Lossy DNG compression: impressive savings, but is it right for your RAW files?

An interesting article by Greg Benz caught my attention: compress your RAW files by “90% or more”. That is an attractive proposition for anyone with a substantial photographic collection. Greg describes using lossy DNG compression for over a year and freeing 3 TB of storage across several years of photographs.

I wanted to take that investigation further. Having already examined the opportunities for lossless compression across my own photographic archive, I could ask two useful questions: could I reproduce these substantial savings, and how much of the saving actually depended on discarding information?

The results confirm that lossy DNG can produce dramatically smaller files. But the comparison matters. My tests also explain why I have chosen to preserve my RAW files losslessly and use high-quality lossy JPEG XL for finished photographs instead.

Putting the storage savings to the test

I compared 47 photographs across 24 camera and recording-format groups, spanning 2006–2025. The collection included Canon, Fujifilm, Sony and Leica cameras, along with iPhone DNG files. Using Adobe DNG Converter 18.6, I compared optimised lossless DNGs with six lossy settings.

This builds on my separate lossless compression research, including 23,078 original camera RAW photographs. That larger study established the storage opportunity available without discarding RAW image data. The 47-photograph comparison examines what lossy compression adds.

The headline result is striking. Against the original camera-file baseline, the default lossy setting produced estimated reductions of about 91% for uncompressed Sony RAW, 89% for uncompressed Fujifilm RAW, 83% for compressed Sony RAW and 76% for Canon CR2. These results support the scale of the savings Greg describes.

To make those figures easier to picture, here is the storage that would remain from 10 TB of each camera-file type.

Illustrative storage equivalents, rounded from the camera-type comparisons. Original camera-file comparisons use the recorded original-size baseline; the direct comparison between lossless and lossy versions of the same 47 photographs appears below.

But what do those impressive figures really tell us?

Taken alone, these results can give a misleading impression of the benefit of choosing lossy compression. They compare the final lossy file with the camera original, overlooking the space that better lossless compression could already have saved.

That distinction makes a substantial difference. In the Sony uncompressed comparison, 10 TB becomes about 4.1 TB with optimised lossless compression, before considering lossy compression at all. Lossy then reduces it further to about 0.9 TB. For Fujifilm, the corresponding figures are about 4.6 TB and 1.1 TB.

The pattern differs with Canon CR2: 10 TB becomes about 7.7 TB losslessly and about 2.5 TB with the default lossy setting. Here, lossy compression provides more of the total reduction. The starting format is a major part of the explanation.

The useful question is how much additional space lossy compression saves once the RAW data is already stored efficiently. That is the comparison I would use when deciding whether to discard any of that data.

The additional saving is still substantial

Comparing the lossless and lossy versions of the same photographs gives the following results. Here, each row starts with 1 TB of optimised lossless DNGs.

Storage equivalents calculated from the measured group reductions; decimal units, with 1 TB = 1,000 GB. These figures measure file size, not image quality.

Across the full sample, I measured a size-weighted saving of approximately 76% beyond optimised lossless compression. That means roughly a quarter of the storage remains. These are meaningful savings, even after making the more relevant comparison.

So the choice is not between keeping 10 TB and reducing it to 1.3 TB. In this illustration, I can keep all the RAW image data in about 5.3 TB. The remaining decision is whether the additional saving of about 4 TB is worth changing what I preserve.

What changes when RAW files become lossy DNGs?

Lossless compression stores the image data more efficiently and restores the same values when decoded. Lossy compression permanently changes some of those values to achieve smaller files.

There is another change in the usual Adobe lossy conversion of a mosaic RAW file. The camera’s pattern of individual colour measurements is converted into full-colour pixels, a process called demosaicing. The result is a linear DNG. It retains considerable editing flexibility, but the original sensor mosaic is no longer available for a different conversion later.

The distinction matters because a RAW file is the starting point for processing. It is not necessarily the photograph as it will eventually be seen. A very dark area in a cathedral photograph may be lifted substantially during editing, revealing detail that was barely visible in the initial image.

JPEG XL’s reference encoder describes quality using a visual distance: zero is mathematically lossless, while one is a visually lossless target. That is useful guidance about image appearance. It is not a promise that every subsequent edit or reconstruction will produce identical results. Preserving data and preserving a finished appearance are different requirements.

What other photographers and developers have found

The external evidence is valuable precisely because it covers both impressive practical results and circumstances where the differences matter.

