Blanket

Image processing in Python. Pixel work in Rust.

Blanket is a Rust-backed image library with a familiar, Pillow-shaped Python API. Open, transform, and save images using native pixel kernels, with no Pillow or other Python packages required at runtime.

from blanket import Image, ImageOps

with Image.open("photo.jpg") as image:
    upright = ImageOps.exif_transpose(image)
    thumbnail = ImageOps.fit(upright, (256, 256), method=Image.Resampling.LANCZOS)
    thumbnail.save("thumbnail.webp", quality=85)

Installation · Examples · Formats · API notes · Contributing

Why Blanket?

• Familiar image APIs. Work with Image, ImageOps, ImageChops, ImageEnhance, ImageFilter, ImageMath, ImagePalette, and ImageStat.
• Native processing. Rust kernels use SIMD and parallel execution for supported operations, with small-image fast paths.
• Modern formats. Read and write JPEG XL, WebP, AVIF, and HEIC/HEIF alongside PNG, JPEG, TIFF, BMP, GIF, and ICO. Export single-page PDFs.
• Explicit precision. Keep 10-, 12-, and 16-bit samples in supported workflows, and choose when to convert to 8 bits.
• Python interoperability. Import array data, work with binary streams, and convert explicitly to Pillow when you need it.

Blanket is currently alpha software. It implements a focused subset of Pillow's API, centered on images in the 1, L, LA, RGB, RGBA, HSV, CMYK, YCbCr, and LAB modes. It also supports integer I and I;16 modes, floating-point F, and indexed P and PA images to a limited extent. It is not yet a drop-in replacement for PIL.

See ImageChops for arithmetic and blend modes, and ImageStat for per-band statistics.

Installation

Blanket requires Python 3.12+. Install it with pip:

pip install pyblanket

See the development notes for information on building from source.

Examples

Convert between formats

from blanket import Image

with Image.open("input.png") as image:
    image.convert("RGB").save("output.jpg", quality=85)
    image.save("output.jxl", lossless=True)

File extensions select the output format. For binary streams, pass it explicitly:

from io import BytesIO
from blanket import Image

with Image.open("input.png") as image:
    buffer = BytesIO()
    image.save(buffer, format="PNG")
    png_bytes = buffer.getvalue()

Optimize a save without additional loss

from blanket import Image
from blanket.Compressor import LosslessImageCompressor

compressor = LosslessImageCompressor(effort=7)
with Image.open("input.png") as image:
    image.save("optimized.png", compressor=compressor)
    image.convert("RGB").save("optimized.jpg", quality=90, compressor=compressor)
    image.save("optimized.jxl", lossless=True, compressor=compressor)
    image.save("optimized.heic", lossless=True, compressor=compressor)

The compressor keeps the smallest encoding that preserves the normal save's decoded pixels. It supports PNG, JPEG, JPEG XL, and HEIF/HEIC; higher effort tries more settings and takes longer. JPEG optimization preserves DCT coefficients. This optimizes the requested save, not the original source file: lossy save settings still introduce their normal loss. See the Compressor module for the API, format-specific behavior, and compression benchmarks.

For smaller output with bounded additional pixel error, pass LossyImageCompressor(max_rmse=2.0, effort=7) from blanket.Compressor. It searches PNG color precision or lower JPEG, WebP, JPEG XL, AVIF, and HEIF/HEIC quality settings, preserving alpha and checking decoded RGB error against the normal save. The result never exceeds the normal save's size.

Resize, enhance, and filter

from blanket import Image, ImageEnhance, ImageFilter, ImageOps

with Image.open("photo.jpg") as image:
    resized = ImageOps.contain(image, (1200, 800))
    enhanced = ImageEnhance.Contrast(resized).enhance(1.2)
    sharpened = enhanced.filter(ImageFilter.UnsharpMask(radius=2, percent=150))
    sharpened.save("edited.png")

resize() supports all six Image.Resampling filters. ImageOps includes cropping, padding, flips, color adjustments, and EXIF orientation correction. ImageFilter provides blur, convolution, and rank filters, as well as 3D color LUTs.

