Zone-based photo restoration at the heart of Wonder 4

The new Wonder 4 model from Topaz combines upscaling, sharpening, and noise reduction through an architecture that adapts corrections to each area of the image.

A slightly blurred face, a noisy background, and textures damaged by compression do not require the same corrections. Until now, restoring this type of photograph often meant chaining several processes together, adjusting their intensity to avoid one enhancement degrading another part of the image. The new generation of Wonder, developed by Topaz Labs, aims to bring these interventions together in a single model, capable of determining the necessary corrections according to the different areas of a photograph.

Available since November 3, 2026, Wonder 4 introduces an architecture known as Mixture of Experts, or mixture of experts, applied to image restoration. In its official November announcement, the publisher explains that the model analyzes several regions of the same photograph and applies the treatments suited to their characteristics. Upscaling, noise reduction, blur correction, and detail recovery are thus grouped into a single intervention, without manual adjustment of their intensity. Restoration that adapts to the different areas of the image

Classic photographic restoration tools generally work through a sequence of operations. A first step removes noise, a second improves sharpness, then a third increases the image dimensions. The order and intensity of these treatments can significantly influence the result, especially when the photograph contains areas of very different quality.

The new architecture chosen by Topaz virtually divides the image into thousands of regions to be analyzed. According to the explanations provided on the publisher’s official forum, each region calls upon the corrections it needs. A blurry subject can therefore receive a different treatment from that applied to a background that is already sharp but affected by digital noise.

The process combines five functions: upscaling, noise reduction, blur correction, sharpening, defect removal from compression, and facial detail recovery. These operations are coordinated in a single pass rather than carried out as independent filters.

This organization also aims to preserve texture consistency. In photography, excessive correction can produce artificial edges, erase natural materiality, or reconstruct details that did not exist in the original image. Coordinating the treatments within a single model could limit these effects, even if Topaz does not publish an independent numerical comparison making it possible to measure precisely the improvements compared with previous versions. Fewer settings, but also less control

The main difference from traditional tools lies in the absence of adjustment parameters. The user imports their image, selects Wonder 4, and lets the model decide which corrections to apply. This approach is aimed in particular at people who want to quickly enhance photographs without mastering the different restoration steps.

The trade-off concerns control over the result. An old photograph may contain intentional grain, slightly soft areas, or textures whose imperfections are part of its identity. An automatic correction risks altering these characteristics, even when the resulting image appears technically sharper.

The first reactions posted on Topaz’s forum reflect this difficulty. Some users praise the results of the new model, while others still prefer Wonder 3.5 and regret the lack of settings that would make it possible to modulate the reconstruction. These testimonials remain anecdotal and do not constitute a representative assessment of its performance.

The previous generation notably offered several intensity levels to strike a balance between fidelity to the original photograph and more pronounced reconstruction. Topaz Photo’s documentation describes this distinction, while also noting that generative models can produce details not present in the source. The new automatic processing does not eliminate this risk: it simply delegates more decisions to the model. A new HDR conversion without altering the original image

The November update also includes Image HDR, a separate feature designed to convert standard dynamic range photographs into images compatible with HDR displays.

The processing analyzes the visual information in the photograph and produces a gain map used to extend brightness and contrast on compatible screens. This information is embedded in an HDR JPEG file, using a method compatible with the ISO 21496-1 standard and Google Ultra HDR or Apple Adaptive HDR devices.

Unlike a conventional adjustment of brightness levels, this representation preserves the standard version of the photograph. On a screen that does not support HDR, the image therefore appears without the extended brightness, while a compatible device can reproduce more intense highlights and wider tonal variations.

The conversion remains a computational interpretation of the initial photograph. It does not necessarily recover brightness information absent from the source file, but instead produces an estimated HDR representation based on the available elements. Limited availability on the web and mobile

The Wonder 4 update is currently available in Topaz for Web and Topaz for Mobile. The publisher has not yet confirmed its availability in the Topaz Photo or Gigapixel desktop applications, which have their own versions of the Wonder models.

The web service also benefits from several practical improvements: batch uploads without a limit on the number of files in TAR archives, preservation of the JPEG format when it is used as input, and improved progress indicators during processing.

The model is also referenced by third-party inference services. The Runware documentation indicates support for Wonder 4 for upscaling up to a sixfold factor, while its availability via an API opens the possibility of integrating restoration into automated processing workflows.

For Topaz users, both new features are accessible from the web and mobile interfaces: Wonder 4 for automatic restoration and Image HDR for creating photographs adapted to