It never was to be honest. I can’t speak on it too much as its protected IP but the engine update in this version has allowed us to try a few new things with our models that should get rid of the strict model scales.
1 Like
I hope so, because 4x is not sufficient for me, for my Neural Style Transfer images, I need at least 5x or more.
I re downloaded Gigapixel AI 5.5 from this link, and have the same results. I can open the software and it seems to work fine, but if I start to try the other modules or make changes to the Auto settings, the software quits. Trying it out as stand alone, I haven’t used it as a plugin. Mac OS Mojave.
Taylor, I just tried this Beta version on my 2017 iMac 27" running High Sierra, and it worked! There was no crashing at all. I tried the new Low Compression on an older jpg and it came out nicely. Also, I used a raw file and cropped it then did a 2x on the Standard, and there wasn’t a crash either. Thanks so much!
Hi Taylor the Beta version works on my Mac Mojave . But now only with a watermark , unless I buy it : But I own Giga Pixel already . ( Do not want to pay twice for it ) So what now
@taylor.bishop I believe these FP16 OpenVINO models were accidentally left out from being uploaded to the server:
Standard:
http://models.topazlabs.com/v1/ggn-v1-fp16-256x256-1x-ov.tz
http://models.topazlabs.com/v1/ggn-v1-fp16-384x384-1x-ov.tz
http://models.topazlabs.com/v1/ggn-v1-fp16-256x256-2x-ov.tz
http://models.topazlabs.com/v1/ggn-v1-fp16-384x384-2x-ov.tz
http://models.topazlabs.com/v1/ggn-v1-fp16-256x256-4x-ov.tz
http://models.topazlabs.com/v1/ggn-v1-fp16-384x384-4x-ov.tzArt & CG:
http://models.topazlabs.com/v1/ggi-v1-fp16-256x256-1x-ov.tz
http://models.topazlabs.com/v1/ggi-v1-fp16-384x384-1x-ov.tz
http://models.topazlabs.com/v1/ggi-v1-fp16-256x256-2x-ov.tz
http://models.topazlabs.com/v1/ggi-v1-fp16-384x384-2x-ov.tz
http://models.topazlabs.com/v1/ggi-v1-fp16-256x256-4x-ov.tz
http://models.topazlabs.com/v1/ggi-v1-fp16-384x384-4x-ov.tzLow Resolution:
http://models.topazlabs.com/v1/gmp-v1-fp16-256x256-1x-ov.tz
http://models.topazlabs.com/v1/gmp-v1-fp16-384x384-1x-ov.tz
http://models.topazlabs.com/v1/gmp-v1-fp16-256x256-2x-ov.tz
http://models.topazlabs.com/v1/gmp-v1-fp16-384x384-2x-ov.tz
http://models.topazlabs.com/v1/gmp-v1-fp16-256x256-4x-ov.tz
http://models.topazlabs.com/v1/gmp-v1-fp16-384x384-4x-ov.tzLines:
http://models.topazlabs.com/v1/gde-v1-fp16-256x256-1x-ov.tz
http://models.topazlabs.com/v1/gde-v1-fp16-384x384-1x-ov.tz
http://models.topazlabs.com/v1/gde-v1-fp16-256x256-2x-ov.tz
http://models.topazlabs.com/v1/gde-v1-fp16-384x384-2x-ov.tz
http://models.topazlabs.com/v1/gde-v1-fp16-256x256-4x-ov.tz
http://models.topazlabs.com/v1/gde-v1-fp16-384x384-4x-ov.tz
It installed and did not crash on the first few attempts. I will continue to test.
How will I know when the final version is available?
Hitchcock, when saving the image there was a box that popped up about the Watermark, but it also had a way to LogIn, so I just logged in and saved the image as usual.
1 Like
Hmm, looks like they are there for fp32. For Video Enhance AI, OpenVINO uses only fp16, which is why I thought they would be fp16 for Gigapixel as well. But I guess for Gigapixel they have to use fp32?
Again, check the .json… does it specify fp16 or fp32 for OpenVINO, or both? All of those URLs work if you change fp16 to fp32.
