Both DxO PureRAW and Lightroom AI Denoise use neural networks to remove noise from RAW files, but they intervene at different stages of the conversion process. DxO processes your file before the image is fully assembled — the most powerful position in the pipeline. Lightroom integrates directly into your existing catalog. For maximum detail recovery, DxO leads. For convenience, Lightroom is the stronger choice.
You pressed the shutter on a keeper — decisive moment, sharp focus, the right light. Then you opened the file. Noise was crawling through every shadow: luminance grain obscuring fine detail, color speckles destroying the feather or skin texture you worked to capture.
Understanding why your RAW file contains noise is the first step toward reclaiming it.
Two tools dominate the professional conversation right now: DxO PureRAW and Adobe Lightroom’s AI Denoise. Both use artificial intelligence to clean noise from RAW files. But they work at different points in the conversion pipeline, with meaningfully different results and meaningfully different trade-offs.
This article covers the physics behind why noise appears, how AI denoising differs from traditional noise reduction, and what each tool does well and poorly. By the end, you will have a clear framework for choosing between them — or for knowing when to use both.
- Why Your Camera Can't Avoid Noise
- Why Traditional Editing Makes Noise Worse
- What AI Denoising Actually Does
- SNR Inforgraphic
- The Technical Truth of SNR
- Avoiding "SNR Debt"
- The "Step Zero" Workflow
- The Gold Standards of Purification
- DxO PureRAW: Denoising Before the Image Exists
- Lightroom AI Denoise: Inside Your Existing Workflow
- Head-to-Head: Where Each Tool Wins and Loses
- Which Tool Should You Use?
- Conclusion
- FAQ
- References
Why Your Camera Can’t Avoid Noise
Every digital image starts as a count of photons. The sensor doesn’t interpret a scene the way your eye does — it measures how many light particles struck each photosite during the exposure. That counting process is where noise begins. It is not a manufacturing defect or a camera flaw. It is the physics of light.
The Physics of Photon Capture
The primary source of noise in a digital image is photon shot noise — not an artifact of the camera’s electronics, but a consequence of how light itself behaves. Photons arrive at the sensor following a Poisson distribution, an inherently random process.
Because shot noise scales as the square root of the signal, the Signal-to-Noise Ratio (SNR) improves with more light: signal increases linearly while noise increases only by its square root.
Electronic read noise — generated during analog-to-digital conversion — becomes dominant only when signal levels are very low. The sensor’s Full Well Capacity (the maximum number of photoelectrons a single pixel’s photosensor can hold before saturating) defines the theoretical ceiling of achievable SNR.
In Plain English: More light means less noise — not because your camera works harder, but because statistics favor larger samples. A pixel that counted 10,000 photons has a far better signal-to-noise ratio than one that counted 100.
Why Traditional Editing Makes Noise Worse
The Multiplier Problem
When you drag the Exposure slider in Lightroom, you are multiplying every pixel value by the same factor. The signal gets brighter. So does the noise — at exactly the same rate. No traditional editing tool can separate signal from noise. It can only scale both together, which is why aggressively brightening an underexposed file always looks worse than it did before.
The “Shoot Low ISO, Fix It Later” Myth
Many photographers deliberately underexpose at a low ISO, planning to recover shadows in post. This strategy frequently produces more noise than a correctly exposed frame at a higher ISO. Lifting shadow data amplifies the sparse photon count captured in those regions — and the noise that dominates them. The sensor needs light, not protection from it.
Underexposed shadows carry a second problem. The signal in the red and blue channels often falls entirely below the noise floor. When white balance is applied, those channels are scaled more aggressively than green, producing the characteristic pink and purple blotches visible in pushed files from most major camera brands. Denoising cannot fully reverse this — the signal was never there.
“A correctly exposed frame at ISO 6,400 will typically contain less recoverable noise than an underexposed frame at ISO 800 pushed two stops in post.”
What AI Denoising Actually Does
AI denoising tools — including DxO PureRAW and Lightroom AI Denoise — use Convolutional Neural Networks (CNNs) trained through supervised learning. During training, the network is shown matched pairs of noisy and clean images and learns to recognize the statistical signatures of specific noise types: the grain texture of shot noise, linear read-noise banding patterns, and color speckle in low-signal chrominance channels.
