OxiMedia 0.1.6 Released — Stub Resolution, Codec Completeness, and a 9-Module EXR Refactor

The pure-Rust FFmpeg + OpenCV replacement just closed a wave of codec gaps and got faster spectral math.

Today we released OxiMedia 0.1.6 — a patent-free, memory-safe reconstruction of both FFmpeg (multimedia processing) and OpenCV (computer vision), unified into a single pure-Rust framework.

No C. No C++. No FFmpeg binaries. No OpenCV Python bindings. No pkg-config, no LD_LIBRARY_PATH, no brew install ffmpeg opencv. OxiMedia compiles to a single static binary (or to wasm32-unknown-unknown for the browser) and runs everywhere — laptops, edge servers, cloud, and the web — with zero system libraries and zero patent royalties.

Why OxiMedia 0.1.6 matters

For decades, professional media and computer-vision work meant living inside the massive C/C++ ecosystems of FFmpeg and OpenCV — with their complex build systems, segfault-prone manual memory management, heavy system dependencies, and patent royalties on common codecs (H.264, HEVC, AAC).

0.1.6 is a consolidation release: it turns scaffolding into working code and pays down internal debt. Concrete wins, all grounded in this cycle’s work:

• 13 stubs resolved. Color-conversion paths in oximedia-accel (RGB↔YCbCr, RGBA↔BGRA, NV12, HSV, linear↔sRGB) now produce real output; the Vorbis codebook VQ decode is wired into the active path; ACES Output Device Transform variants (P3-D65, Rec.709, Rec.2020, D60-sim, sRGB) emit correct results; the DASH segment HTTP fetch is no longer a placeholder; and system font directory scanning is implemented across macOS, Linux, and Windows.
• OxiFFT upgraded to 0.3.0. OxiFFT 0.3.0 delivers a Makhoul-reduction DCT-II (~4× faster than 0.2.0), plan caching for the R2r/R2c solvers, and hand-optimized AVX-512 codelets for sizes 16/32/64. All 13 FFT-dependent crates migrated cleanly with zero performance regressions on the spectral-flux and MIR benchmarks.
• 81,582 tests passing (0 failures, 245 skipped, cargo nextest run --workspace --all-features), with zero clippy warnings workspace-wide and a clean WASM check.

At 0.1.6 the workspace is 108 crates and ~2.69M lines of pure Rust — every crate marked stable.

Technical Deep Dive: how the layers fit

The framework is organized into clean functional layers across its 108 workspace crates:

1. Foundation (oximedia-core, oximedia-gpu, oximedia-simd, oximedia-accel) — wgpu compute shaders, hand-written SIMD kernels, an async Tokio job queue, zero-copy buffers, and the color-space conversion helpers filled in this cycle.

2. Codecs & Containers (oximedia-codec, oximedia-container) — royalty-free codecs only: AV1, VP9, VP8, Theora, Opus, Vorbis, FLAC, MP3, packed into MP4, MKV, MPEG-TS, and OGG. Streaming via HLS, DASH, RTMP, SRT, WebRTC, and SMPTE 2110.

3. Computer Vision & Effects (oximedia-cv, oximedia-effects, oximedia-image) — detection, tracking, denoising, stabilization, scene and shot detection, color management (ICC, ACES, HDR), professional image I/O (DPX, OpenEXR, TIFF), and quality metrics (PSNR, SSIM, VMAF).

4. Pipeline & Production — DAG filter graphs, live streaming, broadcast switching, timeline editing, packaging (CMAF, DRM/CENC), and media asset management.

Workspace-wide guarantees hold: unsafe_code = "deny", single static binary deployment, full WASM + WebGPU support, and zero C/Fortran dependencies in default features (the math comes from the pure-Rust OxiFFT and SciRS2).

