Folks,

So somewhere on this forum someone posted either a thread or a post to a thread talking about using Gigapixel AI (a program which resizes digital images applying "AI" to bring more detail to them, particularly humans) and Headshop so I thought I'd pick up the trial (which seems completely full-featured for 30 days) and check it out.

My impression overall is that it isn't worth using for Headshot.  The samples this person posted (sorry, too old to remember who) showed some close-up details but they were VERY close up (like just an eye).  At that range yes, you do indeed pick up some detail.  But for general head looks the resolution HS produces even from a fairly low-res image (say 1K or so) can't be improved.  

Note there are lots of different ways of using this, but the most reasonable is to increase the resolution of the base image you are using for HS (you can also, post HS, increase the textures produced but this is pretty silly and in my tests didn't help even closeups).  So if you only have a 1K image you increase it to 4K.  I tried all combinations, even very low res (300x200) and very high-res (4K and above) and no combination used with HS had any appreciable improvement.

The problem is as you step back from the improved detail the eye can't discern it (this, by the way, is why we can't perceive 4K but I don't want to open up that can of worms).  Increasing the detail for the base texture just isn't something that pays off in any manner whatsoever (and, of course, the file sizes GP produces start to approach obscene).

For comparison I used Photoshop with it's (newer) "Preserve Details" enlargement and even there it doesn't make any real difference.  I had been hoping this would somehow make using very low res images possible in Headshot and while you can, with or without enlargement by any means does the result look any more details or real.

I have LOTS of samples -- I gave it a pretty good workout -- but rather than post the good news is anyone can try for themselves and make up their own mind (there doesn't appear to be any special pricing for GP -- it's $99 and no discounts for Black Friday/Cyber Monday seem available -- for $50 or so I might pick it up for other purposes but for now I'll save the cash).

And if anyone DOES think there is major improvement (again, at distances where, say, the entire head fills the screen -- so a Real Test) I'd love to see samples.  My own demonstrate otherwise.

---

Alienware Aurora R12, Win 10, i9-119000KF, 3.5GHz CPU, 128GB RAM, RTX 3090 (24GB), Samsung 960 Pro 4TB M-2 SSD, TB+ Disk space
Mike "ex-genius" Kelley

Your mileage may vary, of course. But I do use it every day to resize 960x540 frames out of keyshot to 1080p to save our industrial designers 75% of their rendering time.

It’s the highest ROI of any graphics app I’ve bought in 25 years.

But please try it out! As mike said the demo is free.

I can see that it might be very useful in other applications.  I'm just saying for Headshot, not so much.

And if anyone wants to prove me wrong, just post images showing the difference (but, again, at *normal" sizes -- so a headshot comparison would be great, as that's as close as anyone needs to see a face).

---

Alienware Aurora R12, Win 10, i9-119000KF, 3.5GHz CPU, 128GB RAM, RTX 3090 (24GB), Samsung 960 Pro 4TB M-2 SSD, TB+ Disk space
Mike "ex-genius" Kelley

Hmmmm...
I'm gonna say I am with you on this one here, KT.  While AIG can enhance the look of an up-rezzed lower res photo, I think for the crisp realism of human texture here it doesn't meet it for me.
I mean, you can tell yourself it looks better, and for protocol's sake, you can create, say, a 4K texture.  But it is not the real-deal 4K capture. 

Fair enough. For me, it's enough that it lets me resize my renders. Anything else is a bonus.

Why would you use Gigapixel AI for faces? That is arguably the weakest point of that program in situations where the source is low quality. Topaz has several other programs that often produce better results with faces. I wonder how much time people spend studying Topaz's series of AI tools. Gigapixel AI is not a one-size-fits-all kind of program. The results these tools produce obviously depend on the type of problem the source suffers from in terms of quality (motion blur, compression artefacts, noise, traces of video interlacing, etcetera). The fact that the software has "AI" in its name doesn't mean that the user can just blindly trust the software to make the right choices, and can stay ignorant him-/herself about the enhancement process while the AI software takes over your PC.
I don't see how this Gigapixel AI discussion would have any relevance to users of Headshot.

Simple, some other user (I think the one on this thread) posted where he thought it improved things and I was just testing this and finding out, at least for me, that it doesn't.

It's a reasonable test -- using very high quality images for Headshot is very important and GAI is supposed to be able to sharpen such images.  That it doesn't work well for HS is good for a whole lot of folks to know.

---

Alienware Aurora R12, Win 10, i9-119000KF, 3.5GHz CPU, 128GB RAM, RTX 3090 (24GB), Samsung 960 Pro 4TB M-2 SSD, TB+ Disk space
Mike "ex-genius" Kelley

Sorry, I think I missed that other thread. Still though, Gigapixel AI isn't the most logical first choice of all the available Topaz tools in situations were the picture quality is not just low resolution, but also bad due to distortion, artefacts, compression, etcetera. If you look at the other tools they sell, you'll see that there are much better AI tools to clean up a picture that work better with faces. After enhancing a distorted picture using these more appropriate AI tools, maybe a Gigapixel AI upscale would help, but not the other way around.

It's possible, but I doubt it.

I don't think you are ever going to get anything better than the original in terms of Headshot use.  You can clean up the original, of course, and Photoshop is more than adequate for that.  But there aren't any "magic" solutions to making lower res images "better" for Headshot (part of it, as I explain above, is just related to the limitations of HS -- refining details isn't the big issue with lower res images.  Not having them in the first place is).

I've posted a tutorial on using PS for helping but RL has posted a much better (and longer) one.  Using the tools in PS (or whatever your favorite editor is) can help a lot with HS.  Just no "one-button press" fix.

---

Alienware Aurora R12, Win 10, i9-119000KF, 3.5GHz CPU, 128GB RAM, RTX 3090 (24GB), Samsung 960 Pro 4TB M-2 SSD, TB+ Disk space
Mike "ex-genius" Kelley

I have seen a few videos on AIGigapixel and faces and humans is it's weakest point as it cannot interpret them just sharpen out and upscale what it is given be it shadow from a nose whatever so you can get some scary faces.
Landscapes and architecture more it's strong point with clear lines and boundaries, it would be great for upsizing patterns and textures too.