The result is a mathematically predicted reconstruction of the probable underlying signal — not a filtered approximation. Traditional filters like Gaussian or median blurring smooth pixel neighborhoods to suppress noise, inevitably destroying fine edge detail in the process.
In Plain English: Traditional noise reduction blurs the problem away — and the detail with it. AI tools predict what the original signal probably looked like, then rebuild the image from that prediction. The difference is visible in fine fur, fabric, and foliage.
SNR Inforgraphic
Reclaiming the Ratio
How AI Denoise and DxO PureRAW manipulate the immutable physics of Signal-to-Noise Ratio (SNR) to rescue modern photography.
The Technical Truth of SNR
Signal-to-Noise Ratio (SNR) is determined the exact millisecond your camera’s shutter closes. It is a physical metric based on the number of photons hitting the sensor. Once captured, traditional editing cannot improve this ratio. Linear multipliers, like exposure sliders, blindly amplify both the clean photographic signal and the digital noise interference simultaneously.
Signal vs. Noise Amplification
This stacked visualization reveals the limitation of traditional post-processing. When an underexposed RAW file is pushed using standard exposure sliders, the signal increases, but the noise floor is dragged up right along with it. The ratio remains locked.
AI tools break this physical constraint mathematically. By intelligently identifying patterns, they isolate and suppress the noise floor while maintaining—or even selectively boosting—the underlying signal, resulting in a synthetically improved SNR.
Avoiding “SNR Debt”
Because you cannot naturally add more photons after the capture, post-processing requires a “preservation” mindset. SNR Debt occurs when a photographer applies heavy global adjustments—dragging midtones down into the noise floor or aggressively lifting shadows—thereby degrading the image quality irreversibly in a traditional workflow.
The Degradation Trajectory
The line chart maps the perceived image quality through progressive editing stages. In a traditional workflow, every shadow lift or exposure adjustment incurs “SNR Debt,” dropping the overall fidelity.
By applying AI purification at the beginning of the process, the image establishes a resilient, elevated baseline. Subsequent edits cause minimal degradation, preserving a clean canvas through to the final output.
The “Step Zero” Workflow
Both Adobe and DxO recommend applying AI noise reduction before any other adjustments. Modern editing relies heavily on AI masks (like “Select Subject”). Heavy noise confuses these algorithms. Performing purification first provides a clean canvas, making all subsequent automated and manual tools vastly more accurate.
Raw Capture
Fixed Signal & Noise
Step Zero: AI Denoise
DxO DeepPRIME / LR AI Denoise
Creates Linear DNG
Base Tone Edits
Exposure, Shadows, Highlights
Targeted Masks
AI Subject Selection
Edit Persistence Note
In Lightroom, running AI Denoise creates a new DNG from the raw data and automatically reapplies existing sliders. However, because shadow tonality shifts after denoising, executing this at “Step Zero” prevents the need for extensive backtracking and slider readjustments.
The Gold Standards of Purification
While they cannot change the physical photons captured, tools like DxO PureRAW and Lightroom AI Denoise represent the apex of computational photography, separating signal from noise with unprecedented accuracy.
Technical Performance Breakdown
DxO PureRAW (DeepPRIME): Performs denoising and demosaicing (reconstructing color from the Bayer pattern) simultaneously. This holistic analysis yields a highly “elastic” file with superior color accuracy.
Lightroom AI Denoise: Operates directly on unprocessed RAW data to output a pristine Linear DNG. It offers exceptional convenience by keeping the entire workflow within a single application ecosystem while drastically improving the noise profile.
DxO PureRAW: Denoising Before the Image Exists
DxO PureRAW occupies a unique position in the workflow: it processes your file before the RAW data has been converted into a visible image. This "Step Zero" approach gives it access to information that every other tool — including Lightroom — receives only after the data has been partially assembled and committed.
Why Demosaicing Matters
A camera sensor records one color channel per photosite — red, green, or blue — in a repeating Bayer array. Demosaicing reconstructs full-color information for every pixel by interpolating from its neighbors.
Traditionally, this happens after a separate denoising pass. The problem: noise in the raw data corrupts the interpolation algorithm, producing artifacts — color moiré, false-color fringing, or the sinuous "worm" patterns common in Fujifilm X-Trans files — that become permanently embedded in the image.
