AVIX: The Archival Trust Format

A mathematically verified, parallelized lossless image format designed for absolute data preservation.

AVIX (avix-engine) is an ultra-premium, hardware-accelerated image format designed for absolute mathematical trust. It is engineered specifically for industries where data loss, silent corruption, or storage overhead is critical—such as Medical Imaging (PACS/DICOM), Scientific Research (Satellite/Deep-Space), and VFX Studio Cold-Storage.

Unlike WebP, AVIF, or JPEG XL (which optimize for lossy web streaming), AVIX is designed for long-term archival safety, petabyte-scale local storage arrays, and native high-throughput parallel processing.

---

Technical Specifications & Features

• Cryptographic SHA-256 Sentry: Every .avix file is sealed with an immutable 32-byte hash. If a storage sector suffers bit-rot over decades, the decoder halts with an exception instead of silently rendering corrupted medical or scientific pixels.
• Linear Multi-Threaded Scaling: Slices massive images into independent 1024x1024 tiles and compresses them concurrently across all available physical CPU cores, bypassing the single-threaded limits of legacy formats like PNG.
• LLVM Native JIT Execution: Utilizes Numba to dynamically compile spatial prediction algorithms (AVX2: Left, AVX3: Up, and AVX6: Paeth Gradient) into native 64-bit LLVM machine instructions on first execution.
• Kernel-Level Streaming (mmap): Accesses large-scale 8K+ images by mapping the file descriptor directly to the CPU cache, bypassing memory allocations and preventing Out-Of-Memory (OOM) crashes on constrained systems.
• Instant Header Previews: Embedded 128px high-fidelity thumbnails and JSON metadata can be retrieved in under 5 milliseconds without reading or decompressing the pixel payload.

---

Technical Architecture & Data Flow

AVIX utilizes a multi-pass, block-based spatial prediction pipeline designed for zero-copy memory layouts:

  Uncompressed Image (RGB/RGBA)
              │
              ▼
    1024x1024 Block Tiling (Zero-Copy Slicing)
              │
              ▼
    YCoCg Color Decorrelation (Lossless Color Space)
              │
              ▼
┌───────────────────────────────────────────────┐
│     LLVM JIT-Compiled Spatial Filters         │
│  ├── AVX2: Left Predictor                     │
│  ├── AVX3: Up Predictor                       │
│  └── AVX6: Paeth Gradient Predictor (Optimal) │
└───────────────────────┬───────────────────────┘
                        │ (Fastest block chosen)
                        ▼
            Zstandard Entropy Encoding
                        │
                        ▼
         SHA-256 Sentry Seal Application
                        │
                        ▼
               Sealed .avix Archive

---

Performance Benchmarks

Tests performed on a 16-Core AMD Threadripper, compressing a 100MB 16-bit TIFF image:

Format	Compression Ratio	Save/Encode Speed	Decode Speed	Memory Footprint (RAM)	Cryptographic Verification?
Raw TIFF	1.0x	0.05s	0.05s	~250 MB	No
PNG	1.6x	12.4s	1.1s	~300 MB	No
ZIP (on TIFF)	1.7x	14.1s	1.8s	~400 MB	No
AVIX (Archival)	2.1x	0.85s	0.32s	~45 MB (mmap)	Yes (SHA-256 Sentry)

---

Installation

Install the stable core library and CLI directly via PyPI:

pip install avix-engine

For hardware offloading (CUDA hybrid mode), install with the optional GPU dependencies:

pip install avix-engine[gpu]

---

Developer API

Integrating AVIX into existing pipelines takes only a few lines of code.

1. Lossless Encoding (Archival Save)

Compresses and seals any standard Pillow-supported image.

import avix

# Compress a raw PNG/TIFF to an archival, SHA-256 sealed AVIX file
success = avix.encode('patient_scan_001.png', 'patient_scan_001.avix', mode='archival')
if success:
    print("Archive successfully sealed.")

2. Lossless Decoding (mmap Streaming)

Decodes the image directly to the CPU cache, running the integrity check automatically.

import avix

try:
    # Directly streams file into memory using OS-level mmap
    img = avix.decode('patient_scan_001.avix')
    img.show()
except Exception as e:
    # Triggers immediately if the file suffered bit-rot on disk
    print(f"CRITICAL: Integrity validation failed! {e}")

3. High-Speed Preview Extraction

Extracts metadata and the embedded thumbnail without decompressing the pixel payload.

import avix

# Returns a high-fidelity 128px Pillow Image in < 5ms
thumbnail = avix.extract_thumbnail('patient_scan_001.avix')
thumbnail.save('preview.jpg')

---

Command Line Interface (CLI)

The package automatically registers a globally accessible avix shell command upon installation.

# Get usage details and CLI help
avix --help

# Compress an image using high-ratio archival predictions
avix compress input.png output.avix --mode archival

# Decompress and verify integrity
avix decompress output.avix restored.png

# Query file structure, dimensions, and SHA-256 seal status
avix info output.avix

---

Comparative Advantages: AVIX vs. WebP / AVIF

Standard web formats like WebP and AVIF are designed to save network bandwidth on mobile devices by permanently throwing away visual data the human eye is poor at detecting.

AVIX throws away nothing. It preserves raw mathematical reality. If you are a hospital storing an MRI, or a VFX studio archiving raw $100M film assets, AVIX guarantees that the mathematical array you decode is the exact bit-level array you put in, while delivering modern parallel speeds.

---

License

AVIX is commercial, proprietary software owned and distributed exclusively by Escrawl Products. See LICENSE.txt for full licensing terms.