DxO has been doing something unusual for a few years now: shipping genuinely sophisticated AI-assisted editing tools inside software you buy once. No monthly charge, no feature locked behind a rolling subscription. PhotoLab 10, released in late 2024, continues that pattern, and the headlining addition is an AI-powered depth mask — a selection method that attempts to separate image layers not by subject category (person, sky, background) but by estimated distance from the camera.
That distinction matters more than it might first sound.
What a Depth Mask Is Trying to Do
Most AI masking tools in photo editors work by classifying pixels into semantic categories. The algorithm looks at image content and labels regions: this is a sky, this is a person, this is foliage. That approach works well when the subject you want to isolate has a clear category. When it doesn’t — when you want to brighten only the mid-ground rocks in a landscape, or pull down exposure on a crowd without touching a foreground figure — semantic classification gets awkward fast.
Depth estimation takes a different approach. Rather than labeling what something is, it attempts to infer how far it is from the camera by analyzing perspective cues, focus falloff, atmospheric haze, relative scale, and occlusion patterns. The result is a continuous depth map across the image — essentially a grayscale gradient representing near-to-far distance — which PhotoLab 10 then translates into a paintable, adjustable mask.
The limitation is inherent to the problem: no consumer editing tool is actually reading real depth data from your sensor. This is monocular depth estimation from a flat image, and it can be fooled by images with shallow depth of field, flat lighting, repetitive patterns, or scenes where depth cues are ambiguous. A telephoto landscape compressed by a long focal length will confuse the algorithm more than a wide-angle shot with strong foreground-to-background separation. That’s not a criticism specific to DxO — it’s a constraint of the underlying machine-learning approach that every implementation shares.
Where depth masks pull ahead of subject-category masks is in scenes with spatial complexity: indoor environments with foreground furniture and background architecture, street photography where you want to separate crowd layers, or environmental portraits where separating a subject from a visually cluttered but same-category background (two people, different distances) would otherwise require manual selections.
How PhotoLab 10’s Implementation Works in Practice
The depth mask appears alongside DxO’s existing suite of local adjustments, which already included subject, sky, and background masks informed by the company’s DeepPRIME denoising infrastructure. In PhotoLab 10, you generate a depth map from any image, then use a range selector to isolate a depth band — near, mid, far, or a custom slice of the gradient. That selection becomes a standard mask layer that you can invert, combine with other masks using Boolean logic, and paint over to refine.
The underlying depth estimation runs as a neural network inference step, so it’s computationally meaningful — expect it to be slower on older hardware than simple luminosity or color masking. DxO has built GPU acceleration into the process where drivers support it, which helps on recent machines.
One practical note worth flagging: because the depth map is estimated at the point you generate it, not dynamically updated as you adjust the image, it reflects the tonal and spatial information present at the raw processing stage. Significant exposure adjustments made before generating the depth map may shift what the algorithm interprets as depth cues. Generating the map on a well-exposed, color-corrected version of the image is generally more reliable.
The Non-Subscription Model — What It Means Long-Term
The pricing structure deserves a direct look because it shapes who PhotoLab 10 is actually for. DxO sells PhotoLab as a perpetual license — pay once, own the version you bought. Updates within the same major version are included; major new versions require a new purchase (with an upgrade discount for existing owners, at least historically, though pricing specifics should be confirmed at the time of purchase).
For comparison, the subscription-first model that dominates most professional photo software means your access to already-edited files can technically be severed if you stop paying — the files remain yours, but the editing environment that understands your catalog may not. With a perpetual license, the software you paid for continues to function regardless of what happens to your payment relationship with DxO. This matters to photographers who edit archives over years and don’t want their workflow tied to a recurring billing decision.
The tradeoff is that DxO’s new features arrive in annual release cycles, not as rolling updates. If a competitive capability ships in a subscription tool mid-year, PhotoLab users wait. That’s a genuine cost depending on how rapidly the AI masking landscape evolves. Given how quickly tools like Lightroom’s AI masking have been updated — Lightroom’s editable AI masks changed the non-destructive editing model in meaningful ways — the annual cadence is something to weigh honestly.
What Sits Around the Depth Mask
Depth masking is the marquee addition, but PhotoLab 10 didn’t arrive with only one new feature. The DeepPRIME XD2S denoising engine, which handles luminance and chroma noise separately at the demosaicing stage rather than treating them as a post-processing afterthought, received further refinement. DxO’s lens-correction database — built from physical optical measurements of specific body-lens combinations rather than algorithmic estimation — remains one of the more technically grounded implementations of aberration correction in the consumer editing space. Chromatic aberration, vignetting, and distortion corrections keyed to verified optical profiles are a different proposition from applying a generic profile to a lens family.
The software also retains its approach to partial editing export: you can work nondestructively in PhotoLab’s catalog structure and push to an external editor for retouching work the application doesn’t handle as well. That round-trip workflow is worth understanding before committing to PhotoLab as a primary editing environment, because the catalog and the raw processing engine are tightly coupled in ways that don’t always translate cleanly to simple folder-based file management.
Who This Actually Suits
PhotoLab 10 is a reasonable fit for photographers who work heavily with raw files, prefer owning their software outright, and want AI masking tools that go beyond sky-and-subject dichotomies. The depth mask expands what’s addressable without manual painting, though it won’t replace careful masking for complex selections — it’s a starting point, not a finished result, on anything other than the most spatially clear images.
For portrait work specifically, DxO’s combination of lens-corrected raw processing and AI-assisted subject isolation is coherent. The depth mask adds a spatial dimension that pure subject detection lacks when, say, multiple people at different distances need different treatment. The photo-editing section of this site covers a broader range of AI masking approaches across different tools if you want context for where depth-based masking sits relative to other current methods.
If you’re evaluating whether the perpetual model is worth the higher upfront cost versus a monthly subscription, the honest calculation depends on how long you stay on a given version before upgrading. Do that math against your actual upgrade cadence, not the theoretical maximum.
The one concrete next step: DxO offers a trial version of PhotoLab with full feature access for a fixed period. Download it, open an image with genuine spatial complexity — layered landscape, street scene with foreground and background figures — generate a depth map, and see whether the mask boundary falls where you need it to. That test will tell you more than any general description of how the algorithm works.