GigaPixel vs another?

Hi there im a long time user of Topaz Gigapixel, im currently using version 5.6 because i’ve not as yet renewed the licence, but I am here to challenge Topaz users if they can upscale this image better than an amazing totally free upscaler AI model i found elsewhere on the internet. I found the image on yandex and i’ve checked the search similar and there is no larger version of the image available and i am not cheating.

It completely out boxes Topaz Gigapixel especially when working on very small images… It does have some limitations currently like input image size so i had to stitch two halves together.
The results are much more sharp detail, no noise, no color noise.





The original is the lower res image “original”… The new online upscaler is “new” the others are named after the topaz AI model name. The images were large about 4800x2500, they are downscaled by this forums limit… But i think you can still tell!


Can I ask you, why would you bother posting here if you think this is better? What you post here you should be posting on their forum, scale it up to 6x with GigaPixel very compressed and see what the difference is as that model captures the colors better than the new version you have posted.

Yes the reason im posting here is 1. I am on an older version of Gigapixel and im wondering if the newest version is superior or competes with this ‘other AI’. 2. I am hoping to get some discussion or information on what the future of Gigapixel is and if it too will introduce new AI models that push the upscale quality further. Reason being is I dont know whether to renew my subscription or not.

1 Like



From what i can see the ‘new AI’ (4X) upscale completely out boxes Topaz (6X very compressed) and its other models.

If you don’t believe me im willing to do a test image you post or suggest.

@AiDon Actually, shouldn’t Topaz feel concerned getting outperformed by a free algorithm ?
Gigapixel almost always adds thick color outlines or an incredible amount of noise when there isn’t initially. (And the noise issue is the same as with Sharpen).
Just look at the cropped close-up, in Gigapixel it is full of color artifacts and looks unnatural, while the other appears to be artifact-less, especially when we look at the stainless tools on the left, the lid knobs, the plant, and the black casserole. The whole image cannot look more natural. I don’t know what software was used but it clearly outperforms Topaz, no doubt.

@sono2000 This got me curious. What is the website ? This almost makes me regret my purchase :expressionless:
Just, as with the 6.1.0 version of Gigapixel, the results are exactly the same as your images, no improvements. I have just tried it with the original image you provided. 4x Lines and Very Compressed lead to the same results.

2 Likes

Yes the lack of color noise and artifacts is what stood out to me, and the consistency of the small scale detail.

I shall give you the abstract technical info it says about the model.

Image restoration is a long-standing low-level vision problem that aims to restore high-quality images from low-quality images (e.g., downscaled, noisy and compressed images). While state-of-the-art image restoration methods are based on convolutional neural networks, few attempts have been made with Transformers which show impressive performance on high-level vision tasks. In this paper, we propose a strong baseline model SwinIR for image restoration based on the Swin Transformer. SwinIR consists of three parts: shallow feature extraction, deep feature extraction and high-quality image reconstruction. In particular, the deep feature extraction module is composed of several residual Swin Transformer blocks (RSTB), each of which has several Swin Transformer layers together with a residual connection. We conduct experiments on three representative tasks: image super-resolution (including classical, lightweight and real-world image super-resolution), image denoising (including grayscale and color image denoising) and JPEG compression artifact reduction. Experimental results demonstrate that SwinIR outperforms state-of-the-art methods on different tasks by up to 0.14~0.45dB, while the total number of parameters can be reduced by up to 67%.

1 Like

Thank you for sharing. :smiling_face_with_three_hearts:
The image you posted looks great.
Would you mind share what software or app you used ?

Thank you very much :grinning:
Have a nice day

Im not sure im allowed, I might get banned from discussion if I did that… I would rather Topaz stay at the cutting edge and put some of their revenue into evolving Topaz Gigapixel, the fact that free cloud processed AIs with unlimited use(not credits based or limited) are out boxing it is concerning, and ive only seen tiny improvement in gigapixel over the years… It’s frustrating.

1 Like

I have sent you a private message.
If you don’t want to post the link on public, you can reply me on that message.
Thank you :smiling_face_with_three_hearts:

1 Like

Thanks for the info, this looks promising :wink:

1 Like

Just had to mention this new AI also seems to be able to take the upscaled image and add yet more detail again! which i’ve never seen Gigapixel do.