If you are in Ps you can do an easy, accurate crop in the Ps layer after using the GAI plugin. Or, you can crop in Ps prior to entering the GAI plugin.
There is no option to do that with the standalone, so it must be more self-contained.
I actually found them in the json files originally, otherwise I wouldn’t have guessed both openvino and coreml have fp16 & fp32 models.
These are the only fp16 models present in the server atm, and that’s what prompted me to think that the others were missing:
Standard:
http://models.topazlabs.com/v1/ggn-v1-fp16-192x192-1x-ov.tz
http://models.topazlabs.com/v1/ggn-v1-fp16-192x192-2x-ov.tz
http://models.topazlabs.com/v1/ggn-v1-fp16-192x192-4x-ov.tzArt & CG:
http://models.topazlabs.com/v1/ggi-v1-fp16-192x192-1x-ov.tz
http://models.topazlabs.com/v1/ggi-v1-fp16-192x192-2x-ov.tz
http://models.topazlabs.com/v1/ggi-v1-fp16-192x192-4x-ov.tzLow Resolution:
http://models.topazlabs.com/v1/gmp-v1-fp16-192x192-1x-ov.tz
http://models.topazlabs.com/v1/gmp-v1-fp16-192x192-2x-ov.tz
http://models.topazlabs.com/v1/gmp-v1-fp16-192x192-4x-ov.tzVery Compressed:
http://models.topazlabs.com/v1/ghc-v1-fp16-192x192-1x-ov.tz
http://models.topazlabs.com/v1/ghc-v1-fp16-192x192-2x-ov.tz
http://models.topazlabs.com/v1/ghc-v1-fp16-192x192-4x-ov.tzLines:
http://models.topazlabs.com/v1/gde-v1-fp16-192x192-1x-ov.tz
http://models.topazlabs.com/v1/gde-v1-fp16-192x192-2x-ov.tz
http://models.topazlabs.com/v1/gde-v1-fp16-192x192-4x-ov.tz
Thank you , that worked . When will there be an update so I can use it with Mojave ? ( And strange that the Beta version works, but not 5.5 ? , How come ?)
1 Like
Taylor, I just tested your latest 5.5.0-maccrashfix and confirmed that it installs, launches and works fine when switching modes, processing and saving in Mojave (2013 i7 MBA).
It did offer a confusing update which I ignored:
I’m curious if this installer is something new you just created or if it is the next-to-last beta, which didn’t have the issue in Mojave during testing.
I was reading on internet that fp16 models has faster processing and use less memory but they offer less accuracy then fp32. I wana ask is there any actual difference in quality between fp32 and fp16 and can we force rtx gpu to use fp32 instead of fp16 model.
You were reading about 3D graphics. When it comes to 2D graphics FP16 is way more (higher precision) than you’ll ever need.
Color is very muted in 5.5.0 after processing. (It seems to be the low res model)
Yes, you can actually trick gigapixel into using the fp32 models instead if you edit the .json files in the models folder.
I gave it a go and noticed that in some instances, the fp32’s produced a fairly crispier and more detailed image, especially around the eyes and hair, while also handling noise and compression artifacts more smoothly. But for the most part, the difference is barely noticeable (you need to REALLY zoom-in to catch it) and sometimes even, inexistant.
Here are the tweaked .json files if you decide to give it a whirl (w/ fp32 models included, it’s on me 
): https://anonfiles.com/544369s8ue/force_fp32_zip / .json only
Make sure to backup your stuff first, then extract in “C:\ProgramData\Topaz Labs LLC\Topaz Gigapixel AI\tgrc” and replace everything when prompted.
I suggest you take the time to prepare some upscales with the fp16 models beforehand for comparison’s sake. It’ll help you make up you mind on weather the boost in image quality is worth the speed reduction or not. Because to me, hell yeah it does!
2 Likes
Thanks I have already downloaded the models and will try this.
Ah… Sorry for not including a link w/ only the .json files. Better late than sorry I guess: File on MEGA
