BEST Way to ENLARGE Images - PS Super Resolution vs Topaz Gigapixel

Photoshop Super Resolution vs Topaz Gigapixel AI - If you’ve ever tried to resize an image to make it larger, you probably know that this can have a negative impact on image quality.

In this video, I do a thorough test of this new feature to see how enlarging various image files impacts the overall quality. The results were quite surprising!

In the first test, I compared the same image, one version at its original 100% resolution and one at 50% resolution. I then enlarged the 50% resolution image and compared the results with the original.

As a second test, I then compared the same image, taken on both a 35mm format Sony A7R2 and medium format Hasselblad H6-100c, enlarging the smaller Sony image to compare with the Hasselblad image.

Finally, I also tested Photoshop’s Super Resolution feature against an alternative software, in this case, Topaz Gigapixel AI, to see how the results stacked up.

Chapters:
00:00 Introduction
00:33 What is Photoshop Super Resolution?
00:51 Test 1 (Super Resolution on Larger TIFF File)
06:33 Test 2 (Super Resolution on Smaller JPEG File)
07:40 Test 3 (Super Resolution with Sony A7R2 file vs Hasselblad H6)
10:44 Test 4 (Photoshop Super Resolution v Traditional Enlarge)
16:04 Test 5 (Photoshop Super Resolution vs Topaz Gigapixel AI)

[jozefgeusens]

In my experience working with Topaz AI for a few hours, I must say it's very impressive. Especially with jpeg-images it will also remove the annoying artefacts of the jpeg-compression. The effect on RAW images is excellent, but I'm speaking of 24 megapixel images upscaled to 2x and 4x. The 6x enlargement is marginally better than 4x and looks artificial, but from detailed RAW to 4x is mostly impressive especially on textures like leather and textile (and even skin). Upscaling to 2x is for a lot of images a real treat. Also, using images up to 24 megapixel as source, will not take very long to process: in my configuration (i5 Intel Core i5-8400 and NVIDIA GeForce GTX 1660 Ti) it takes less than 3 seconds to preview and 2'50 for saving to tiff - resulting in a 12068 x 8056 pixel image. It's not perfect and I can sometimes see artefacts on small gradients, but overall I'd say it does a better job than what I've tried before.

[jlwilliams]

This is an interesting comparison, but I think a lot of people -- NOT to say Mr. Taylor specifically -- are expecting too much from this technology. The question shouldn't be "Will it look as good at the individual-pixel level as a native high-resolution image?" because of course it won't... the real question is, "Will it look good enough to get by? Will it let me mix in a favorite image from my old camera with images from my new camera, without people immediately saying, 'What went wrong with THAT one?'" In many cases I think the answer is yes, especially if you're comparing prints at matching sizes, because printing helps mask the fine-detail differences. (I hope more reviewers eventually will do comparisons at fixed output sizes, since that's where this type of product is most needed -- when you're producing a set of same-size prints or a print-on-demand photo book, and need to match all your images to the printing device's optimum resolution.)

Also worth noting: Adobe said explicitly in its Super Resolution announcement that SR works best on original camera-raw images, because they provide more information for the AI to work on. As Mr.Taylor's first demonstration showed, a high-quality TIFF image seems to work just as well. But it's probably expecting too much to expect a similar performance on a JPEG image, especially one made with an older, likely noisier camera. Also, on some images, Topaz Gigapixel AI might produce "sharper-looking" results, because it includes a user-adjustable sharpening function; Adobe SR specifically does not do this, instead expecting you will want to adjust sharpening and noise reduction in Camera Raw after upsampling (again, this was in the announcement post.) And as Mr. Taylor also showed, sometimes a plain Photoshop upsample with the Preserve Details 2.0 option is all you need; I think it works especially well on "organic" images that don't have a lot of hard edges or prominent textures.

AI upsampling has exciting potential, especially for those of us with collections of older camera-raw images, and I appreciate Mr. Taylor's comprehensive demonstration of what it can and can't do.

