Product Direction in 2024

Thanks Andy, I ordered a one year extension for Gigapixel. Hope it will get some more updates in the future. :blush:

Generative AI to fix a problem is not the same as generative AI to create or add something to an image. Are you sure your interpretation of the rules is correct?

You are welcome. I’ve also added some info to your thread under ‘Bugs and Issues’.

I’d like to see Chromatic Aberration and Lens barreling correction relative to the lens used.

Yes, I’m confused too, are they different packages that actually work differently, better or even worse than other packages? or is Photo AI the whole package one should get if only wanting photo editing and resizing? a good streamlining of the software would be great.

I tried Supir and I’m not impressed for several reasons.

First and foremost: It requires a minimum of 60GB RAM and 30GB of graphics VRAM.

Second, and also important: It hallucinates and adds details that it seems fit and that would be missing in the image, but which do not actually exist. For example, in some images, it adds earrings to women, in others: teeth, i.e. a smile, even though it is clearly visible that the person is only smiling with their lips. So it hallucinates details and needs prompts to improve prediction. On the GitHub page itself, you can see that it adds a car logo that is different from the original photo… etc. I think it can be used for “playing” but it is hard to use for professional purposes. By the way, since it is too demanding, it is impossible to upscale more than 3x 640*480 photos with an 80GB GPU which is pointless.

Replicate demo is available at: cjwbw/supir – Run with an API on Replicate

I remain loyal to Topaz :slight_smile:

Thats why we want this implementation how the CEO did write.

woho, the website you linked shows a whole different level of this ai implementation and is by far more realistic than the ad website.

Now you can see which failures it does and how good Topazlabs is already in this regard.

To me, it’s not that fascinating when it makes up details that don’t exist. It’s generally okay and I understand the hype, but I repeat, it’s not for professional use. At least from my perspective. The most banal example. With the best model and the same settings, it gave me different results, and quite a bit. The third image is the original upscaled bicubic.

Eric, thanks for asking for input. In scrolling through the responses, I see many that express the same things I wanted to bring out.

Most of my work this year has been with extended families archiving. I’ve restored and processed hundreds, perhaps a 1000 or more old photos - some going back to the late 1800s. Many were only available as files scanned in by others, so the variation in quality was enormous, and not under my control. For newer digital images, PAI was all I needed to process most of them. The others required lots of time and effort. I learned from that what would be valuable improvements to PAI for this purpose.

  1. Face enhancement worked remarkable well on some extremely old grainy photos. Where it struggled, as others have mentioned, is in the surrounding hair, so the nice faces floated in a sea of immediately surrounding blur, noise and lack of detail. What might be helpful, is to be able to manually allow brushing in the “face” area to include the portion of hair that automated processing ignored. That would go a long way in fixing how obvious the difference is.

  2. A set of filtering for scanned image artifacts would be nice. I dealt with a lot of moire patterns, matte finished photo scans, and some line artifacts on a huge number of poorly setup scanned pics. No idea how or why these had lines all through them. Interestingly, Gigapixel AI did a far superior job of of removing some of these artifacts.

I also photographed several indoor talent show events this year. Even with my decent Sony cam with great autofocus, there were still a lot of noisy blurry photos when shot with burst mode to capture action. I must say, PAI, did a great job fixing these - even in full auto mode. I did experiment with the color and lighting beta filters. Some photos showed very good improvement, but others actually turned out in worse shape with the filters on. I wish the sliders could be adjusted without having the entire auto processing of ALL filters having to be redone. I’m not sure where in the process chain these color filters occur, but if at the end, perhaps the earlier applied noise, sharpening, upscale, etc. could remain as is without having to reprocess everything.

Regarding Video AI: Most of my work with it so far has been on old film, VHS, Hi8 and DV restoration, so SD cleanup is very important. I do still get requests to have the material returned on DVD. It would be nice to also include an SD interlaced output option directly in VAI. I know that would be low priority, though. I’m having trouble finding a good way to reinterlace processed VAI deinterlaced footage externally.

