holistically-Nested Edge Detection](https://arxiv.org/abs/1504.06375) from *ICCV 2015*. It predicts dense edge maps from photos and optionally applies post-processing (non-maximum suppression and thinning) for cleaner sketch-like lines.

**Output:** Returns a grayscale `PIL.Image` (`"L"` mode). In the example below, the saved file is `angel_hed.png`.

### PhotoSketch

The method implements [Photo-Sketching: Inferring Contour Drawings from Images](https://arxiv.org/abs/1901.00542) from *WACV 2019*. It uses a Pix2Pix generator to map natural photos to contour-like sketch images.

**Output:** Returns a grayscale `PIL.Image` (`"L"` mode). In the example below, the saved file is `angel_photosketch.png`.

### SwiftSketch

The method implements [SwiftSketch: A Diffusion Model for Image-to-Vector Sketch Generation](https://arxiv.org/abs/2502.08642) from *SIGGRAPH 2025*. It predicts Bézier control points and converts them into SketchKit's vector `Sketch` representation.

**Output:** Returns a `Sketch` object. In the example below, it is rendered and saved as `angel_swiftsketch.png`, and can also be exported to SVG.

### AniLines

The method integrates [AniLines](https://github.com/style2paints/AniLines-Anime-Line-Extractor), an anime-style line extractor. It processes raster images to predict high-quality continuous line art for anime illustrations. It operates primarily in two modes: `basic` and `detail`.

**Output:** Returns a grayscale `PIL.Image` (`"L"` mode). In the example below, the saved file is `angel_anilines.png`.



## Code

The main interface is the `SketchGenerator` class.

```python
from sketchkit.image2sketch import SketchGenerator
from sketchkit.renderer import CairoRenderer, CairoRenderOptions
from PIL import Image

# Load an input image
input_image = Image.open("angel.png").convert("RGB")

# 1) HED: raster edge map
hed_generator = SketchGenerator(method="HED")
hed_result = hed_generator.run(input_image, size=512)
hed_result.save("angel_hed.png")

# 2) PhotoSketch: raster contour sketch
photosketch_generator = SketchGenerator(method="PhotoSketch")
photosketch_result = photosketch_generator.run(input_image, size=512)
photosketch_result.save("angel_photosketch.png")

# 3) SwiftSketch: vector sketch
swiftsketch_generator = SketchGenerator(method="SwiftSketch")
sketch = swiftsketch_generator.run(input_image, size=512)

render_options = CairoRenderOptions(canvas_size=(512, 512))
renderer = CairoRenderer(render_options=render_options)
rendered_image = renderer.render(sketch, render_options=render_options)
rendered_image.save("angel_swiftsketch.png")

# Optional: save vector output
sketch.to_svg("angel_swiftsketch.svg")


# 4) AniLines: anime-style line extraction
anilines_generator = SketchGenerator(method="AniLines")
anilines_result = anilines_generator.run(input_image, size=512)
anilines_result.save("angel_anilines.png")
```

## Usage Notes

**Constructor**

- `SketchGenerator(method=...)`
  - `method`: backend name. Supported values are `"HED"`, `"PhotoSketch"`, `"SwiftSketch"`, and `"AniLines"` (case-insensitive).

**Run Method**

- `run(input, size=None)`
  - `input`: accepts a `PIL.Image`, a `numpy.ndarray`, or an image path.
  - `size`: accepts an integer (square output) or a `(width, height)` tuple. If omitted, output keeps the input image size.