[Dstonephoto]

One often overlooked point of superresolution is the improved ease with which you can mask and edit out parts of an image which would normally be extremely tricky (eg a building behind hair). When used sparingly and creatively Gigapixel offers some truly spectacular results. Now, where a lot of this stuff falls short is the one-size-fits-all aspect of the software. A lot of the AI branded products are clever software solutions which only harness a fraction of the potential of true machine learning. In a perfect world we would feed the neural engine a set of perfect cat pics and use that as a base from which to optimise our poorly lit image of Whiskers the one eyed cat and transform him into a runway-ready majestic feline of epic proportions! On the video front some of the stuff being done is batshit insane out of this world when compared to the stuff we get for stills! Still, super resolution and other computational photography advances are an insane gift to the world of photo enhancement. Try grabbing a 50 KB google image from David Bailey and upresing it, then zoom in on it. It’s pretty crazy stuff.

[balamina]

I can't thank you enough Karl for all the stuff I've learnt from you! Very helpful tutorial and comparison test that I really was looking for! God bless you!

[colinweir]

Thank you for your compaison information. i guess it all comes down what you are looking for.

[mrbob]

Thank you Karl for taking the time to do this

[relativwork]

thx, also for the clean english (for me as a german guy ;-) )

[DrJorgeReynal]

Every day you learn something new with the Master Karl !!!!

[Allen-Trav]

Thank you for a precise, thorough comparison. It is spot on with my experience using Gigapixel AI. I am reasonably pleased with upscaling cropped images coming out of my Canon EOS R6. Upscaling has assuaged my temptation to upgrade to a larger MP camera like the R5 for now.

[RingWraith]

Why would you have Gigapixel AI Mode set to "Architectural" instead of "Standard" for this image?

[davidmilisock]

Great video Karl thanks. Being in the sign industry we've been up sampling images for nearly 3 decades. Depending on the quality standards of my clients the commercial use of such procedures has in the last decade required that I have multiple programs and methods for achieving this while in the past we were limited to our own ingenuity.

Multiple applications for noise removal are also required as much of the work comes from 3D Architectural software. You'll get a 6, 000 pixel sRGB image that must be cleaned up for noise then resolution enhanced, the last five images I did went to 35 feet at 100 DPI for large banners.

It's not uncommon to run the original through multiple noise reduction applications to choose the best cleaned result and then through multiple up sampling applications to get the best high-resolution recornered.

In my experiance no one programs has the market on quality cornered.

[skyscraperfan]

Thanks! I was already looking for a video where somebody tries to restore a 50% resolution to 100%. That is the only way to see if "Super Resolution" really doubles the resolution.

[danbrowning]

Thanks for this critique. For me, PS is good enough. I can’t imagine waiting 20 minutes to process the image with Topaz and don’t see the need for another software program. I also would rarely need to upscale for such a large print.

[lupo]

Great video. It would have helped if you noted the Topaz version number, since they update their software every couple months.

[dct]

With the image detail already so high from enhance and Topaz. Doing a sharpen or sharpen ai, I wonder if it'd be over kill. Would it fix some of the blur, but leave out artifacts. I'd assume it'd introduce even more artifacts.

[mavfan]

I found Gigapixel superior for my taste to Photoshop working with Sony & Fuji GFX 100 files. I was really unimpressed with Gigapixel a year ago but it’s been improved a great deal.

[homewurks]

Why does your 'File Handling' menu support JPEG, HEIC and TIFF Handling, Karl? I cannot find the HEIC support anywhere

[ayushclicks]

Does this feature available for old photoshop version or camera raw filters??

[SenseiKreese]

My impression is that if you added some sharpening to the super-resolution afterwards, you'd get the same result as Gigapixel. Do the print test

[CloudStrife]

Would be cool if you did a video on how to shoot on Musou black and ir flock sheet.