In the preferences you can select the “strength” of the models. Adjusting these may fix your problem.
It’s actually ridiculous that there’s no option to disable logging - it’s like we are running a debug version of the application instead of the release version.
This is a hill I will die on! ![]()
Same. It’s doing it’s thing for a few seconds, but right before it shows the results, it crashes. I had to downgrade to the previous version which doesn’t have the updated upscale models.
Please submit a bug report to the support team so that they can look into this. I did. The more people submit one, the higher the chances of them fixing this problem.
Forget what I wrote here, what TPAI enhance depicts here is probably more accurate than the original.
No kidding, these are steel bars photographed from the side and they are flat and not round or pillow-shaped.
These are roof supports from the Century Link field.
Hi …
I have been trying to solve the problem from my point of “view.”…
It seems like the system-pagefile was too small…
I only changed this… from 8 GB to the system recommended size to 16384 MB…
- and it works for me fine… haven’t any crashes after that…
Perhaps it can help others
)
My pagefile is system managed, and is currently at only 2048 MB. TPAI has 32691 MB of system memory and 8192 MB of video memory to work with, and that seems to always be enough for TPAI the way I’ve been using it.
I encounter the following Issue with the new version v2.3.0:
Whenever I upscale Images using one of the two new Models (Standard v2 or High Fidelity v2) the export will freeze after the “processing” step the progress bar will stay at 100% but not continue to the “saving” step. I can wait forever and it won’t continue. This problem only occurs on the two new models, I don’t have any issues with the four old models.
What I already tried:
- Switching between CPU and GPU processing,
- Toggled the “Allow these images to be uploaded” back and forth
- Changed the input and output directory between multiple separate hard drives,
- Tried each of the options for the output format (jpg, png, tiff, etc.),
- Tried it with single images as well as with bulk exports,
- Used a variety of input and output resolutions,
- Tried it on a variety of image sources (photos, ai-generated, graphic designs)
… but nothing changed anything about the Issue.
My system is:
Mac Mini m2 Pro (2023)
32 GB RAM
2 TB Hard drive
16 Core CPU/GPU
MacOS Ventura - Version 13.3.1 (a)
Topaz Version is obviously 2.3.0
I really hope anyone can help, so I can use this badly needed new models ![]()
Can you drop me a log file for using the two models?
Open the app and running the two models using any image → wait a bit when it got stuck → Help → Open log folder → Get the log file that has the most recent modified date.
Have you enabled upscaling function when you had crash? Do you see crashes without running any upscaling models?
No, I haven’t got upscaling enabled when the crashes happen. I downgraded to the previous version 2.2.2 and I no longer get any crashes. Seems to be a problem with the new update 2.3.0.
Then it is not upscaling models related. There is removal tool model changes too in 2.3.
In the Adobe support documents, they advise that the pagefile should be 3x the size of the ram, for Photoshop.
Please, point me to that.
What I see are just many obsolete “recommendations” at Adobe site, some contradicting to each other. e.g.,
April, 2021: Create a paging file with maximum size of three times amount of installed RAM | Windows XP
And another from the same April, 2021: Set the virtual memory paging file to the default size | Windows XP
Still, none of these is “recommended”, they just describe how one can make these sizes if desired.
I believe that 30 Mp is physical limitation which Canon cannot overcome to make other specs up to speed, e.g. computing/writing. As discussed at the same rumor site.
Otherwise, one could say that 10 Mp is better than 20, 20 better than 30, and so on.
If it would be that optimum, Canon would never make 5DS, R5, as well as other brands which make >30MP FF sensors, or crop sensors >12MP.
I did not really understand all your calculations of “real resolution” of 96 and 180 Mp, but it is good that you have time for that.
Lets say you have 3x3 pixels.
To really make a difference in resolution you need 6x6 pixels at the same space size of the original 3x3 pixels.
I did spend some day with the EOS R5 and did end up with the R3.
The R5 did look very soft and noisy, its pixels are already to small for my taste.
Diffraction does drop in at f/4 (R5) and after this its resolution does become less.
Yes these suggestions are very old but still up to date.
I did remind me to these but didnt search for it.
It’s not about taste, it’s about detail/noise signal that a sensor can resolve, which potentially can be further improved in the postprocessing using algorithms where the real details are recovered rather than created de novo by AI.
Assuming that a lens is capable to resolve at that level too.
As for diffraction… Do you want to tell me that you photograph your butterflies with open aperture?
So, you do not point me to the valid current recommendation, e.g. for Photoshop?..
Because those that I referenced above are not only for Windows XP but also talk about such amounts of RAM like 64 Mb, who on earth has such RAM at present? And, Adobe copied this section to their recent recommendations for improving performance of Illustrator but I did not see anything relevant for Photoshop.
Sometimes yes. ![]()
f1.2, f3.5, f4.0 depending on the lens.
Its impossible to create my images without AI denoising, bc color noise will kill off all the color details, and “advanced algorithms” do smear all the color details away, bc with a single image its impossibe to restore color details.
But you are free to get an image from me and try your best if you want and show me those algorithms, i’m curious.
With AI its like to have a next level camera.
The only images that are better without AI are Astro images.
Hi all - I have created a request to allow users to disable logging - please vote for it: