Gigapixel v6.1.0

yes, that’s what I was doing, but it was making too much pain (i have damaged hands, using a wacom touch as a mouse to minimise pain).
now i’m doing the crop outside of Gigapixel until it’s fixed ! Thx !

I’m trying it out, but I’m having the same problem. This makes me waste time when dealing with a lot of images.

Windows 10 21H1
AMD 3900X
Nvidia RTX 3080

After further investigation, it seems related to Face refinement. After I turn it off, I can save images. Reinstalling didn’t help. Online/offline models didn’t help. As I said, the preview works fine. Part of the image is saved. Problems occur going from top to bottom - just where the face starts.

My support request is ongoing and the issue is being investigated.
If you inspect the log files (just open them with notepad), you’ll see there are errors when the engine tries to load a model, it fails to choose between TensorRT and CUDA. It then loads the same model three times in a row, and that may be why the loading hangs for so long.

It would be great if you could make a request and send your logs too, so they can fix it faster.
Also it’s a good thing that you have an RTX card while mine is a GTX, and both of us are experiencing the same issue.

For my part the logs say :

22:30:37.740 INFO [0x155c] Cancelling processing (when the model is changed and preview initiated)
[…]
22:30:47.069 DEBUG [0x155c] TEffect procFinished (when the preview has finished processing)

First of all, it tries to load a TensorRT model which is not compatible with my GTX card, and also fails to switch to CUDA, and that alone takes near than 3 seconds.
Then it eventually initializes properly and now starts loading the exact same model three times in a row, which takes 2 seconds each, for a total of another 6 seconds.
When models are finally loaded, it starts processing and that step only takes less than one second.
It took a total of almost ten seconds to complete the preview !

Never mind, the disk was full.

1 Like

We got the described issue. We will try to optimize it in a following patch of v6.1.

2 Likes

And concerning my problem, will you also make a patch?

My problem was this:

In version 6.1 there is still the big bug that I reported in 6.0. That is to say that as a Photoshop plugin it does not manage the transparency of the layers. Indeed, when sending to Gigapixel, my transparent background becomes white in Gigapixel and remains white after applying the modifications and returning to Photoshop. As if the layer was flattened. This bug appeared with the 6.0 update. And the only temporary solution I had to solve the problem was to make a mask that takes the shape of my object on the layer. I didn’t have to do this on versions before v6 because it preserved the transparency of the layer without the use of layer masks. In standalone and external mode, it works well though. Only plugin mode has this problem.

I am having trouble with the app becoming bloated. GPAI is now more than 10GB, mostly in a folder labeled tgrc inside the package. Should I reinstall from fresh? If this is not a normal size, why isn’t the installer cleaning the folder? Thanks.

1 Like

Uninstalling nd then deleting files from the tgrc folder then re-install should clean up. I moved older .tz files (2021 date) to another folder and then used GAI. It ran with no problem so I will now delete them. I kept the tz files that had real names since they didn’t have a newer version.

Thanks. I used a symbolic link to move the bloated tgrc folder to an external drive, but It’s nice to know that I could reinstall and get some relief from the ever expanding app. Why can’t it clean up after itself???

1 Like

I’m certainly not alone, and just as a suggestion, but no matter the models / settings tried, not one single take care correctly face of Readhead women.
even the standard one with everything turned off (blur , denoise at zero), remove the freckles, seeing it certainly as “noise” and they are cleaned from the face.
Hope future training / workaround can be done in a future update / version.

1 Like

I agree. My tgrc folder had 11.8 GB of files. After some clean up it is still 6 GB. I would guess that all the new instal has to do is delete the tgrc folder at the beginning of the install, seems easy. Of course you would have to download the models again.

1 Like

With freckles you have to do two upscales (with and without face enhancement) and airbrush the freckles back in manually.

I think this is one area AI may never get right as freckles and noise can look almost identical, especially in a lower resolution shot.

hey, thanks , i unfortunatly know nothing to picture manipulation (photoshop or a other software), what airbrush them manually mean ? how it’s done ? (maybe a link to a youtube’s tutorial to do this ?) thanks.

One method is to have two files (same picture) open in your photo program. One goes though Gigapixel with Face Refine on and the other with Face Refine off (to keep the freckles ). Then you copy and paste the one with no freckles over the one with them. You then use a mask to bring back the freckles.
This may seem complicated to you if you are not experienced in this but it is not hard.

The other method is just to use a paintbrush to paint some freckles on a new layer.

1 Like

i understsand what you mean, explained like this i know how to do it (first method).
but doing it on +100 pictures one by one, i gave up, i’ll use no face recovery on this model.
eventually I’ll do it properly on something more important (like a friend face lol).

(the painbrush method is out of my capacity).

Can anyone confirm having issues with the overall speed processing when face recovery is turned on even with super low resolution files ?

I have watched the Topaz tutorial video on Youtube and it seems to not take that much long on their MAC.

Despite having a beast PC with a RTX 3090 face recovery is very slow to process.

There are a lot ov variables such as picture size , number of faces, etc. In one test case I upscaled a photo with two faces 1.5X and it took 5 seconds with FR off and 25 seconds with it on. That is a large difference but 25 seconds is not real slow unless you are processing a large batch of files. I have a Ryzen 5 2600X CPU and a Radeon RX 5600 XT GPU (6GB VRAM). In preferences I have the AI Processor set to Auto. Let me add that the picture was set at 50% zoom and I used low resolution model.

Is there a quality difference between low and high GPU model? On low I get much faster results.

Face recovery has to run an AI process for every single face found so you can imagine how quickly that can add up both in RAM and GPU time.

I tried to do a 1080p school photo with about 100 faces in it and my Mac Mini 8GB just seemed to get stuck. My Macbook Pro (M1 Pro 16C) 16GB did it in about a minute and a half, my PC with an RTX 2080 and 32GB RAM took 23 seconds.