Getting Started

cargo add oximedia

A minimal transcode-to-AV1 pipeline reads, encodes, and writes — all in pure Rust, no system FFmpeg required:

use oximedia::prelude::*;

fn main() -> oximedia::Result<()> {
    // Decode an input, re-encode video to royalty-free AV1, and mux to MP4.
    let input = MediaReader::open("input.mkv")?;
    let mut output = MediaWriter::create("output.mp4")?;

    output.set_video_codec(VideoCodec::Av1)?;
    output.set_audio_codec(AudioCodec::Opus)?;

    for frame in input.frames() {
        output.write_frame(frame?)?;
    }

    output.finalize()?;
    Ok(())
}

What’s New in 0.1.6

• 13 codec/CV stubs resolved across oximedia-accel, oximedia-codec, oximedia-audio, oximedia-image, oximedia-lut, and oximedia-caption-gen — color conversion, Vorbis VQ decode, ACES ODT variants, DASH HTTP segment fetch, and system font discovery.
• OxiFFT 0.3.0 across the workspace — Makhoul-reduction DCT-II (~4× faster), plan caching, and AVX-512 codelets; all 13 dependent crates migrated cleanly.
• exr.rs refactored into 9 modules with splitrs — the previously 2,000+ line OpenEXR reader is now split into focused core, compression, channels, metadata, scan_lines, tiles, deep, multipart, and mod files, all under the 2,000-line policy, with no public API changes.
• AWS SDK version constraints aligned to the actual Cargo.lock pins for the S3/MediaConvert/MediaLive/MediaPackage cloud paths (cosmetic).
• RUSTSEC-2026-0104 documented and ignored — a reachable panic in rustls-webpki CRL parsing, transitive via the AWS SDK; OxiMedia never performs CRL checks, so the path is unreachable. The patched route requires a C dependency excluded by the Pure-Rust Policy, so the advisory is documented and cargo audit exits 0.
• 81,582 tests passing, zero clippy warnings, WASM check clean.

Tips

• Profile the new DCT path. Anything routing through oximedia-mir, oximedia-audio-analysis, or the spectral-flux metrics benefits directly from OxiFFT 0.3.0’s faster DCT-II and plan caching — re-run your spectral benchmarks and you should see the speedup with no code changes.
• ACES is real now. The P3-D65, Rec.709, Rec.2020, D60-sim, and sRGB Output Device Transforms produce correct output in this release, so HDR/wide-gamut grading paths that were previously stubbed will now render as expected.
• Color conversion in oximedia-accel. RGB↔YCbCr, NV12, RGBA↔BGRA, HSV, and linear↔sRGB helpers are implemented — reach for these instead of hand-rolling pixel format math.
• Default builds stay sovereign. The default oximedia build pulls in zero C/Fortran symbols. The AWS/cloud SDK surface is the only place native TLS advisories surface, and OxiMedia’s usage keeps the affected code path unreachable.
• Keep files small. oximedia-image/src/exr.rs was over 2,000 lines; we split it with splitrs. If you vendor or fork EXR handling, the new module boundaries make it far easier to navigate.

This is the foundation

OxiMedia is the pure-Rust media and computer-vision layer of the COOLJAPAN ecosystem, and 0.1.6 leans on its siblings directly:

• OxiFFT 0.3.0 — the spectral engine behind audio analysis, MIR, alignment, and CV math.
• SciRS2 — scientific-computing primitives (linalg, random, SIMD) underpinning the CV and signal paths.
• OxiARC — pure-Rust compression (deflate, lz4, zstd) for containers and archival.
• OxiONNX — the pure-Rust ONNX runtime that powers OxiMedia’s opt-in ML pipelines without any C++ ort dependency.

Repository: https://github.com/cool-japan/oximedia

Star the repo if you’re tired of FFmpeg/OpenCV build hell and patent worries.

Pure Rust media and computer vision is here — fast, safe, patent-free, and sovereign.

— KitaSan at COOLJAPAN OÜ April 26, 2026