zenjpeg

zenjpeg is a pure Rust JPEG encoder and decoder ported from Google's jpegli.

It produces smaller files at the same visual quality as libjpeg-turbo and mozjpeg through adaptive quantization, optional trellis optimization, and XYB perceptual color space support. The decoder is faster than both mozjpeg and zune-jpeg across all image sizes, with parallel decoding reaching 10x the throughput of mozjpeg on 4K images. #![forbid(unsafe_code)], no C dependencies.

Quick start

Encode

use zenjpeg::encoder::{EncoderConfig, ChromaSubsampling};

let config = EncoderConfig::ycbcr(85, ChromaSubsampling::Quarter)
    .progressive(true);

// One-shot encode from an RGB pixel buffer
let jpeg: Vec<u8> = config.encode(&rgb_pixels, width, height)?;

Decode

Requires the decoder feature:

[dependencies]
zenjpeg = { version = "0.6", features = ["decoder"] }

use zenjpeg::decoder::{Decoder, DecodeResult};

let result: DecodeResult = Decoder::new()
    .decode(&jpeg_data, enough::Unstoppable)?;
let pixels: &[u8] = result.pixels_u8().expect("u8 output");
let (width, height) = (result.width, result.height);

Streaming encode

For large images or memory-constrained environments, push rows incrementally:

use zenjpeg::encoder::{EncoderConfig, ChromaSubsampling, PixelLayout, Unstoppable};

let config = EncoderConfig::ycbcr(85, ChromaSubsampling::Quarter)
    .progressive(true);

let mut enc = config.encode_from_bytes(1920, 1080, PixelLayout::Rgb8Srgb)?;
for chunk in rgb_bytes.chunks(1920 * 3 * 16) {
    enc.push_packed(chunk, Unstoppable)?;
}
let jpeg: Vec<u8> = enc.finish()?;

Performance

Measured on a Ryzen 9 7950X (WSL2), default compiler target (no -C target-cpu=native), using noise+patches test patterns. Full benchmark tables and methodology are in the CLAUDE.md project guide.

Encoder

The encoder is ~15-20% slower than C++ jpegli (median 1.15-1.2x), with effectively identical output quality (size within +0.63%, DSSIM within +0.41%). auto_optimize(true) adds hybrid trellis quantization for +1.5 SSIMULACRA2 over C++ jpegli at matched bit rate.

Decoder (sequential, baseline 4:2:0)

Image size	vs mozjpeg	vs zune-jpeg	vs C++ jpegli
512x512	0.80x	0.86x	0.55x
2048x2048	0.94x	0.89x	0.61x
4096x4096	0.95x	0.91x	0.65x

Values < 1.0 mean zenjpeg is faster. mozjpeg here is libjpeg-turbo with NASM SIMD (C).

Decoder (progressive, sequential, no DRI)

Image size	vs mozjpeg	vs zune-jpeg
512x512	0.60x	0.55x
2048x2048	0.62x	0.57x
4096x4096	0.69x	0.63x

1.4-1.8x faster than mozjpeg, 1.6-1.8x faster than zune-jpeg on progressive files.

Decoder (parallel, --features parallel, baseline 4:2:0)

Image size	vs mozjpeg	parallel speedup
1024x1024	0.37x	2.6x
2048x2048	0.16x	5.8x
4096x4096	0.10x	9.5x

Fused parallel decode (entropy + IDCT + color convert per restart segment via rayon). Requires restart markers; progressive images show no parallel benefit.

Encoder features

Color modes

• YCbCr -- Standard JPEG. Maximum decoder compatibility.
• XYB -- Perceptual color space from JPEG XL. Better quality per byte, but requires a jpegli-aware decoder (libjxl, zenjpeg).
• Grayscale -- Single-channel output.

use zenjpeg::encoder::{EncoderConfig, ChromaSubsampling, XybSubsampling, Quality};

let ycbcr = EncoderConfig::ycbcr(85, ChromaSubsampling::Quarter);
let xyb   = EncoderConfig::xyb(85, XybSubsampling::BQuarter);
let gray  = EncoderConfig::grayscale(85);

