Deep Prime vs Topaz Photo AI

I just saw this video by Dave Kelly. Have to admit , I could hardly believe what I saw.
Jump to 14 min to see the comparison of Topaz vs Deep Prime in Pure Raw (I guess Deep PRIME in PL5 could remove more noise )

It’s still in beta , but are supposed to be released in September.

In my experience, the Topaz Sharpen and DeNoise AI tools already do a better job than DeepPRIME in blurred areas. It’s much the same with Photo AI according to this video. However, taking everything into consideration, I think Topaz Photo AI is doing too much noise reduction. At 15:20, look at the green leaves. Horribly uniform, with loss of fine detail / microcontrast. Even in the blurred regions the noise reduction seems to go a bit too far for my tastes - at least the way the software is being used in this video.

My own workflow is to stick with PhotoLab and then add Sharpen AI or DeNoise AI when I need a better result in a given part of the photo.

5 Likes

It may well be superior in some circumstances, but the guy presenting the video is schilling for the company since he makes a profit on the sale of Topaz products, so whatever he says or shows has to be taken with a grain of salt.

In his so called comparison with Pureraw’s version of DeepPRIME, it is true that Topaz appears to have removed more noise, but it also removed more fine detail which he failed to mention. It is similar to a magician’s redirection so you only focus on what he wants you to see. I looked very carefully at his comparison on my 28inch 4K monitor and there was significantly less fine detail in the Topaz version. And remember he was comparing the two images at 400%. To me, loss of fine detail in pursuit of low noise is unacceptable…

Here is his side by side comparison between the Topaz version and the PureRaw version zoomed out. Although it is zoomed out the superior retention of fine detail in DxO’s version on the right is very obvious even in this screen grab when you view it at full resolution. It is also sharper with more vibrant colors. Remember DeepPRIME is applied as part of DxO’s superior demosaicing.

Mark

10 Likes

I have a 32" 4K monitor , and having trouble finding differences in detail.
There should be no question that Deep PRIME has potential for more development.

My workflow is usually PL5 > Topaz sharpen/Denoise > Affinity.

2 Likes

Depending on your eyesight and viewing distance, a larger monitor will make it more difficult to assess the effectiveness of detail retention, for the same reason as viewing at 400% does.

Every time I zoom in beyond 100% to look at details in a specific area of a photo, I lose sight of what it actually looks like. For me, zooms beyond 100% are only of use for ‘scientific curiosity’. The look of the final exported image at typical viewing size is what ultimately counts.

For what it’s worth, I trialled Topaz DeNoise AI last year and found that on some photos it did a visibly better job than DeepPRIME while on some photos it was visibly worse. To me, with both technologies being based on machine learning, that says they trained their systems on different types of photos.

3 Likes

In the video , he also tested Topaz Denoise , but the new Photo AI is just next level of removing noise.

Topaz has never been good with RAW file (even though they recommend using RAW file) , often giving some strange artifacts. It’s better to use a decent RAW editor > tiff> Topaz.

While I’ve had some great results from DeNoise AI, I’ve also had some dreadful results. DeNoise has a habit of doing strange stuff with straight lines. Things like communications towers, open stairway structures, strong rectangular architectural details (several stories of windows), and close parallel lines can all end up distorted. Face melt on distant humans, and leaves turned into globs are also not uncommon. DeNoise AI can work for you, but I don’t trust it enough to just let it loose on automatic.

Photo AI combines the Topaz AI Triad (Gigapixel, Sharpen, and DeNoise), so the software gives you several variables for which you don’t have the same level of control that you do when using them individually.

I was offered an install of Photo AI when I renewed my Topaz support (I let it expire and was just waiting around for the inevitable “discount”). First thing I noticed after I installed it was my file association for JPEGs shifted to Photo AI. I don’t recall being asked, and I didn’t feel like I should have to go into Windows and change it back. I could see no method of changing that in the Photo AI menus. I wouldn’t be surprised if Photo AI changes those file associations every time there is an update.

I ran Photo AI on a some files, and I just couldn’t see the advantage for my workflow, and figured my best course of action was to uninstall the software. I think that the market for Photo AI is the smart phone/social media crowd.

And as a point of comparison, I got PureRAW2 when DXO was finally able to handle Fujifilm X-Trans files. My experience there is also varied. To be safe, I sometimes process RAW files through both PureRAW2 and Iridient X-Transformer because sometimes there is just too much of a mess left behind with PureRAW2. But it varies, and PureRAW2 has given new life to files I could only take so far previously.

