RAW Photo Formats: CR3, NEF, ARW, DNG, and More

When someone says "shoot RAW," they're telling you to capture the sensor's full data instead of letting the camera process it into a JPEG. Good advice. But "RAW" isn't a format — it's a category. Every camera manufacturer invented their own RAW format, each with different containers, compression methods, and metadata structures. Canon's CR3 has nothing in common with Sony's ARW except the fundamental concept.

This guide covers what RAW files actually contain, how each manufacturer's format differs, what Adobe DNG solves, and the practical implications for storage, processing, and long-term archival.

What a RAW File Actually Contains

A RAW file stores data from four distinct sources, packaged together.

1. Sensor Data (The Main Payload)

The image sensor is covered by a Bayer filter (or X-Trans in Fujifilm cameras) that places a colored filter over each photosite. Each pixel records only one color: red, green, or blue. A 45-megapixel sensor has ~22.5 million green photosites, ~11.25 million red, and ~11.25 million blue (green gets twice the count because human vision is most sensitive to green).

Bit depth: Each photosite records at 12-bit (4,096 brightness levels) or 14-bit (16,384 levels). Compare to JPEG's 8-bit (256 levels). This extra precision is where RAW's editing headroom comes from — you can push exposure +3 stops on a 14-bit RAW and still have detail in the shadows. Try that on an 8-bit JPEG and you get blocky noise.

Not an image yet: The sensor data is a mosaic of single-color values. It must be "demosaiced" (interpolated into full RGB pixels) before it looks like a photograph. Every RAW editor performs this step differently, which is why the same RAW file can look subtly different in Lightroom vs Capture One vs DxO.

2. Metadata

EXIF: Camera model, lens, focal length, aperture, shutter speed, ISO, date/time, GPS (if enabled), metering mode, flash, white balance as shot. Identical to JPEG EXIF but also includes RAW-specific fields like sensor temperature, noise characteristics, and lens correction profiles.

Makernotes: Camera-specific proprietary metadata. Canon stores autofocus point information, scene classification, and custom function settings. Nikon stores VR (vibration reduction) data, D-Lighting settings, and active focus point. Sony stores lens compensation data and creative style parameters. Much of this is undocumented by the manufacturers.

3. Embedded JPEG Preview

Every RAW file contains at least one JPEG preview — typically a full-resolution JPEG that the camera generated using its default processing pipeline. This is what you see on the camera's LCD and in file browsers. Some cameras embed multiple previews at different resolutions (thumbnail, medium, full).

Size impact: The embedded JPEG can be 2–10MB depending on resolution. A 50MB RAW file might contain 8MB of JPEG previews. Some tools can strip these to save space, but this breaks thumbnail display in many file browsers.

4. Camera Processing Settings

White balance, picture style/profile, color space setting, noise reduction level, sharpening level, contrast, saturation. These are recorded but not applied to the sensor data. They're instructions: "if this RAW is processed with default settings, use these values." You can change any of them in your RAW editor without penalty.

Manufacturer RAW Formats

Each manufacturer uses a different container and compression scheme. Here are the major ones.

Canon: CR3 (and Legacy CR2)

CR3: Introduced with the EOS M50 in 2018. Based on the ISO BMFF (Base Media File Format) container — the same family as HEIF and MP4. Supports Canon's C-RAW lossy compression (40% smaller files with minimal visible quality loss in most shooting scenarios).

CR2: Previous Canon RAW format based on TIFF container. Used by every Canon DSLR and early mirrorless (EOS R, RP). Lossless compression. Fully supported by all major RAW editors.

File sizes (Canon R5, 45MP): CR3 standard: ~45–55MB. CR3 C-RAW: ~25–30MB. The C-RAW savings are significant for high-volume shooters — a 128GB card holds ~4,300 C-RAW vs ~2,500 standard RAW.

Convert CR2 to JPG | Convert CR2 to PNG | Convert CR2 to TIFF

Nikon: NEF

Container: TIFF-based. NEF has been Nikon's RAW format since the D1 (1999). The format has evolved but maintained backward compatibility.

Compression options: Uncompressed (largest), Lossless Compressed (recommended — no quality loss, ~40% smaller), and Lossy Compressed (additional ~30% savings with minor quality reduction in extreme shadow recovery scenarios).

Bit depth: 12-bit or 14-bit, selectable in camera. 14-bit lossless compressed is the recommended setting for maximum quality without excessive file sizes.

File sizes (Nikon Z8, 45.7MP): 14-bit lossless compressed: ~35–45MB. 14-bit uncompressed: ~55–70MB. 12-bit lossy compressed: ~20–28MB.

Convert NEF to JPG | Convert NEF to PNG | Convert NEF to TIFF

Sony: ARW

Container: TIFF-based. Sony has used ARW since the Alpha DSLR line. The format has gone through several internal revisions (ARW 2.0, 2.3, etc.) but the extension remains .arw.

Compression: Sony uses a combination of predictive delta coding and Huffman compression. The A7 IV and newer cameras offer lossless compressed and uncompressed options. Older cameras (A7III and earlier) use a somewhat lossy compression that clips 11-bit data to fit, though the impact is only visible in extreme shadow recovery.

File sizes (Sony A7R V, 61MP): Lossless compressed: ~60–75MB. Uncompressed: ~120MB. The 61MP sensor produces the largest RAW files of any current full-frame camera.