// Quality scales: target a perceptual metric instead of an arbitrary number
let ssim  = EncoderConfig::ycbcr(Quality::ApproxSsim2(90.0), ChromaSubsampling::None);
let btrgl = EncoderConfig::ycbcr(Quality::ApproxButteraugli(1.0), ChromaSubsampling::Quarter);
let moz   = EncoderConfig::ycbcr(Quality::ApproxMozjpeg(80), ChromaSubsampling::Quarter);

Optimization presets

Eight presets covering the jpegli, mozjpeg, and hybrid optimization families:

Preset	Scan type	AQ	Trellis	Tables	Notes
JpegliBaseline	Baseline	Yes	No	3	Matches cjpegli --jpeg_type baseline
JpegliProgressive	Progressive	Yes	No	3	Default. Matches cjpegli output.
MozjpegBaseline	Baseline	No	Full	2	Matches C mozjpeg baseline profile
MozjpegProgressive	Progressive	No	Full	2	Matches C mozjpeg progressive profile
MozjpegMaxCompression	Progressive	No	Full	2	Matches JCP_MAX_COMPRESSION + scan search
HybridBaseline	Baseline	Yes	Adaptive	3	Best of both: jpegli AQ + trellis
HybridProgressive	Progressive	Yes	Adaptive	3	Best quality/size tradeoff
HybridMaxCompression	Progressive	Yes	Full	3	Slowest, smallest files

Or skip presets and use auto_optimize(true) for CMA-ES butteraugli-tuned parameters (equivalent to HybridMaxCompression with lambda 14.5):

let config = EncoderConfig::ycbcr(85, ChromaSubsampling::Quarter)
    .auto_optimize(true);

Chroma subsampling

4:4:4, 4:2:0, 4:2:2, and 4:4:0. Downsampling methods: Box, GammaAware, and SharpYUV (via the yuv feature, default).

Other encoder capabilities

• Progressive JPEG with jpegli, mozjpeg, and exhaustive scan scripts
• Streaming row-by-row API -- no whole-image buffering
• Parallel encoding via rayon (parallel feature)
• 19 pixel input layouts including RGB/BGR/RGBA/BGRA in 8-bit, 16-bit, and f32
• ICC profile embedding and EXIF metadata builder
• Per-image metadata via the three-layer pattern (EncoderConfig -> EncodeRequest -> encoder)
• Memory estimation -- estimate_memory() and estimate_memory_ceiling() before encoding
• Cancellation -- cooperative Stop token on all push methods

Decoder features

The decoder API is in prerelease. Enable with features = ["decoder"].

• Streaming decode -- single-pass for baseline, multi-pass for progressive
• Scanline reader -- ScanlineReader for row-by-row output with bounded memory
• 4 strictness levels -- Strict, Balanced, Lenient, Permissive
• Parallel decode -- wave-parallel and full-buffer modes via rayon (parallel feature)
• Fancy upsampling -- bilinear chroma upsampling (disable with fancy_upsampling(false) for ~5-10% speed gain)
• Crop regions -- decode a sub-rectangle without decoding the full image
• Auto-orient -- lossless DCT-domain rotation from EXIF orientation
• CMYK support and raw coefficient access (DecodedCoefficients)
• Metadata preservation -- EXIF, ICC, XMP, JFIF, Adobe, MPF segments
• Deblocking filters -- optional H.264-style / Knusperli post-processing
• Dequantization bias -- Laplacian bias for improved decode quality at low Q (matches C++ jpegli decoder behavior)
• Resource limits -- max_pixels, max_memory for DoS protection

Feature flags

Feature	Default	Description
decoder	No	JPEG decoding (enables zenjpeg::decoder module)
parallel	No	Multi-threaded encode/decode via rayon
trellis	Yes	Rate-distortion trellis quantization (mozjpeg-style Viterbi)
yuv	Yes	Fast SIMD RGB-to-YCbCr via the yuv crate (10-150x faster)
cms-lcms2	No	ICC color management via lcms2 (needed for XYB decode)
cms-moxcms	No	Pure Rust ICC color management (alternative to lcms2)
ultrahdr	No	UltraHDR HDR gain map encode/decode (implies decoder)
layout	No	Lossless transforms + decode-resize-encode pipeline