DxO's DeepPRIME engine performs demosaicing and denoising simultaneously. A single neural network analyzes noise patterns during color reconstruction, distinguishing real detail from sensor interference at the point where separation is most effective.
In Plain English: Most tools clean a photo after it has been assembled from the sensor's raw data. DxO cleans the data while the image is still being built — catching noise before it gets locked into the color information.
DeepPRIME vs. DeepPRIME XD: When to Use Each
DxO offers two engine variants. DeepPRIME is the standard option: fast, broadly compatible, and sufficient for most ISO 800–3200 files. DeepPRIME XD uses a larger, more computationally intensive neural network designed for the most demanding high-ISO work, with higher processing times to match.
According to DxO's own documentation, DeepPRIME XD recovers the equivalent of two to three additional stops of ISO performance compared to conventional demosaicing.
An independent review by CaptureLandscapes broadly supports a meaningful advantage at ISO 3200 and above, though the degree of improvement varies by camera and sensor architecture.
"DxO intervenes before the RAW file is fully assembled. Every other tool — including Lightroom — works with data that has already been partially constructed."
Lightroom AI Denoise: Inside Your Existing Workflow
Lightroom's AI Denoise operates on the original RAW data, not the developed preview. It outputs a new Linear DNG file, which is reimported into your catalog with existing edits automatically re-applied. The original RAW file is never altered. For photographers already working in Lightroom, the integration requires no change to their standard process.
What "Linear DNG" Actually Means
A Linear DNG is a RAW file that has undergone demosaicing — full-color reconstruction from the Bayer array — but has not been subjected to gamma correction, color rendering, or any non-linear transformation.
The data remains a linear function of original scene luminance, preserving the full dynamic range of the proprietary RAW file: equivalent highlight and shadow recovery potential. The 32-bit floating-point encoding provides extensive headroom for large tonal adjustments, preventing the quantization banding that would appear if identical edits were applied to an 8-bit or 16-bit processed file.
In Plain English: A Linear DNG gives you everything the original RAW file offered — full highlight recovery, full shadow depth — with noise already cleaned. It behaves like a RAW file but without the noise problem.
Why Order of Operations Matters
Applying AI Denoise after pushing shadows or blacks in Lightroom will trigger automatic re-application of your edits to the new DNG. Here is more information on the correct order of Editing Operations.
This is convenient but not consequence-free. Denoising changes shadow tonality — removing noise that was contributing to perceived brightness. Check your blacks and shadows after processing and fine-tune as needed.
Lightroom's masking tools — Select Subject, Select Sky, and Content-Aware Remove — analyze image edges to build their selections. Heavy noise mimics edge signals, misleading these algorithms. Running AI Denoise yields more accurate selections, particularly for high-frequency subjects such as bird plumage or animal fur at ISO 6400 and above.
Head-to-Head: Where Each Tool Wins and Loses
Choosing between these tools is not a matter of one being categorically better. Each has clear strengths in specific scenarios. The critical factors are what you shoot, at what ISO values, and what your existing workflow looks like.
| DxO PureRAW | Lightroom AI Denoise | |
|---|---|---|
| Processing stage | Before demosaicing (Step Zero) | After demosaicing |
| Best subjects | Wildlife, sport, fine texture | Portraits, people, moderate ISO |
| Output format | Linear DNG (external file) | Linear DNG (within catalog) |
| Demosaicing | Simultaneous with denoising | Performed separately first |
| Skin rendering | Can over-smooth at defaults | More natural by default |
| Batch processing | Faster for high volume | Better for single-file edits |
| X-Trans support | Specialized support | Limited |
| Workflow step | Dedicated preprocessor | Integrated into Lightroom |
| Storage overhead | Additional DNG per file | Additional DNG per file |
Detail Recovery: Feathers, Fur, and Fine Texture
The gap between these tools is not uniform across all subjects. It is most visible on fine, high-frequency texture — feathers, fur, fabric weave, foliage — and the reason is structural, not a matter of algorithm quality.
Demosaicing reconstructs color by reading neighboring photosites. On a smooth surface like sky or skin, neighboring pixels are similar in tone and color, so the interpolation has little room to go wrong.
However, on a feather or a patch of dense foliage, neighboring pixels are legitimately different from each other — rapid changes in brightness and color are real detail, not noise. The problem is that at high ISO, noise creates a pattern of rapid, unpredictable variation between adjacent pixels.
