True, I use rendering for the entire image. I don’t use batch processing or direct export, so it’s probably as you write.
Have you checked if the results of whole image preview and direct export are identical? Having read posts about the preview not matching the export, I have little trust in this. But maybe the preview is better, I’ll have to try. 
I accepted your challenge 
and here is my result: I tried to compare the result between Preview entire image and direct Export image. A randomly selected photo, 800x600 pixels. Source and result in *.jpg format. I used my favorite Redefine (C=2, T=3, Upscale=4x). See Fig. 1 for task parameters. (Windows 11 Home, Nvidia, Gigapixel 8.3.4.)
I always started Gigapixel again from the beginning, including specifying the source. After saving both results, I performed a visual comparison – it seemed the same to me. Then I compared both results bit by bit – the program found no difference, the content was identical (see Fig. 2) as well as the size of both outputs.
I attach the resulting image (Fig. 3). It is, of course, simplified, I did not try all possible functions of Gigapixel repeatedly 100x, but it is enough for me as the first approximation: There is no difference. I don’t know whether there are (and if so, what kind of) differences in the results of other colleagues in their photographs; maybe there are (maybe not), but they would have to speak for themselves:
I have the same suspicion, but I didn’t time it (I don’t have a stopwatch). So the suspected perpetrator has not been convicted yet.
Using the stopwatch on my iPhone, I upscaled a 633x949 JPG to 1467x2200, C2,T1 with image description generated by Gemini. It took approximately 10m25s on my M4 Mac. The preview had generated in 10m18s. So nothing in it really. The upscaled images are exactly the same.
1 Like
Interesting findings, thanks for sharing. I am pretty sure that one of the points that were criticized as to GPAI 8.3.x was that the preview does not match the final result. But that may be related to an earlier version, or it may occur if the preview window is used instead of the full image preview.
I have the impression that this originally concerned the preview window (only the selected part) of the image. If only a part of the image is selected, less information about the whole is available, so there may be a difference there. If I remember correctly, in Gigapixel there was an optimization in some version that significantly shortened the rendering time of the entire image, so the problem may no longer be critical – of course, if you are not processing large images or a large number of them. Fortunately, I do not have to process hundreds or thousands of photos at once, so I always use the entire image (I don’t have top-of-the-line hardware).
The topic of keeping ‘showoff’ posts in another section needs to be brought up again.
Since the onslaught of ‘creative’ posts started, this forum section has been rendered useless for bug reporting/tracking/feature requests as everything gets drowned out with ‘topaz girl’ posts.
I urge anyone who just feels like posting ‘hey look what the funny AI did’ pictures to post them somewhere else, unless it’s to help report/fix a bug.
It’s become incredibly…something…IMO
9 Likes
Absolutely agree.
1 Like
i agree with you as well
1 Like
As do I.
1 Like
Preferably in its own Redefine “Fantasy pictures” heading in the Show Off section.
Regards.
3 Likes
Will there be an update today?
Hello.
I do hope my suggestion about moving the “Fantasy pictures” to its own section in the Show of Spot, hasn’t offended anybody. It wasn’t my intention to do so. I really do enjoy looking at your work but as a new user to Gigapixel it’s a bit overwhelming at times.
Best regards.
1 Like
Photo AI 4.0.0 was released yesterday and it seems to have fixed the Super Focus 2 bug that I reported on 3/28. I also had reported the same bug for Gigapixel AI when using Recover 2. Topaz support told me several weeks ago that these were the same problem so I expect that there will be a Gigapixel AI release with this bug fixed soon. I also hope my bug report about Redefine will be fixed in the same release.
1 Like
Does the Redefine model get updated every now and then?
Not sure if it’s due to the size of images i’m upscaling or due to the model itself. But results now are a lot worse than they were some 6+ months ago. As in, it completely changes the initial image. Even with creativity at 4 and texture at 1 which wasn’t the case previously.
Hi.
The difference Face Recovery is only available for Creativity levels 1 and 2 now in addition, the recommended image size is, One Megapixel 1024 x 1024px or smaller
Also, Creativity set at 4 is considered quite a high setting for photographs giving images an almost surreal look.
Try keeping the Creativity for photographs between 1 to 2 and definitely no higher than 3 whereas, levels 4 to 6 can only be best described, as being more for the artist or surrealist hence, the name Redefine and Creativity.
Are you aware, each one of the six Redefine Creativity Levels, are in fact, separate AI Models and that’s, why the differences between each one can appear more diverse and very different in appearance
1 Like
The latest release still does not let me use Recover V2 on my M1 Mac. It simply either crashes or says " an error encountered while processing".
In my perception, dealing mostly with animal pictures, C1 and C2 are quite similar, especially at T1. C3 is a completely different cup of tea, it creates way different textures than C1/2 and, dependent on the prompt, sometimes false items in the picture.
@Devs: I’d love to have a setting close to C2 that doesn’t “reinvent” animal fur as aggressively as C2 does. C2 turns every creature, even one with relatively short fur, into something that looks like it lives in an arctic environment. Not that a tiger doesn’t look nice with mammoth fur but it is definitely too far from the real thing. 
By the way, if you walk the upscale factor up from 1 to 2 to 4 to 6, the “too long fur” issue will become less prominent with the higher digits. So, if you experience the issue, using 6x and downscaling to the desired resolution can yield much better results than using a lower upscale factor. As long as your GPU allows for using the higher upscale factors and you don’t mind the long processing times involved.
I certainly agree with you and with wildlife, care has to be taken with the Creativity and Texture settings on fur and especially birds feathers.
An Idea was put forward to Topaz about refining the Creativity between levels 1,2,3 and 4,5,6 I myself suggested a percentage scale would have been a better option and that’s when Topaz explained although, the six increments appear to be part of the same AI Enhancement Model. In fact they are six individual AI models working independently from one another therefore, a percentage scale wouldn’t be feasible
So, going by that information I don’t, see why they couldn’t introduce an AI model level between 3 & 4 which, has the biggest jump from what is real and what’s surreal.
A couple of things I’ve noted, firstly Redefine seems to understand people because, yes at the higher Creativity levels Redefine turns everyone into Topaz Girl but, hasn’t she got fantastic hair which can easily be composited within your image in Photoshop.
Secondly, as I’ve already mentioned you certainly don’t get fantastic animal fur or birds feathers. The Text Prompt helps a little but, what would really help enormously would be the ability to include and use a reference image along with the Text Prompt.
Andy