Common configurations

# Encode only (default features)
zenjpeg = "0.6"

# Decode + encode
zenjpeg = { version = "0.6", features = ["decoder"] }

# Server workload (parallel encode + decode)
zenjpeg = { version = "0.6", features = ["decoder", "parallel"] }

# Minimal (no trellis, no SharpYUV)
zenjpeg = { version = "0.6", default-features = false }

# UltraHDR
zenjpeg = { version = "0.6", features = ["ultrahdr"] }

UltraHDR

With features = ["ultrahdr"], zenjpeg can encode and decode HDR gain maps following the Ultra HDR specification. The codec handles JPEG assembly, MPF structure, and XMP metadata; the caller provides tonemapped SDR, gain map data, and ICC profiles.

Lossless transforms

With features = ["decoder"], DCT-domain rotate/flip operations are available with zero generation loss. The layout feature adds a decode-resize-encode pipeline integrating zenlayout and zenresize.

SIMD and platform support

SIMD dispatch is handled by archmage at runtime -- no compile-time target flags required. Supported instruction sets:

• x86_64: AVX2/FMA, AVX-512
• wasm32: SIMD128 (1.6-1.7x encode, 1.5-2.0x decode speedup)
• Scalar fallback on all other targets

CI tests on 6 OS targets: Ubuntu x86_64, Ubuntu ARM64, macOS ARM64, macOS Intel, Windows x86_64, and Windows ARM64.

Limitations

• The encoder is ~15-20% slower than C++ jpegli. The gap is mostly in the DCT where C++ Highway AVX-512 outperforms Rust AVX2. Other hot paths (AQ, entropy coding) are at parity.
• XYB mode produces 5-11% larger files than C++ jpegli at equivalent visual quality. The quality is identical; the size gap is under investigation.
• The decoder API is in prerelease and will have breaking changes.
• Progressive images do not benefit from parallel decoding (no restart segments to parallelize).
• NEON SIMD is not yet implemented. ARM targets use the scalar fallback.

Image tech I maintain

State of the art codecs*	zenjpeg · zenpng · zenwebp · zengif · zenavif (rav1d-safe · zenrav1e · zenavif-parse · zenavif-serialize) · zenjxl (jxl-encoder · zenjxl-decoder) · zentiff · zenbitmaps · heic · zenraw · zenpdf · ultrahdr · mozjpeg-rs · webpx
Compression	zenflate · zenzop
Processing	zenresize · zenfilters · zenquant · zenblend
Metrics	zensim · fast-ssim2 · butteraugli · resamplescope-rs · codec-eval · codec-corpus
Pixel types & color	zenpixels · zenpixels-convert · linear-srgb · garb
Pipeline	zenpipe · zencodec · zencodecs · zenlayout · zennode
ImageResizer	ImageResizer (C#) — 24M+ NuGet downloads across all packages
Imageflow	Image optimization engine (Rust) — .NET · node · go — 9M+ NuGet downloads across all packages
Imageflow Server	The fast, safe image server (Rust+C#) — 552K+ NuGet downloads, deployed by Fortune 500s and major brands

_{* as of 2026}

General Rust awesomeness

archmage · magetypes · enough · whereat · zenbench · cargo-copter

And other projects · GitHub @imazen · GitHub @lilith · lib.rs/~lilith · NuGet (over 30 million downloads / 87 packages)

License

Dual-licensed: AGPL-3.0 or commercial.

I've maintained and developed open-source image server software -- and the 40+ library ecosystem it depends on -- full-time since 2011. Fifteen years of continual maintenance, backwards compatibility, support, and the (very rare) security patch. That kind of stability requires sustainable funding, and dual-licensing is how we make it work without venture capital or rug-pulls. Support sustainable and secure software; swap patch tuesday for patch leap-year.

Our open-source products

Your options:

• Startup license -- $1 if your company has under $1M revenue and fewer than 5 employees. Get a key
• Commercial subscription -- Governed by the Imazen Site-wide Subscription License v1.1 or later. Apache 2.0-like terms, no source-sharing requirement. Sliding scale by company size. Pricing & 60-day free trial
• AGPL v3 -- Free and open. Share your source if you distribute.

See LICENSE-COMMERCIAL for details.