Oops… price I pay for having GPAI and TVAI Pages both open…
BTW is there a page anywhere for the upcoming iOS version, I don’t see one publicly available or in the GPAI Beta forum?
You need to post on the Video AI forum. This one is for Gigapixel.
Dakota and Angie are responding on the TPAI forum. It does seem a bit unprofessional bringing out a new release and then not being around for feedback. What if there had been a severe bug that required an immediate patch?
Hi, I am very frustrated, my Gigapixel 8.1.1. is very slow, not processing/ upscaling any pictures and crashes all the time. The app version I had before produced damaged pictures after a veeery long processing time, and I am only upscaling to x1,5 to x2. I even switch off Denoise etc., so I just need larger files, but the application just doesn´t work.
If a new release has a bad bug for you, then go back to a previous version.
When you preview the whole screen you get in fact a window telling you “render completed in …” and the you can save the result in a few seconds.
Unfortunately the given time is very rough after it exceeds one hour, e.g. just saying 6.5 hours.
Anyway, I now (quite randomly) selected two images, a low res one from WhatsApp with terrible quality and a size of 785x1000 pixels and a “normal high res” 12 MPix image from the iPhone with the standard resolution of 4032x3024.
I primarily did want to incorporate higher resolution images but seeing the time needed here for the iPhone image on my machine quickly lost interest in that 
For low-res and low-quality WhatsApp images Redefine can do wonders:
Redefine gave me this really good result:
Recover was much less impressive but had much longer processing time.
I created a thread with the results and the used images for benchmarking:
Feel free to add your numbers…
1 Like
The local computation in v 8.1.1 seems to have totally stopped working on my Mac Studio M1 Ultra. I have tried multiple clean new installs by now and I am totally annoyed and fed up, because each and every upscaling/redefine fails and never finishes, while all Mac Cores and GPUs are idle. Only Cloud computing works. But I don’t want to use any paid cloud computation at home (it is a welcome option on my notebook though!) and I currently feel betrayed, although the current behavior might be just a bug (I hope).
Just updated on my M4 Mac mini 16gb; previous version was working fine but the new version immediately fully crashes if I try to run the generative models. If I try to do a small preview it crashes immediately, if I do the full image is loads for a few moments and then crashes. This wasn’t happening with the last version.
It’s possible your 16GB memory is not sufficient. I recently purchased the 24GB M4 Mac Mini. TGPAI uses just over 18GB when I run it.
However, others have commented that Topaz has not yet calibrated (if that’s the right word) its generative AI models to take full advantage of the M chip’s neural engine. This is not something I understand, but I am just offering it as a thought.
That’s reasonable, but the previous release was working perfectly fine prior to this update. It would take about 30 minutes to run one of the generative models, which is slow, but it worked without issue so long as you were willing to wait. With 8.1.1 it just crashes immediately (presumably while attempting to load the model). I’ve tried rebooting and uninstalling/reinstalling and it hasn’t made a difference. 
Do a complete uninstall using AppCleaner. TPAI users sometimes are successful when they remove a file from the Library called CoreLib or something like that. Check out the recent thread on how two do that. I think I have also heard that .plist files associated with Topaz can be a problem.
I have only just started using TGPAI again since purchasing my Mac mini M4. Prior to that I had an M1 16GB. So can’t compare.
Also note that no one from Topaz has contributed to this thread since it opened almost three weeks ago. You could do worse than open a support ticket.
Maybe your installation is affected by the Core ML Cache folder problem described here: [RESOLVED WITH COREML CACHE DELETION] - If you get MAC OS 15 CRASH - Steps to solve this here
The link includes instructions how to solve this problem.
1 Like
The HDR effect on this very special image stings the eyes a little.
I’m a bit new to the beta program, how do you get the beta version to recognize you as a licensed user or are they for testing without saving?
My understanding is that, if you don’t have a current licence you can still beta test, but you won’t be able to save anything.
That makes sense, and it looks like you need to have the license of the base version that you are trying to use like 8 for 8.13 or so
I think it works to the first decimal point. For example, my licence for TPAI has expired. I currently have the latest version, 3.4.3. This means that, if there is a 3.4.4, I’ll be able to upgrade to it; but not to 3.5.0 and above.
4090 vs 5090 Performance
Shader Only = +33%
I’m in hope still that in FP16 and above (AI data Formats) we will see an increase in Performance.
When i look at the hopper architecture it had 51 Shader Teraflops (fp32) but 204 Teraflops in fp16.
That’s true. I tried to slightly improve a photo of a small, old stone Romanesque church (it is maintained and supposedly sometimes functional) and then (just for fun) played with Redefine. Well, I wasn’t satisfied. Redefine is apparently (only in its beta version?) a very limited generative function. One single, already worn-out Topaz Girl tries her best with the strength of her hairy arms, but she can’t quite satisfy the required demands 
.
The results of generative AI are often kind of strange. For example, the road paved with stone cubes – after enlarging those cubes (preferably in the lower left corner of the photo), a very dense development of tiny houses appears… The church looks like a 2D backdrop made of cardboard… and the unfortunate sky… I couldn’t get my prompts to work, I tried in vain, maybe wrongly. But it was just a meaningless attempt. The church, without imposing not very meaningful generative changes, came out quite well, about up to using Creativity values < 3. So I don’t know what Topaz intends with Redefine in the future.