A demosaicing algorithm that runs before denoising cannot reliably tell the difference. It interprets some of that noise as real edge information, bakes it into the color reconstruction, and produces false texture that survives the subsequent denoising pass.
DxO's simultaneous approach sees both problems at once and resolves the ambiguity before it gets locked in. The result on a bird's wing at ISO 6400 can be the difference between individual barbs and a smeared approximation of plumage.
On smooth, low-frequency subjects — open sky, a plain background, human skin — this distinction largely disappears. The demosaicing step has nothing complex to misread, so Lightroom's sequential approach produces results that are visually indistinguishable from DxO's.
If the majority of your work involves portraits or studio subjects, the processing-stage advantage DxO holds is largely theoretical in practice.
Skin Rendering: Where Lightroom Often Wins
DxO's XD mode can produce an unnaturally smooth appearance on human skin. Adobe's implementation is generally more conservative, yielding more natural skin texture by default.
If DxO results look over-processed on portrait work, reduce the Luminance slider — it directly controls the aggressiveness of the smoothing effect without affecting edge sharpness significantly.
Speed and Batch Processing
Both tools use GPU acceleration, and processing speed scales directly with GPU capability. DxO has a reported advantage for large batch operations. Lightroom's catalog integration makes it faster for single-file edits within an active session. Neither tool has a decisive edge — the right choice depends on your typical file volume.
Documented Weaknesses — DxO PureRAW
DxO adds a dedicated preprocessing step before your primary editor opens the file. It offers no creative editing capabilities. XD mode on portrait subjects can appear over-processed at default settings. Each processed output is a full-resolution DNG file, adding significant storage overhead at volume.
Documented Weaknesses — Lightroom AI Denoise
Lightroom AI Denoise operates after demosaicing — it cannot access the structural advantage available at Step Zero. On files above ISO 6400 with significant shadow content, the gap in fine-detail recovery compared to DxO is visible and repeatable. Shadow and black settings require review after processing, as tonality shifts with noise removal.
Which Tool Should You Use?
Use DxO PureRAW if you regularly shoot subjects with fine high-frequency detail — birds, wildlife, foliage, sport, and at ISO 3200 and above. It is also the right choice for Fujifilm X-Trans files, which benefit significantly from DxO's specialized demosaicing support compared to any downstream alternative.
Use Lightroom AI Denoise if you shoot at moderate ISO values, primarily photograph people, or prefer to stay within a single application. Its results on portrait subjects are consistently more natural than DxO XD at default settings, and catalog integration removes a preprocessing step from the pipeline.
The most demanding workflows combine both: DxO as the preprocessor, handing off a clean Linear DNG to Lightroom for all tonal and color decisions. This maximizes signal quality at the earliest possible stage. The DxO-to-Lightroom handoff workflow has storage and time costs — weigh those against your output volume before committing.
"A blurry frame at ISO 1,600 is unusable at any processing stage. A sharp frame at ISO 10,000 is now a portfolio-level asset." (23 words ✓)
Conclusion
No software can add photons that were never captured. The Signal-to-Noise Ratio is fixed at the moment of exposure. What AI denoising does is separate the existing signal from the noise entangled with it — and it does this more effectively than any tool that preceded it.
For working photographers, this changes the calculus in the field. The decision is no longer between a sharp, noisy frame and a blurred, clean one. Prioritize shutter speed, expose correctly, and let the software handle the noise. The results justify that trust.
DxO and Lightroom are not competing answers to the same question. DxO addresses signal purity at the point of conversion. Lightroom addresses convenience within an established workflow. Both are valid. The correct choice depends on what you shoot, at what ISO, and how much friction you can tolerate in post-processing.
FAQ
Is it worth paying for DxO PureRAW if I already have Lightroom? Yes, if you regularly shoot wildlife, birds, heavy foliage, above ISO 3200. DxO's integrated demosaicing and denoising recover detail that Lightroom cannot match. The two tools are complementary — not alternatives to each other.
Does AI denoising work on JPEG files? It works, but the results are significantly worse. AI denoisers are optimized for linear RAW data. Applying them to a JPEG — already processed and compressed — yields far less recoverable detail.