Convert ARW to JPG | Convert ARW to PNG | Convert ARW to TIFF

Fujifilm: RAF

X-Trans sensor: Fujifilm uses a unique 6x6 repeating filter pattern instead of the standard 2x2 Bayer pattern. X-Trans reduces moiré without an optical low-pass filter, producing sharper images at the sensor level. The tradeoff: third-party RAW editors must implement X-Trans-specific demosaicing, and some (notably Adobe) produce slightly wormy artifacts in fine detail areas.

File sizes (X-T5, 40MP): Lossless compressed: ~40–50MB. Fujifilm's compression is effective; RAF files are typically smaller than equivalent Bayer RAWs at the same resolution.

Best software: Capture One handles X-Trans well. Fujifilm's own RAW Studio (free) uses the camera's processor remotely for the most authentic demosaicing. Lightroom has improved significantly but still shows occasional artifacts in fur, fabric textures, and fine repeating patterns.

Other Manufacturer Formats

Panasonic RW2: TIFF-based, similar to NEF. Used by Lumix cameras. Well-supported in all major editors.

OM System (Olympus) ORF: Proprietary container. Micro Four Thirds sensors produce smaller RAW files (20–25MB at 20MP) due to the smaller sensor area.

Leica DNG: Leica is the only major manufacturer that uses DNG natively. Leica M, Q, and SL cameras record directly to DNG, giving you the archival benefits without conversion.

Hasselblad 3FR/FFF: Medium format RAW (100MP+). File sizes of 100–200MB per image. Processed primarily in Phocus (Hasselblad's free software) or Capture One.

Adobe DNG: The Universal RAW

Digital Negative (DNG) is Adobe's answer to the proprietary RAW format fragmentation problem.

Why DNG Exists

The problem: There are 600+ unique RAW formats in existence (counting each camera model as a variant). Every time a new camera launches, RAW editing software needs an update to support it. If a manufacturer goes bankrupt or abandons a format, there's no guarantee any software will read those files in 20 years. Kodak's DCR format and Sigma's X3F are already at risk.

DNG's solution: An openly documented RAW container based on TIFF/EP. Adobe publishes the complete specification (DNG 1.7 as of 2023). Any developer can implement a DNG reader without reverse-engineering or licensing. The sensor data is stored in standardized layouts (CFA pattern, linearization tables, color matrices) that don't depend on manufacturer-specific knowledge.

Converting to DNG

Adobe DNG Converter: Free standalone tool (Windows/Mac). Batch converts any supported RAW format to DNG. Options include embedding the original RAW file (safety net, doubles file size), lossy compression (significant space savings, minor quality impact), and choosing DNG version compatibility.

Lightroom import: Lightroom can convert to DNG on import (Preferences > File Handling > "Convert to DNG during import"). Adds a few seconds per image to import time.

What changes in conversion: Sensor data is re-packed into the DNG container. Metadata is mapped to DNG standard fields. Camera-specific makernotes are preserved as-is (in a private tag). The image data itself is mathematically identical if lossless compression is used.

What you might lose: Some camera-specific metadata that editors use for manufacturer-specific processing (Canon's Dual Pixel autofocus data, Nikon's picture control profiles). Most users never notice. If you depend on manufacturer-specific features in your editor, test before committing to full DNG conversion.

DNG File Sizes

Lossless DNG: Typically 0–5% smaller than the original proprietary RAW. DNG uses a more efficient lossless compression (Adobe's custom predictor) than most manufacturer defaults.

Lossy DNG: 50–70% smaller than original RAW. Applies linear demosaicing and reduces to 8-bit or 10-bit before compressing. This is a significant quality reduction for extreme editing but imperceptible for normal workflow. Not recommended for archival.

DNG with embedded original: Original file size + DNG size. Roughly doubles total storage but provides a perfect safety net.

Why RAW Matters: The Quality Difference

The practical editing advantages of RAW over JPEG, quantified.

Editing Headroom in Numbers

Exposure recovery: 14-bit RAW gives approximately 6 stops of highlight/shadow recovery. In practical terms, an image shot 3 stops underexposed can be pushed to correct exposure with no visible quality loss. A JPEG in the same situation produces banding, noise amplification, and color shifts after even 1 stop of recovery.

White balance: In RAW, white balance is metadata — you can change it with zero quality penalty because the sensor data is pre-white-balance. In JPEG, white balance is baked in. Correcting white balance in a JPEG requires shifting channels (adding blue to fix a warm image, for example), which compresses the available tonal range in the adjusted channel.

Highlight detail: RAW files often contain highlight information 1–2 stops above what the embedded preview shows. Clipped highlights in the JPEG preview may have recoverable detail in the RAW sensor data. This is especially valuable for wedding photography (white dresses) and landscape photography (bright skies).

The Storage Cost

File size comparison (45MP camera):

• RAW (14-bit lossless): ~45MB per image
• RAW (C-RAW/lossy compressed): ~28MB per image
• JPEG Fine (quality 96): ~15MB per image
• JPEG Normal (quality 85): ~8MB per image

For a 1,000-image shoot: RAW = ~45GB, JPEG Fine = ~15GB. Storage cost in 2026: ~$1/TB on hard drives. The RAW premium for that shoot is about $0.03. In exchange, you get non-destructive editing flexibility for every single shot. The math overwhelmingly favors shooting RAW.

RAW Software Compatibility

RAW formats exist in an awkward state: technically proprietary and fragmented, practically essential for professional photography. Every serious photographer should shoot RAW, understand what their camera's specific format contains, and have a DNG backup strategy for long-term archival safety.

The fragmentation problem is real but manageable. Adobe, Capture One, and the open-source community maintain support for hundreds of RAW variants. DNG provides insurance against any single manufacturer's format becoming abandoned. The bigger risk isn't format obsolescence — it's not shooting RAW in the first place.