GP AI v4.0.0 - Grain Created in Output Image


I cant create a new thread unfortunately,

with GP AI 4.0 the final images have some weird grain, which the previous version of GP AI didnt produce.

Here are a few examples of what I mean:

Grain Example

I tried using CPU/GPU without any difference. Also changing “Suppress Noise” and “Remove Blur” didnt change the grain alot.

Again previous version didnt have any grain in the final images at all.

Can you please look into it @ Topaz devs?

1 Like

Not a problem. Please drop your original image and output image into so we can start troubleshooting the issue.

Josh what I see in the posted image is that there is no Profile and it is being converted from PNG to TIF with Keep Color Profile selected. It doesn’t make sense maybe the app should pick up on that.

Please go to your profile and update your username and name. The forum doesn’t allow links/uploads from strange entries.

So it’s not just me. I noticed the same thing with some images. It happens only at 6x scaling and when using GPU. Using the CPU doesn’t seem to add this grain. Also, 4x scale and lower doesn’t seem to produce this grain even with GPU. I will try sending some samples.

Thank you. I uploaded the original image, one version that was created using GP AI 3.xx and one that was created using GP AI 4.0.

You can see the grain pattern (kinda checkerboard-looking pattern) on the wall, on the ground etc.

It doesnt matter if I convert the file or keep it the same format, because the grain is visible in the preview window also.

Also the grain is present at 4x upscale, it is even present at 2x upscale, especially if the “Suppress Noise” option is disabled (none).

I hope you can replicate and fix this issue. Thanks in advance!

here are 2 examples with source file, a GP 3.xx version and GP 4.0 version of the final image:


Also after some further testing, it seems that using GPU adds alot more grain than CPU, but if you zoom in very closely, the grain is also present with the CPU (examples in the folder):

As a sidenote- the quality using CPU upscale is slightly better than using the GPU (in my Case GTX 1060, 6GB, latest NV driver as of march 28).

The cpu upscale removes Aliasing/combing artefacts better, which the GPU upscale doesnt do to the same degree and the CPU upscales some small detail better. The AI using the CPU seems to be slightly “smarter” than using the GPU, but process is quite a bit faster using the GPU. I was using the enhancements-> Remove Blur “high” and Suppress Noise “high”. And so after some testing, for the best quality possible, the CPU seems to be the better choice if processing time is not a big factor.

You can check the examples in the linked folder and compare them zooming in to original size.


That it does though I think it isn’t information the GUI provides to the user. Maybe some extra visibility we can add in the future.

I can confirm the same issue with Nvidia 970 (4 GB) of VRAM - the effect (pattern) changes when you change amount of ram usage in the settings. Also preview is not accurate - on finished images the sharpening is not as strong as in the preview. CPU algorithm seams to work better and the inaccuracy is less noticeable, but it is still visible.

Does more RAM make the grain better or worse?

Makes no difference I have 32 gb of ram in my system.

I think that the question is about vram (video ram) - It makes difference - the pattern is more visible with lower vram settings but its obvious even on high.