My disappointment is with the current state of the Redefine models. Even the two supposedly subtle models are anything but subtle. At both 2× and 4×, I’m seeing excessive artifacts, invented details, and unrealistic textures. (The older Redefine model at Low certainly didn’t turn ordinary skin into hair!)
I’m not sure when the deterioration began, but it’s genuinely disappointing, especially given how good Redefine used to be.
The Wonder models, on the other hand, are excellent in their own right — big props to the team for those. But even if Wonder is intended to compete more directly with Magnific, I still find Magnific’s results substantially better.
I’ve been a loyal Topaz user for years, so this isn’t criticism for criticism’s sake. But at this point, the main reason I’m sticking with Topaz is that it runs locally. That’s an enormous advantage, but I’d hate for it to become the only one.
With “Redefine,” there are 2 modes, and 6 modes in total. These haven’t changed since the original release.
The only changes that have occurred to these models are their presentation in the UI, the parameters exposed for adjustments, and whether Face Recovery can be used in tandem with the creative models.
“Redefine” with 6 Creativity modes was divided into Redefine realistic and Redefine creative to better convey the use of 2 models, not 1.
Sharpen and Denoise were removed as parameters due to the confusion they were causing users and the number of bad results being produced.
Face Recovery was disabled for Creative levels 3-6 because of the bad results that were occurring.
If you’re getting unexpected results with renders here are a few recommendations:
Reduce the scale factor to a lower number. If you’re using 4x, use 2x instead. If you’re using 2x, use 1x instead.
Reduce the Texture slider to a lower number.
Use a lower Creativity level. If you’re using “Low,” step down to Redefine realistic “Subtle.”
Use the Image Description and attempt different styles of prompting, always using a descriptive approach, not an instructive approach.
If you’re working with a photograph, use a different model. While it can work well to restore old or small images, Redefine’s intention was to work with genAI images.
