Topaz Labs | Wonder 3.5 and Bloom 2 Improve AI Image Upscaling
Topaz Labs has introduced its Definition Update with two new image upscaling models, Wonder 3.5 and Bloom 2, alongside workflow improvements for Topaz Image Web and Mobile. The update focuses on a familiar challenge for designers and AI creators: increasing resolution and detail without making textures look artificial or pushing an image too far away from its original appearance.
Topaz expands image enhancement with Wonder 3.5 and Bloom 2
Wonder 3.5 is designed as a more general-purpose image upscaler. It combines upscaling, sharpening, and denoising in a single process, with Topaz focusing this version on producing more detail, more natural textures, and fewer repetitive patterns than previous iterations.
That makes it relevant beyond photography. Designers working with small reference images, compressed assets, illustrations, covers, thumbnails, or other visual material can use this type of enhancement when an image needs more usable resolution before it moves into a larger composition or final export.
Bloom 2 focuses on detail in AI-generated images
Bloom 2 takes a more specialized approach. It is designed for low-resolution AI-generated images and can refine or introduce detail while trying to remain faithful to the original image. Topaz specifically targets the overly smooth or plastic-looking textures that can appear in generated content.
The model also improves the treatment of faces and skin while placing more emphasis on character consistency compared with the original Bloom model. For AI creators, that balance matters because an aggressive enhancement model can technically improve resolution while unintentionally changing facial features, proportions, textures, or other identifying details.
Branching Workflows make multi-step enhancement easier
The Definition Update also adds Branching Workflows to Topaz Image Web and Mobile. After processing an image with one tool, creators can use the menu to add another branched workflow and continue working with additional models or enhancement processes.
This creates a more flexible way to test different processing paths instead of treating every enhancement as an isolated operation. A creator could evaluate one version after an upscale, for example, while continuing another path with additional restoration or enhancement without losing the earlier direction.
Generative Restore targets images traditional restoration cannot recover
Generative Restore is aimed at severely degraded, corrupted, or extremely low-quality images. Unlike a conventional cleanup process, the workflow can rebuild visual information, which makes it better suited to difficult restoration cases where normal enhancement methods are no longer enough.
That generative behavior also makes it important to review the result carefully. Reconstructing missing information is different from recovering detail that was actually present in the original file, so creators restoring archival, reference, or identity-sensitive material should compare the generated result with the source before treating it as an accurate reconstruction.
IMPORTANT: Wonder 3.5 is currently listed for Topaz Image Web, Topaz for Mobile, and the API. Topaz says support for Topaz Photo and Topaz Gigapixel is coming soon.{alertWarning}
Daisuki's Take: What This Means for Designers
The interesting part of this update is not simply getting another pair of upscalers. Wonder 3.5 and Bloom 2 address two different creative situations: improving a broad range of images and refining AI-generated work where preserving the original character or composition can be especially important.
For AI-assisted design, Bloom 2 may be particularly useful when a generated concept is already visually successful but lacks enough resolution or surface detail for a larger final asset. In that situation, preserving the image can be more valuable than creatively reinterpreting it during enhancement.
We would still treat generative restoration and aggressive enhancement as stages that require visual review rather than automatic final steps. The stronger these tools become at inventing plausible detail, the more important it becomes for designers to decide when added detail improves the work and when it begins changing the original.
Sources and Recommended Links
- Definition Update | Topaz Labs Official Site
