Hi, I’d like to ask whether the program now includes preservation of seamless patterns, since most upscale programs tend to destroy them. As a result, when applying textures after upscale in a game, it ends up looking like Minecraft… you can clearly see the texture borders. Thanks for the answer, René
1 Like
Could you please publish the roadmap?
TopazLabs efforts seem to got diverged to the cloud based subscription products mostly…
TPAI and GAI haven’t gotten updates in a long time now… And what about Redefine on Mac?
2 Likes
BTW, do you use legacy models?
I recently found out that for upscaling some of the legacy models are better under certain circumstances than the newer versions.
For example, HiFi v1 often handles animal fur better than v2, being less aggressive in sharpening the hairs, while v2 often ends up in subtle over-sharpening even at sharpness set to zero.
Anyone else still cherishing the legacy models?
4 Likes
I’m really curious to see what the results from ON1 Resize AI 2026 will be like. It’s coming out in a couple of days.
I noticed a bug in this version of Gigapixel. Mainly with the Recover V2 model. If you add, say, 4 or more images and set a Recover parameter for each image. For example:
- The 1st image, I set it to “none.”
- The 2nd image, I set it to “low.”
- The 3rd image, I set it to “medium.”
- The 4th image, I set it to “high.”
And then when I export all 4 images at the same time, it only takes into account the last setting (high), and this will be overwritten on the other 3 images. So, I’ll get 4 images with the “high” quality setting.
It’s quite restrictive.
Hi, this doesn’t download, please refund the $64 I just paid.
This is the message I get, I have no clue what to do with it.
RequestHeaderSectionTooLarge
Your request header section exceeds the maximum allowed size.
8192
9FEB7TR9YCGR6T5G
XlfcA9s9L/Z05ohEpm6qFGnOohHYHhi9/BruSw60KDbF8tis1IeaJdWsxEY7aPEup4QoMA+w5zY=
Hi William.
RequestHeaderSectionTooLarge
Typically points to a problem with how your system or network is communicating with Topaz’s servers.
Disable VPNs or proxies, which can sometimes overload request headers.
Check for software conflicts, like antivirus or firewall settings that might be interfering with Gigapixel’s connection.
Once that’s done I would recommend redownloading Gigapixel to ensure all model files and configurations are correctly set up, sometimes corrupted installs can cause strange behavior.
Here’s the link to contact Topaz Support directly
Topazlabs Support
support@topazlabs.com
Hope this helps
1 Like
Your link and mine both take you to Topaz Support
2 Likes
Hi everyone. Could someone with a big graphics card improve this photo with Supir AI to a resolution of 8160x6120? = 50 megapixels?
https://i.postimg.cc/9fqd48T7/IMG-385.jpg
With several tests in Gigapixel, I cannot get the most faithful result possible without modifying the original photo.
Here is what I have obtained so far by first combining Flux context and gigapixel + Photo AI with super focus afterwards.
https://i.postimg.cc/MpJMP6jv/IMG-385-v1.jpg
C’est un peu mieux qu’avant mais pas vraiment génial à mon goût.
The original photo was taken with my Galaxy S25 Ultra with its 5x telephoto lens + digital zoom. So it’s very blurry if you zoom in. Like paint lol.
Also, if it helps, here is the prompt I created with Gemini:
"This photo captures a serval in mid-stride, moving from left to right. Its slender body is in profile, highlighting its magnificent tawny coat spotted with black and its long legs.
Its head is turned forward, its gaze focused on its destination. A slight motion blur on its paws and head enhances the dynamics of its movement, giving the impression of a snapshot captured in the moment. It is walking on natural ground composed of green grass, dirt, and dry twigs. The photo conveys a sense of grace and stealth, typical of this feline."
The competition’s new “make images bigger” release, and GPAI 8.4.2, both at default settings, 4X upscale… GPAI is at R in all comparisons.
The original Facebook ad image:
GPAI at work, different view:
6X shootout:
1 Like
But here, you’ve used Redefine Creative to its fullest extent. I knew how to do that too. But Creative mode doesn’t work on this photo. It no longer looks like a serval. Plus, all the unwanted artifacts have been added. I wanted an improvement from Supir AI. But my RTX 3070 isn’t powerful enough to use the model. Because you need at least 12 to 16 GB of VRAM.
I have tried Supir some time ago and found it to be picky as to input picture quality. As far as I know, it uses a Stable Diffusion model, and none of those is really good at adding details to blurred pictures. That is why I doubt Supir can significantly improve the original pic.
However, I can try what I can get with GPAI and Stable Diffusion alone.
I think GPAI using the recovery model will give better result than the Low Res model. 
Here is a quick comparison between GPAI and free open-source model. 
I think this is not too bad. It took 4 steps with different software:
-
Gigapixel redefine realistic 4x, setting “subtle”, prompt “serval cat” on the 1065x800 picture from this board (tried 2x as well but it is worse), yielding a 4256x3200 pic,
-
Downsized to 2128x1600 using a simple Lanczos filter (just because the next step involves Stable Diffusion for which the 12k pic is too large),
-
Used Stable Diffusion image2image with the model realisticVisionV60B1_v51HyperVAE, denoise 0.3, to improve the fur texture,
-
Then re-upscaled 2x with GPAI High Fidelity V1 (better for this than V2) to 4256x3200. 50 MP would sure have been just as good but the file size would probably have been too large to upload.
The “secret” is to use Stable Diffusion as a “refiner” for GPAI output. There are several SD models that can salvage the over-agressive fur texture that GPAI redefine tends to generate in animals.
2 Likes
BTW, the original result is even sharper, the board software reduced the upload in size and compressed it. I could try to upload the full size result as a .zip if anyone is interested.
Sorry, I misunderstood the request…
It’s weird that you have this. Or the image is limited in time on the site. 
Yeah, it’s not too bad.