• Excellent results in normal use. In his 2024 investigation, Greg Benz reports that differences generally require very substantial enlargement to see, while noting occasional edge artefacts and changes in extreme shadow adjustments. Alex Armitage’s practical demonstration likewise makes a strong case for the editing flexibility retained in very small lossy DNGs.
• Demanding shadow recovery can reveal differences. In his Canon EOS R5 comparison, Nando Harmsen found detail loss, colour noise and artefacts after substantial shadow recovery. He also found that denoising before conversion removed the visible differences in his comparison. The processing sequence matters, as well as the format.
• High-ISO processing and enlargement deserve attention. Greg’s 2025 follow-up describes colour errors in deep shadows when an ISO 6400 photograph was compressed before denoising, and much better results when denoising came first. He also reports problems with Adobe Super Resolution on lossy DNGs, while obtaining good enlargement results with Topaz Gigapixel.
• Software can respond differently. Donncha Ó Caoimh’s experience is broadly positive about appearance, but includes an example of automatic perspective correction behaving differently after conversion. This is a useful reminder to assess the editing process as well as the initial view.

Older articles also describe restrictions that have since changed. Adobe’s current Enhance documentation lists linear DNGs among the formats supported by Denoise. Raw Details still requires mosaic RAW data; Super Resolution also supports formats such as JPEG and TIFF. The issue is therefore not a blanket inability to process lossy files, but which information and processing options remain available.

There is a positive development on the software side too. DxO explains its compressed linear DNG approach in PureRAW 6, claiming files about four times smaller while preserving perceived quality, including during heavy editing. Its pipeline combines dynamic-range compression with JPEG XL after its RAW processing. This is a different implementation from the Adobe conversion tested here, but it reinforces the value of considering where compression occurs in the process.

These reports explain why lossy DNG can be an excellent practical choice for one use and an unnecessary restriction for another. My measurements establish the storage benefit; these external findings help inform the decision about editing and preservation.

Why I keep my RAW files lossless

My photographs are material I may return to for many years. Better processing tools, different displays and new ideas can all change what I want to do with them. My experience of bringing old RAW photographs to life with True HDR has made that particularly clear.

For cathedral interiors and other scenes with deep shadows and bright highlights, I want the original image data available when I make those decisions. Having already achieved substantial lossless storage savings, I prefer to retain that latitude rather than make an irreversible choice before I know how the photograph will eventually be processed.

Preserving photographs for the future is part of the purpose of taking them. That includes preserving the opportunity to make better use of them. For my archive, efficiently compressed lossless RAW files are the safer foundation.

Lossy JPEG XL has an important place after editing

This is not an argument against lossy JPEG XL. I use it as a central part of my SDR and HDR photography workflow.

Once I have recovered the shadows, adjusted the colour and finished the photograph, I can judge compression against the appearance I actually want to preserve. A high-quality lossy JPEG XL master can avoid the very large files that lossless encoding of a finished HDR image sometimes produces. The optimised lossless RAW remains available if I want to start again.

My approach is therefore:

• Preserve RAW data efficiently using lossless compression.
• Create separate SDR and HDR masters that preserve the finished versions: a high-quality SDR JPEG and a high-quality lossy HDR JPEG XL, following the settings in the workflow guide.
• Create viewing copies from those masters for the website, galleries and other uses.

Lossy DNG copies can also be useful for a portable collection or a particular collaboration while the lossless archive remains intact. They are an option, rather than an additional requirement in my workflow.

The storage savings are real. The more important decision is where to use lossy compression. For my photography, keeping the RAW data lossless and applying high-quality lossy compression to finished masters provides the combination of efficiency, quality and long-term flexibility I want.

More about the test and alternative compression settings

The comparison used Adobe DNG Converter 18.6 on macOS in September 2026: 47 photographs across 24 camera and recording-format groups. The sample covers 23 camera models, with compressed and uncompressed Sony A7R V RAW treated separately. Most groups contain two photographs; the Canon 450D group contains one. The saved analysis records conversions to lossless JPEG XL DNG and six lossy settings, with the settings read back from the generated files. File sizes include embedded previews.

The camera-original comparisons use recorded original file sizes from the archive research and are estimates by camera type. They are distinct from the direct comparison between lossless and lossy versions of the same 47 photographs. The primary experiment reported here measures storage; the discussion of image-quality effects draws on the linked external investigations.

The default lossy setting in the recorded DNG Converter tests used distance 0.5. Gentler distances were also tested. Compared with optimised lossless files, the verified file sizes show approximate overall reductions of 20% at distance 0.1, 49% at 0.2, 62% at 0.3 and 76% at 0.5, for demosaiced output. Distance settings describe the compression choice; they are not measurements of visible quality in these tests.

A separate lossy mosaic option retains the sensor’s mosaic structure instead of converting it to full-colour pixels first. It still changes image values: keeping the mosaic structure does not make compression lossless. Its behaviour also varied with input type in the investigation. It offers another approach to explore, but does not change my preference for preserving the archive losslessly.

For the broader technical background, see the JPEG XL reference implementation and its encoder documentation. For the practical distinction between RAW preservation, edited masters and delivery formats, see my guide to SDR and HDR image formats.

This research also supports my experiment in building SDR and HDR photo galleries: preserve high-quality source material, then measure how best to deliver the photographs people actually see.

Last updated: September 2026