46hjf





Try that with gigapixel!

1 Like

This is like Face Recovery but for the whole image.

We can almost see the atoms lol

1 Like

Comparison :

Full size images


Original Image


Gigapixel Standard (Auto settings) = lot of noise as usual. Okay the tool-tip say doesn’t work good on fur, but still, what the heck. Even the vegetation gets noisier… This makes Standard unusable with most material when Auto is enabled. It puts the Remove Blur at 75 when there is no need, just put it at zero and most of the noise is gone. Design flaw or feature ?


Gigapixel Very Compressed (Auto settings) = Most of the details are smoothed out and some parts turn into paint (look at the head’s fur).


Gigapixel Low Resolution (Auto settings) = This gives the best results among all models, even when Auto is enabled.


Unknown AI (setting A) = The result is impressive and looks very natural, all the details of the fur are preserved and enhanced, although is presents a color shift (must be some color profile or bit depth issue).


Unknown AI (setting B) = This one is a bit smoother than with the setting A, there are slightly less details and it looks more like a real photograph with natural blur, although there are some blurry artifacts in the vegetation around the subject.


Close-up


Original image


Gigapixel Standard


Gigapixel Very Compressed


Gigapixel Low Resolution


Unknown AI (setting A)


Unknown AI (setting B)


Unknown AI (setting A) more close-up = Note the flies are getting upscaled too, and they even get shadows…


Unknown AI (setting B) more close-up


Quick comparison from Original to Unknown AI (setting A)

:face_with_monocle:
No further testing required.

1 Like

you didn’t noticed it but gigapixel is doing a much much better work on the grass and all the background. the “unknown AI” is removing a lot of details there. it’s working great while zooming, but by full comparaison, the fur is different, not only by colour but in the texture too. where i agree, is when zooming ! on the cow, on the last picture, the result is very impressive, but on overall with the background and grass, i prefer gigapixel. i’m really curious to know which new software it is.

I agree the grass loses details, I believe it is because of the color shift, but also the sharpening is too heavy.

1 Like

best of both worlds !! mixing both pictures (i’m not an expert, i’m not able to do that, doing mask on one picture and replace it by the second one etc…),
(check your Private message please :wink: )

One more please.
Original low res
bee-flower-insect-blossom-thumb
Gigapixel standard 6X

Unknown AI - upscaled twice in a row.


Notice the detail on the fur and the wing and such, and as for the flower it has great detail but has probably guessed the actual texture wrongly, however its a good try. The AI probably guesses well known features better such as the bee.

Unknown AI, magnifying just the bees head once again.(this can be done potentially ad infinitum).


Next example input image was just a tumbnail size.
30544302408_d28ae15f09_z-120x120

Gigapixel 8X

Uknown AI 8X (approx)

I just tried on one of my picture, and it’s really depend of the material, i’m on an example where gigapixal did a much better job than this unknown AI. so it’s depend of a lot of factor. I don’t think we can tell that “this is better than gigapixel”, and vice versa. it demands of the material, what the purpose of the upscalling etc.

But which has more potential?? Gigapixel has had years to refine their models. And its subscription based.

The unknown AI is showing amazing potential and if its not perfect in all scenarios then it should be developed further. Topaz should adopt this model as one of its models so we have the choice.

what you want and what topaz can do / will do, are two differents things. you can’t come here, unfortunatly, talk about an another Ai software, and drop “this is better, topaz must use it” :wink: . you’re not the dev of Gigapixel, i’m not them. It’s a commercial product and the engine of picture treatment you talk about is based on a microsoft project. there is a licence with it. so it can imply a lot of things that only topaz is able to answer.

For example, on Video Enhance Ai, actually the software is actually completly rewritten to use Ffmpeg, there is a licence to follow, and they can’t do everything we would like because of some licence limitiations etc.

all this just to tell that, it can be a good idea to post the question, and submit it, but in the end it’s topaz decision. Gigapixel work under a custom engine made by them, using their own model. this project use other software, technic, etc… they can’t import “this model” like this. i don’t think it’s working like this. a model is nothing without the software or the engine which make it running. so it’s great to have suggested it, let’s see now the answer of topaz devs on this !