And lastly, I miss the days of the weekly or monthly product webinars where professional users shared tips on Topaz products. You have no idea how much I learned from some of those!
Also, as at least one other person mentioned, I STILL use Topaz Studio and Studio 2 from time to time. There are some features in those that I can’t seem to replicate elsewhere. I also use the legacy NR, sharpen and Mask plugins as well. Some things just still work . . . . AI ain’t everything.

Regards,

Tom

A small thing, implementing full-sized JPG previews into DNG files created by Photo AI

Descreening halftones.

Lower face AI recovery to 40 instead of default 60, that’s what I’ve used for my old photos. Gives a more natural look.

Well, despite my original statement of “not looking for Topaz alternatives,” I looked for them, found something, will try. I uninstalled Photo AI version 2.4 and went back to 2.3.2 which will last for a while. Not bad. Version 2.4 is unusable for me, but may serve others well. Given the high cost of upgrades, I can’t risk paying for something that goes in the direction indicated in 2.4. As I searched for a possible alternative, I saw many attempts to use AI for photo editing. Perhaps there will be something promising, if not immediately, then over time. The development of the competition has picked up speed. The roadmap indicated by version 2.4 does not suit me. I wish the Topaz PAI product future success and new customers on the new chosen road.

To me it reads as if photo AI is exactly the right software for you to have fun with, it may cost a bit but the alternatives are not there in my opinion.

With the other manufacturers, you have software that is easy to use at first glance, but the output quality is not good.

Incidentally, you can park the operating part of the tool s anywhere.

And TPAI remembers the position, cool thing, like with capture one.



Thanks, Thomas, for your comment and examples. Yes, I chose Topaz Photo AI because it looked perfect for my needs. Before that I had been using three separate tools (Gigapixel, Denoise, Sharpen, plus the – to me – useless and unused Adjust). Then I decided on Topaz Photo AI and it worked quite well for my needs. I would hate to change it. I don’t like 2.4 though; I have returned back to 2.3.2 mainly because of the UI – I’m secretly hoping that the last TPAI 2.4 trend will not continue and maybe some next update/upgrade could take into account a lot of user complaints.

You are right, there is a problem with alternatives (I am still only writing about my purely subjective point of view). I tried maybe 10 applications (PC, Windows 10) and most were as you write, “easy to use at first glance, but the output quality is not good.” I would hardly utilize sophisticated tools like Adobe Photoshop, it would be like “to use a sledgehammer to crack a nut,” partly also because of the price and type of license. I have found that HiPaw Photo AI software might be useful – it gave me nice results on a number of photos, especially zooming in at 30x (camera), and then (computer) cropping, magnifying, and applying the “denoise” tool. HiPaw is easy for me to control. For now, I will make do with Topaz PAI 2.3.2 and will wait for the further development of the situation.

That’s it for my amateur user type “apply some good Photo AI to my imperfect photos”. Jan

Why would you use Photoshop instead of Affinity Photo?

The thing with the new interface is that you are able to use denoise 2x instead of 1x on the same picture.

So ISO 25600 images will look like ISO 100 (Or denoised ISO 3200 image).

Good advice. I have Affinity v. 1 but as I can see there is already a newer v. 2 which is much better. So, I have just upgraded to v. 2. Hopefully Topaz PAI 2.4.0 will be upgraded to something more suitable, too. Jan

What I need to know about the Roadmap is the RAM requirements as Topaz upgrades. Currently I am running with 16gb and Topaz runs to the SSD for almost everything(needing 24-28gb). I would need a minimum of 36GB of unified memory on my MBP to run without drawing from the slow SSD. Will I need 48 or 64 gb as time goes on?

2 Likes

I did go with 128 gb with my workstation bc i do fill up ram very often to 70GB.