Adobes generative AI fill, the data is not complete, you need to noise in to fill the gaps.
And now look at this, i did just add 1% mono noise, even the tiles of TPAI Enhance are gone.
But if you drop in a big file, the magic is gone.
But if you drop in a big file, the magic is gone.
i took a raw file with auto enhance and upscale by factor two which took 9 seconds saving included and with denoise 10 seconds!
1 Like
It was important to me that you take a tiff, then the raw conversion is missing.
Also I can make a tiff in any size but not raw.
But thank you for your effort.
1 Like
Oh I am sorry I will try again! 
1 Like
I am a bit confused. There is no option âenhance standardâ. âsharpen standardâ is the closest i can find. also denoise does not offer three parameters, only two!
I tested this again but with a .tif image. It took 7 seconds for the first test and 6 seconds with denoise.
1 Like
thanks
what was the resolution of the file?
5038 x 3358
1 Like
Unfortunately, slightly dissapointed to where we are so far.
I highly recommend that TPAI focus on improving on what is supposed to be its main strength : better denoising, sharpening, upscaling and most of all, better autopilot (and not really on what other programs can already do, ie. cropping, demosaicing, âŠ).
Please find some samples of why TPAI is still very unreliable on what are supposed to be its strength.
Thanks
2 Likes
Iâm not able to do PC work yet. I just ran the attached thru this 1.4.3 rel. of PAI (Win 11, Standalone) & this is what A.P. produced all alone (Std model) - it chose Only Remove Noise. I have set all A.P. settings except Sharpen Off in Prefs. My usual setup is split screen. I still would prefer whatever gen of models were in 1.4.0.1b. I got excellent results in that rel. - just didnât like the âknobs & switchesâ & collapsed model lists in it.
I do not know the answer, but DXO, Lightroom, Photoshop, On1, and other programs use the GPU also. Do they also have this problem?
This version tends to crash in the preview often when I apply the preserve text option.
We test batches of files internally and release a beta to have users review and help catch issues as well. Our internal file collection may not include all camera models and variations in image data that can cause these artifacts.
We will continue looking for ways to improve our QA for the models like gathering more Raw files from various cameras.
1 Like
Are you able to share these examples so I can reproduce the issue?
You can securely submit your image(s) to my Dropbox using the link below. Please be sure to send me a note to let me know you sent something.
Dropbox File Request
Please attach the logs to my Dropbox link above so I can investigate. Open Topaz Photo AI, go to the menu bar on the top and click the Help > Open Log Folder menu option.
done. 
Love the product but there is a huge remaining problem: except for faces, it makes human skin look like it was formerly owned by a desiccated reptile, all âcrunchyâ and dry. This is a glaring, basic defect far more important than the procedural tweaks and cleanup that seems to be your current focus.
2 Likes
Thanks for looking into this. Perhaps this is a key, you may need to broaden the baseline automated testing. This will pay off in a long run (very significantly), as you can always run same test to ensure the baseline is clean.
You should be able to detect that results of baseline testing are same or comparable (as applicable) release to release. If this is done automatically, you can empower yourself to substantially expand the baseline coverage and thus ensure higher quality of releases, every time.
3 Likes
So Iâve done my usual thing where I leave it a few versions, before testing again, and Iâve got to say itâs still an absolute shambles. I have zero confidence in the software, and Iâm coming up to the point where you will be wanting me to renew, so nearly a year and itâs still barely out of Beta! Get the basics of the app sorted first before you start tarting it up with all kinds of features. If it canât sharpen and denoise with consistence results itâs worthless.
1 Like
