Sono2000, we told that you don’t tell the name of the software publically… you broke this… it’s not very kind.
the licence you tell about the “software” you use, but this software is based on a developpement made by Microsoft, i didn’t went too far, but just do your homework like I did and search a bit.
A test on new material. This is a CG Render of some pots Im working on based on an image earlier. Up-scaling saves a lot of time when i have high settings on my render such as caustics and SSS.
Original took 4 mins to render.
Topaz Gigapixel 4X Art&CG
New ‘Unknown’ AI 4X
I am extremely impressed by the new one its fantastic!! i was thinking it might not handle this type of input well but its falwless, especially when viewed at 3840X2160 full resolution… It would have taken a very long time to render that!
Word is going to get out about these new AIs, it’s not a secret. Topaz must have resources and money, it’s up to them to keep themselves on the cutting edge. This new AI is available as is apparently for free unlimited use.
New AIs are bound to come along and make the old ones obsolete. Gigapixel will become obsolete if they don’t act.
Note that you have a thread where you can continue your discussions here, also do not spam other threads with your discussions.
Note that making statements that this is so much better than GigaPixel makes no sense as no one software in a category can do everything. And, from some of the examples posted, both the new “wonder” and GigaPixel fail to produce a realistic pleasing result.
So remember that provide realistic real world results. “Pixel Peepers” will always try to justify things that they see against other software, my suggestion is that you make up your own mind and don’t try to convince other people that one is better than the other.
You can post the name of the product as it makes no difference.
My current thinking is that the quality of the input makes a difference, probably in both… if its a bad camera, then the image has been saved to medium jpg then maybe manually resized smaller and passed through several formats, then maybe the pleasing image can never be recovered.
Meanwhile here is another test.
Random image from the internet.
Original image size = 290 X 317
Topaz low res setting.
new AI swinIR.
You just shown that your using the wrong model, use compressed as it is different and the image you are using is not low res … comparisons need to show on an even footing.
Here the input is basically a thumbnail. Gigapixel is using its face refinement and is set to low res model.
Gigapixel
SwinIR
Next input a medium resolution b&w photo. 945x 650
Gigapixel lines setting(best for this image)
SwinIR
Zoomed
Test 2. Trying to feed the image back into the AI to get a zoomed in area.(not a good idea in gigapixel as far as i know)
Gigapixel
SwinIR
You need to understand what resolution vs compression is. They are not the same
a purposely compressed jpg image saved to level 0 in photoshop.
Gigapixel set to very compressed model.
SwinIR
SwinIR has better hair texture, better nails, better eyes. Overall they both look fake.
Food photography test with depth of field blur.
Medium resolution jpg 550x550 pixels.
Original
Gigapixel standard model 4X
SwinIR
Gigapixel added noise where there should not be, it added color noise as well.
SwinIR is has much sharper defined details and its textures are more realistic.
Test 2
Original
826x620 pixel jpg
Gigapixel 4X Lines(significantly sharper than standard)
SwinIR
Gigapixel introduced noise and color noise again.
SwinIR was more natural looking, almost perfect looking textures, and significantly finer details.
Test 3 feeding the upscaled image back into the software to get a double upscaled image.(not recommended for gigapixel)
Gigapixel Lines4X
SwinIR
I have used many different applications and have found nothing that can do the job as well as Gigapixel especially on images from the early 2000s,I have also found that Video Enhance that can save images from old video with better results than the competition,Then we have the usual upload sites that claim their bicubic resize is AI.Ive seen all the links out there leading to projects that require that I install Klingon dependencies on to my windows 10 machine then require command lines typed in Klingon.People want a nice clean GUI,this is the reason Topaz is the world leader in this field as it has the foresight to give users what they want.
I partly agree. But this new AI is better and Topaz should train their own AI based on this same technique and then build a nice clean GUI around it.
input image 960x540
SwinIR
zoomed
Gigapixel
swinIR
Portrait Face test.
input image
Gigapixel Standard model 4X
SwinIR
zoomed gigapixel
Zoomed SwinIR
SwinIR showing superior sharpness detail and less noise.
Architectural test.
Input image
Gigapixel Lines model 4X
SwinIR
zoomed gigapixel
SwinIR zoomed
small section of the image re-upscaled at 4X.
Gigapixel
SwinIR
Gigapixel verycompressed 4x
SwinIR
I shall leave it up to you to decide which is better.
@sono2000 what is your actual point with all these repeated posts? Are you trying to get users of GigaPixel to switch?
The other thing is you need to specify the science behind your testing as low res images you are presenting prove absolutely nothing. By that I mean the process you undertake to compare, what output is available because most people deal in DNG/TIFF, what sizing is optimal etc., etc.
If your just going to post more and more images, not even using the latest version of GigaPixel I will close the thread to further posts.
but still all the projects out there are not an all in one package some and only enhance in certain areas and ways so we end up with one is good at that and the other is good at another thing.with 6.1 the face recovery is better than anything i have seen,as well as that Gigapixel is also CPU or GPU and CPU suits me fine for those early 2000s images from the web.Now if I have to capture still/frames from fast moving video of the same era I use topaz video enhance,as its better in those circumstances.Ive tested those projects myself (that I’m not going to name) I wont stray away from topaz , when there is an update you never know how the next set of models is going to interpret an image and what details will be uncovered .I usually compare the output on the same images then one can experience how the models used are evolving from earlier versions.
i’m actually testing Photo Ai in early access, and i’m getting better result than the open source application.
so gigapixel + denoise + sharpen, used correctly, can make much better result than what i could think of or was able to do manually !
have you tested hair in the early access version? I’ve seen massive improvements in other details over the last couple years, but topaz has always had a hard time with hair and text (small compressed text still legible can become mangled garbage).
yes, i converted a bunch of old pictures (actress singer dancer ginger rogers), i worked on them in gigapixel in the past, and took the gigapixel ones (i don’t have the original anymore) and put them in Photo Ai, and Wow, it added extra quality to it. i can post the pictures here if you want.
i’m aware about text, i didn’t tested. but Photo Ai is Gigapixel 6 + denoise + sharpen, which calculate the best setting automatically and use the 3 software at the same time.
Gigapixel alone (not version 6, I don’t remember but the date of the file is july 2021) folllowed by the same picture processed by Photo Ai, in Auto mode with face setting lowered.
Thanks for the mention of this new product, I didn’t even know about it, I just claimed mine now and going to try it out 