Build an image with transparency

from blanket import Image

background = Image.new("RGBA", (640, 480), "white")
overlay = Image.new("RGBA", (160, 160), (30, 100, 220, 128))
background.alpha_composite(overlay, dest=(40, 40))
background.save("composite.png")

Use paste() for masked placement, putalpha() to set transparency, and split() / Image.merge() to work with individual channels.

Work with NumPy and high-bit-depth images

NumPy is optional. fromarray() accepts array-interface objects, including strided arrays:

import numpy as np
from blanket import Image

samples = np.full((64, 64, 3), 713, dtype=np.uint16)
image = Image.fromarray(samples, bit_depth=10)
image.save("exact.png")
image.convert("RGB", bit_depth=8).save("preview.jpg")

image.bit_depth reports sample precision. Copying, conversion, pixel access, splitting, cropping, transposition, and resizing preserve high-bit-depth samples. Other processing operations require conversion to 8 bits. See high-bit-depth images for storage and format-specific behavior.

Supported formats

Use from blanket import features to inspect the installed build: features.check("heif_decoder") reports runtime codec availability, features.version("jxl") reports its backend version, and features.blanketinfo() prints backend and format diagnostics. See the features module for feature names.

Encoder settings and their defaults are listed in the format guide. HEIF lossless compression can still change RGB values during RGB/YUV conversion.

Compatibility and scope

• Image modes: 8-bit processing supports 1, L, LA, RGB, RGBA, HSV, CMYK, YCbCr, and LAB. Integer I, I;16, I;16L, and I;16B and floating-point F support pixel access, conversion, copying, cropping, transposition, and resizing. F also supports transforms and TIFF interchange. Indexed P and PA images have a smaller supported operation set.
• Print and luma/chroma modes: CMYK uses raw channels without ICC transforms; JPEG and 8-bit TIFF retain CMYK on read, and JPEG, TIFF, and PDF support CMYK output. CMYK save-time compression optimization is unsupported. YCbCr saves to JPEG through RGB and opens back as RGB. Convert either mode to RGB for other output formats or direct quantization.
• LAB: 8-bit CIE Lab* supports pixel operations and TIFF interchange. Native, statically linked LittleCMS converts between sRGB and D50 LAB; other conversions pass through RGB. LAB save-time compression optimization is unsupported.
• Metadata: PNG, JPEG, and uncompressed JPEG XL metadata boxes supply EXIF/XMP for orientation handling. Saving writes pixels only; metadata is not preserved.
• Multiple frames: animation and multipage editing are not supported.
• High bit depth: sample retention is supported for selected operations and formats; HDR tone mapping and HEIF HDR/ICC metadata retention are not.
• Pillow interop: image.to_pillow() creates a Pillow image when Pillow is installed. High-bit-depth images must first be converted to 8 bits.

See the usage and API notes for operation-specific restrictions, palette support, and interoperability examples.

Performance

The benchmark suite compares Blanket and Pillow using equal in-memory inputs and eager decoding. After completing the development setup below, run it with a release build:

maturin develop -r --extras test --uv
uv run --no-sync scripts/benchmark.py --sizes web

Use --all for the full suite or --sections Resize ImageOps to select operations. Results depend on image size, operation, codec settings, and hardware. The benchmark guide explains baseline comparisons and reproducible runs.

Contributing

Source builds require the Rust nightly pinned in rust-toolchain.toml for portable SIMD.

With the build prerequisites installed, set up a development environment and run the checks:

uv sync --no-install-project --extra test
uv tool install maturin
maturin develop --extras test --uv
cargo fmt --check
cargo clippy -- -D warnings
cargo test
uv run --no-sync pytest -q
uv run --no-sync scripts/verify_behavior.py

The native core lives in crates/, the Python API in python/blanket/, and compatibility tests in tests/. Include a focused regression test when fixing a bug or adding a feature. To report a problem, open an issue.

License

Blanket is licensed under the MIT License. See third-party licenses for native dependency notices.