Gigapixel AI 4.2.0: Ugly Degaussing/Rainbow Effect on Enlargements

I think there’s something very wrong with the latest version of Gigapixel AI. I’m seeing some kind of weird rainbow-like smearing on enlarged photos, similar to the degaussing effect you see on old CRT monitors. It seems to happen more often at the 6x enlargement setting. Here’s what I am talking about:

I never saw this problem in the previous version of the program. Also, you guys spelled “Manual” as “Mannual” in the processing settings panel.

If it appears in the saved image raise a support request on the Topaz Labs website with the settings, your environment (Help-> Graphics info and copy and save to a text file), and the settings used for the enlargement process.

No idea how relevant it might be but the Changelog now says dated 19th July v4.2.1 is the current one. With no mention of v4.2.0

It would be great if the Topaz team made the log more explicit like the older style one.

I for one am going to let the dust settle a bit on this update?

I just updated. The spelling error is now fixed, but I hope 4.2.1 has more changes than that…

I think the spelling error is the only thing that was changed since I’m seeing the same issue. Changing from GPU to CPU seems to get rid of the rainbow effect, but scaling results are dramatically worse.

Sadly my trial ends tonight, so I won’t be able to do much more testing…

@bitfidelity In the Gigapixel version I see a little colorization around her lip but nothing severe. Even the original has some blue color areas. While not perfect, I think you are being a bit picky since it greatly sharpened the enlargement. If you want an alternative try http://a-sharper-scaling.com

I have the same problem here too with some photos. When it happens it’s rainbow colored horizontal bands from top to bottom. I noticed it only happens with GPU processing and with just a few photos. CPU processing seems to be fine.

What I notice the most is the color banding around her shoulder. These appear to trace back to the part of the calculation that is determining where there are edges that need sharpening. If the original contains any hint of color banding artifacts it amplifies the patterns into the consciously visible range.

Ok, I think I narrowed the problem down. Turns out this happens only with GPU and Max AI models turned on. If Max AI models is turned off with GPU or if it’s on BUT with CPU, the bands don’t appear. I will submit a ticket to report this.

Thank you.

Ticket submitted.

1 Like

@svetlina_design

Thanks for your testing & insight :slight_smile:

Hi Svetlana
You said
“Ok, I think I narrowed the problem down. Turns out this happens only with GPU and Max AI models turned on. If Max AI models is turned off with GPU or if it’s on BUT with CPU, the bands don’t appear.”

You have discovered a user workaround solution as per others findings ~ Problems with the quality of the final enlarged image has been raised a few times as I recall could have been both version 3 and version 4

Your tests have proven and raised again that using Max AI Models in the settings when GPU processing is selected is not the best choice of settings. But as far as I recall there has been no comment or feedback about selecting Max AI Models or not by the Topaz team?

Because it has come up a few times across a few (how many ?) versions I do wonder if Topaz have identified the cause and when there will be a full solution & explanation?

Afteral, if Max AI Models are supposed to give us the best processing speed & output but are failing to do so, what is that set of Models for? And under what conditions are we supposed to be able to select that setting, making that more obvious to both old & new users will likely save a lot of cries for help such as this one :wink:

Sorry, I’m not sure I understand you. Can you please rephrase your reply?

Hi Svetalana @svetlina_design

I have edited my reply and hope that makes more sense? :slight_smile:

By the way Topaz Labs has nothing to do with DxO, totally separate company.

Thank you! :slightly_smiling_face: Although I was kind of confused why you talk abot DxO which is a different company AFAIK.

BTW, I’m seeing regression in quality with CPU upscaling too which historically always gave me better results than GPU. After round about version 4.0.3, I noticed more and more artifacts in my images - blocky clusters of pixels like JPEG compression artifacts (even though I save as TIF) and more noise even at high denoise levels. From my tests, I found that Gigapixel 4.2.0 is by far the worst of all in this regard so I uninstalled it and went back to Gigapixel 4.0.3. Now my CPU upscales look great again. I may create a thread about this with samples and raise a ticket.

Ooooooops! my bad it was late when I first posted, that is my excuse and I’ll stick to that :lol:

Reminder to self ~ proof read my replies more carefully !

PS of note ~ I also do have DxO software and the forum structure they use is identical…so being tired it appears I just got mixed and even managed to leave my error until now uncorrected. Phew!

PPS I hope I have corrected all the wrong attributions OK

As posted in reply to Don…a slip on my part for a couple of reasons :slight_smile:

Hello,

After I submitted a ticket I was given a special test version of Gigapixel to try, but it didn’t change anything. I mentioned this in my reply to the person that gave me the test version, but I didn’t receive any reply from him for two weeks now.

The issue seems to be fixed (or at least reduced) in the newest version with face detection.