Gigapixel v6.1.0

It is a similar model downloading issue as some others have mentioned. Basically no model is found so no preview update or image save is valid. We will fix this but before done a few things can be done:

  1. if used in-app installer. close all the gigapixel programs. download the full installer, and install again.
    if does not work, then
  2. contact our support team by the chat button here https://support.topazlabs.com/. Ask them for the full collection of models.

help->install missing models

Note some users have reported downloading interrupted/stopped when downloading all the models. If that is the case, contact the support team as I mentioned here

Iā€™m impressed with the new face recovery AI in 6.1. No issues installing and was able to download all the models in one go. Nice work Topaz.
Bike11

2 Likes

My connection speed is about 50 mbit/s and this is the maximum in this small village. ~6MB/s download speed and from gigapixel servers come only 1MB/s. So itā€™s not good for me. More comfortable if they upload the full gigapixel pack with AI models. And download once.

you can try:

help->install al missing models
Or
Please chat with our support team from here https://support.topazlabs.com/. They have a full collection of our models and can send to you on request.

This is strange because Iā€™m also experiencing this. When switching models or scale it says ā€œloadingā€¦ā€ for more than 5 seconds, especially with GPU, and sometimes the program hangs and shows ā€œdoes not respondā€ for a very brief moment before finally starting to process.
It was much more responsive with the previous versions, even though there always was a ā€œloadingā€¦ā€ delay, but not that long and without the unresponsive window.

For information Iā€™m on i7 9700K and GTX 1070Ti.
Denoise and Sharpen also have a significantly longer waiting time when switching models with GPU than it is with CPU.

Also I feel like I have some models in double. Is that normal that the Gigapixel models folder is 16 Go by doing the updates from February 2022 to now ?

1 Like

The picture doesnā€™t have a whole face in it as shown in the upper right corner. Canā€™t expect the AI to recognize it correctly.

Gigapixel often causes unwanted color shifts, especially the Low Resolution and Art & CG models. If you are using GAI as a Photoshop plug-in, it puts the new upscaled image in a ā€œgigapixel-layerā€ over top of the original background, which is scaled up using Photoshopā€™s regular Bicubic interpolator.

So hereā€™s a stop-gap measure to use until Topaz gets this color shift fixed: To get my original colours back I set the blend mode of the gigapixel-layer to Luminosity. That allows the original colors to show through but with the new details. Itā€™s not perfect, but usually pretty close and better than having clobbered colors.

1 Like

Exactly. It seems that model loading was much faster in the older versions of GAI. This slowness discourages me from trying out the different models on my images. BTW, this is without the face refinement option.

Face recovery is good, however fine hair strands become thick and chunky. Hoping this will get fixed.

I think hair still mostly relies on the main models which is very hit and miss, especially the Standard model. Its somewhat frustrating that the Low Resolution model often does hair better but then loses detail in the rest of the image.

I have contacted the support about it.

I donā€™t know if this will help but try using a VPN ( I use Cyberghost) and connecting to a server in the USA. This my bypass the slow circuit. The problem is not with Topaz servers which are fast. I have a 400 Mb/s download internet and I can download gigapixel 6.1 in less than a minute.

The installer download is fast, but what @dency01 is talking about is that model download from inside Topaz software is extremely slow.
I also have a 400 Mb/s download speed so the .exe downloads at 50 MB/s, but the models will download at 6 MB/s at great maximum. It sometimes even go as low as 1 MB/s.
It is actually intended and if you ask support they will give you faster links for model download.

Just tried updating a different computer in-app from v5-something, failed (very fast connection BTW):

Screen Shot 2022-05-16 at 4.53.02 PM

Uninstalled the older version and downloaded the full app from the top of this thread. I notice it is asking me to put the app directly in Applications (Mac), whereas the older version was in Applications>Topaz Labs LLC>Topaz Gigapixel AI

Screen Shot 2022-05-16 at 4.55.41 PM

Fresh version seems to be working fine, now I get to download models again :wink:

1 Like

Iā€™m not talking about the face (which has been correctly detected and processed). Iā€™m talking about long hairs. Look at the area below the subjectā€™s shin, thereā€™s this clearly visible cutoff between enhanced and non-enhanced hair. Thatā€™s why I mentioned masking that would make this cutoff less noticeable.

By the way, some hair enhanement tool within Gigapixel would also be welcome.

3 Likes

Yeah its very variable presumably based on how many portraits they trained with a lot of hair in shot vs just faces. I had one that was pretty awful where the facial enhancement blurred the hair then when it went outside the facial area the normal AI had it in focus.

I rather expect this will improve as they continue to train the model, though not sure how they can make it work with really long hair.

My internet plan has data caps. With that said, I just installed 6.0 the other day. A prompt appeared to download model files (2.76GB), which I did. Just installed 6.1 and the same prompt appears again to download model files. Why? Unless the models have changed, I donā€™t see why downloading the same files is necessary?

The first post says the models have been updated in this release.

1 Like

Thank you so much for this wonderful post! I love seeing all of the success stories that our users are sharing with the new Face Recovery model.

If I could put in a request, could you add a link to our update post that explains more of the technical details that went into building Face Recovery AI? Iā€™d love to get more people to see that article because it complements your post really well.

Regardless, thank you again for taking the time to share it!