jxl-encoder

Pure Rust JPEG XL encoder. Lossy (VarDCT) and lossless (Modular) encoding, verified against three independent decoders (jxl-rs, jxl-oxide, djxl). #![forbid(unsafe_code)].

740+ tests passing.

Quick start

use jxl_encoder::{LossyConfig, LosslessConfig, PixelLayout};

// Lossy — distance 1.0 is visually lossless
let jxl = LossyConfig::new(1.0)
    .encode(&pixels, width, height, PixelLayout::Rgb8)?;

// Lossless
let jxl = LosslessConfig::new()
    .encode(&pixels, width, height, PixelLayout::Rgb8)?;

// Full control — limits, metadata, cancellation
use jxl_encoder::Limits;
let jxl = LossyConfig::new(1.0)
    .with_ans(true)
    .with_gaborish(true)
    .encode_request(width, height, PixelLayout::Rgba8)
    .with_limits(&Limits::default())
    .encode(&pixels)?;

Pixel layouts

Rgb8, Rgba8, Bgr8, Bgra8, Gray8, GrayAlpha8, Rgb16, Rgba16, Gray16, RgbLinearF32.

Lossy encoding supports all layouts including alpha (VarDCT for RGB + modular for the alpha channel). Lossless supports RGB, RGBA, grayscale, and gray+alpha.

What works

Lossy (VarDCT): 19/27 AC strategies (all that libjxl evaluates through effort 7), ANS entropy coding, adaptive quantization, chroma-from-luma, gaborish, pixel-domain loss, butteraugli quantization loop, custom coefficient ordering, noise synthesis, error diffusion, EPF sharpness, JPEG re-encoding.

Lossless (Modular): RCT (all 42 variants), ANS + Huffman, LZ77 (RLE + hash chain), histogram clustering, content-adaptive MA tree learning, palette transform, squeeze (Haar wavelet), 14/14 predictors including Weighted.

Animation: Both lossy and lossless, per-frame duration, loop count, frame crop detection.

Input formats: 8-bit sRGB, 16-bit sRGB, linear f32, grayscale, alpha. BGR/BGRA layouts.

Lossy quality vs libjxl: Within 3% of cjxl effort 5 at low distances (d <= 1.0). The gap widens to ~22-26% at higher distances due to missing cost model refinements (iterative rate control, full histogram clustering).

Features

Feature	Default	Description
`std`	yes	Standard library support; enables `encode_to()` for `Write` targets
`butteraugli-loop`	yes	Iterative quant field refinement via butteraugli distmap
`safe-mode`	yes	Guard flag (no behavioral effect; multi-group works correctly)
`rate-control`	no	Iterative encode for precise distance targeting
`jpeg-reencoding`	no	JPEG bitstream re-encoding into JXL
`trace-bitstream`	no	Zero-cost bitstream tracing for debugging

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.

Upstream

Upstream code from libjxl/libjxl is licensed under BSD-3-Clause.

Our additions and improvements are dual-licensed (AGPL-3.0 or commercial) as above.

Algorithms and constants derived from libjxl (BSD-3-Clause).