Also, I forgot. There is also another parameter that can be modified. It is the scale. If I remember correctly, you can go up to 8x. But you really need a good machine to do such a high scaling. For my part, I often use 1x and 2x. In some cases 3x but not more for my use. Most of my photos that I improve are already at high resolutions.
1 Like
Yeah, the 80% does a nice job on photos and seems to retain facial features without going nutsy creative. There’s also more consistency from head to clothing (sharpness wise). Noise cleanup was effective too.
Were you able to get batches to work okay? I don’t batch process. So that’s one less thing for me to contend with.
Thanks, and I have been sticking at 80%. ‘cause you said so ;-).
Batching is mostly working, sometimes weird stuff happens and I need to redo some. I would still love to clear the finished queue though to avoid unintentional re-do’s and to reduce clutter.
I have tons of these old group photos I am currently working on. Back in the -80s and -90s I sold prints of these ;-).
Well, ppl (and companies…) generally do better if they take my advice! 
Agree, the queue cleanup is an important roadmap item…
Pls tell me it didn’t put little yellow squares around all the recognized faces!!!
Some of them look like best guesses (but sorta mutant) by the robot. But there are some decent salvages.
I don’t preview first.
Here is without Face recovery, maybe not much difference (all renders at 2X):
Sometimes I restart GP between batches to freshen it up; I just had to now because it crashed…
A bunch of the more featureless originals end up with the same “mask” applied to the people’s faces. Big nose, high cheeks, smashed mouth. Seems like a horror movie mask (or the lead from Phantom of the Opera).
Yeah, my Wonder renders from medium format film are more realistic for the faces.
1 Like
Do you have to downsize for processing med format or does it just handle that size?
In PS I mashed down my original (way too big) hi-res film scans to 25” W @ 300ppi (about 7500 pixels wide).
1 Like
Considering the noise level in the image, I find the final result on the right quite natural.
1 Like
I think for output quality it is better not to zoom in on faces and watch in full screen
1 Like
A small suggestion that would be interesting would be to have the possibility for the Wonder model to have preview area frames like in Recover and Redefine. This way it avoids wasting time if the final rendering is not satisfactory.
Yes, many of these are quite amazing salvage jobs.
1 Like
As I recall the devs said the final render would be different anyway because of the way generative AI works.
Guess I forgot that. But it makes sense. There’s no way to replicate a gen in Ps if I want to. So, suppose it would be true here too.
Right, and also a small section vs. a full image, never will match.
1 Like
Is this new Topaz Gigapixel v1.0.2 the subscription thing that we don’t have access to when our perpetual license runs out, or is it an update to the perpetual Gigapixel v8 ?