Will AI denoising make my photos look plastic or over-processed? It can, particularly DxO's XD mode on skin tones. Reduce the Luminance slider to dial back the effect. Lightroom AI Denoise is generally more conservative at default settings.
Should I run AI Denoise before or after other Lightroom adjustments? Before. Noise misleads Lightroom's AI masking tools, so clean the file first. Check your blacks and shadows after denoising — tonality can shift when noise is removed.
Can AI denoising rescue a severely underexposed photo? Partially. It cleans the noise, but underexposure means missing photon data — details that were never recorded cannot be reconstructed. Correct exposure is the only real solution.
References
- Scientific Imaging, Inc. "Signal and Noise." scientificimaging.com. https://scientificimaging.com/knowledge-base/signal-and-noise/ (2026)
- Evident Scientific. "CCD Signal-to-Noise Ratio: Read Noise, Shot Noise & Sensitivity." evidentscientific.com. https://evidentscientific.com/en/microscope-resource/knowledge-hub/digital-imaging/concepts/ccdsnr (2026)
- DxO. "DxO DeepPRIME: Still the world's best AI denoising tool." dxo.com. https://www.dxo.com/technology/deepprime/ (2026)
- Adobe Support. "Enhance fine details in raw images." helpx.adobe.com. https://helpx.adobe.com/lightroom-classic/help/enhance-details.html (2026)
- Tucsen. "Photon Shot Noise in Scientific Imaging SNR." tucsen.com. https://www.tucsen.com/learning/photon-shot-noise-in-scientific-imaging-snr/ (2026)
- Adimec. "Read Noise versus Shot Noise." adimec.com. https://www.adimec.com/read-noise-versus-shot-noise-what-is-the-difference-and-when-does-it-matter/ (2026)
- Nikon MicroscopyU. "CCD Signal-To-Noise Ratio." microscopyu.com. https://www.microscopyu.com/tutorials/ccd-signal-to-noise-ratio (2026)
- DxO. "What are Linear DNG files? How do you use them?" dxo.com. https://www.dxo.com/news/linear-dng/ (2026)
- Adobe Blog. "Denoise Demystified." blog.adobe.com. https://blog.adobe.com/en/publish/2023/04/18/denoise-demystified (2023)
- Kaylor, A. "Mastering Noise in Low Light Photography." annalisekaylor.com. https://www.annalisekaylor.com/blog/mastering-noise-in-wildlife-photography (2026)
- Song Hurst, S. "DxO PureRAW 6 vs Lightroom AI Denoise." simonsonghurst.com. https://www.simonsonghurst.com/dxo-pureraw-6-vs-lightroom-ai-denoise (2026)
- CaptureLandscapes. "DxO PureRAW 6 Review." capturelandscapes.com. https://www.capturelandscapes.com/dxo-pureraw-review/ (2026) [Independent third-party testing — methodology: visual comparison across camera bodies at multiple ISO values]
- Kolari Vision. "Understanding and Fixing Color Banding Issues in Photography." kolarivision.com. https://kolarivision.com/understanding-and-fixing-color-banding-issues-in-photography/ (2026)
- Photography Stack Exchange. "Why do strange purple blotches appear in shadows when using a Camera RAW profile?" photo.stackexchange.com. https://photo.stackexchange.com/questions/98187/ (2026)
- PMC / MDPI. "Overview of Research on Digital Image Denoising Methods." pmc.ncbi.nlm.nih.gov. https://pmc.ncbi.nlm.nih.gov/articles/PMC12031399/ (2025)
- MDPI Applied Sciences. "Impact of Traditional and Embedded Image Denoising on CNN-Based Deep Learning." mdpi.com. https://www.mdpi.com/2076-3417/13/20/11560 (2023)
- Paolo Sartori Photography. "Lightroom AI Noise Reduction 2025: Why It's a Game Changer." paolosartoriphotography.com. https://www.paolosartoriphotography.com/blog/2025/6/game-changer-lightroom-ai-denoise-no-longer-creates-dng-files-heres-why-that-matters (2025)
- DxO Forum. "The purpose of DeepPrime 3 (Pure Raw 5 vs Pure Raw 4)." forum.dxo.com. https://forum.dxo.com/t/the-purpose-of-deepprime-3-pure-raw-5-vs-pure-raw-4/49965/ (2026)