3 Likes

Yes, I’m inclined to agree. It has fewer options than the three separate products and (still in beta?) gives unimpressive results when noise levels are high.

I recently made an inadvertent experiment with my OM-1: I set it to ISO 80,000 (twiddling controls in the sunlight without my spectacles) and took several indoor shots. The images are like a messy painting by Seurat.

So when I downloaded Photo AI they were the first images I tested: the results were disappointing. Noise artefacts were gone and colours were more or less uniform. But the details were all smeared as if rendered in oils with a trowel – or perhaps like a quick run from a low-end DALL-E implementation.

Topaz AI (stand-alone) did a better job using the less aggressive “standard” model. Noise remained but it could be further reduced by the “high quality” setting in DXO PL (the image is now a TIFF, so DeepPrime won’t work).

DXO DeepPrime alone did a superior – surprisingly good – job of minimising noise with only slight smearing and a good level of fine detail. The images look like they’d be useable for modest-sized prints. Colour was good, too: not without some aliasing but only on close inspection.

P

1 Like

I did a little test with a 16k ISO OM-1 picture, and while Photo AI effectively removes all the noise in the bokeh area, it lost a lot of details in the bird’s feathers.

So for my purpose, I think that PL5 is still a better choice. But I guess that’s a matter of personal taste!

Here is the Photo AI picture:

And here is the PL5 one:

2 Likes

While I can acknowledge the diversity of other products, Its DxO PL for me

Others are welcome to have fun with Topaz, Capture One, Affinity, Adobe LR, etc… if those workflows and results work for you.

4 Likes

I think you should change the title, because it’s deceiving like the video you are referring to, which is a prime example of a disgusting shill at work.
Noise reduction is not just about “removing noise” (something that can be done with any software), it’s about removing noise while still leaving sharpness and DETAIL. The comparison shows that Deep Prime is still the superior product.

If this guy in the video was genuine I would advise him to visit a doctor and check his eyes. But because he is a shill who makes money with Topaz, while deceiving and lying to people who seem to have no clue what they are looking at, I assume that he is doing it on purpose.

The Topaz result is soft, all details and tone variations are gone! The supposed “noise” that Deep Prime is leaving (which I even find appealing) has to be enlarged to 400%, because otherwise no one would notice it.

P.s. I don’t have Photolab, or Topaz. I’m waiting for PL6, while still looking evaluating alternative products. I thought I would find something useful, but instead I wasted my time on another YouTube shill.

6 Likes

One major difference is that when using DeepPRIME in PL5, you have no need to create a TIFF copy just to do the denoising, so you can change any of the demosaïcing settings at any time prior to exporting the final image for printing or publishing.

One of the primary reasons for my choosing PhotoLab was that I no longer have to stuff my disks full with both RAW and TIFF files of the same images.

7 Likes

It is annoying to have to create TIFFs in e.g. an LR > Topaz workflow and save them to disk alongside the raws.

But it’s also annoying that we cannot see the results of “Deep Prime” processing until we export the file from PL. Admittedly, we then have the option of exporting a JPEG that is likely to be much less weighty than a TIFF.

Fortunately, by exporting to the “Original folder”, PL ‘automatically’ imports this processed version into the catalog and, if you choose to export with “All” metadata, displays it alongside the adjusted raw in the PhotoLIbrary using whatever library sort option you have chosen.

If there were ‘macros’ in PL, this is probably the first one I’d create. As it is, I use a “Keyboard Maestro” macro (on a Mac) to achieve the same result.

Correct
This is also a frequently discussed problem and, according to DXO, results from the fact that the computing power for viewing a complete image is too great and the display update therefore takes too long.
During EA5 we discussed whether it should not be left to the user to choose between DeepPrime display of the whole image or DeepPrime of the section.
Time for a FR?

best regards

Guenter

1 Like

I completely agree. If you look at an earlier post of mine in this thread you will see it that I mentioned more or less the same points.

Mark

Unless Topaz has changed from when I last used my licensed copy of the previous version, you can run it directly against raw files with jpegs as output. That license version is uninstalled and long gone from my system.

Mark

I have PL5 Elite + Filmpack 6 Elite + View Point
I also have Topaz Studio 2 + Denoise + Sharpen

I have just canceled my Adobe subscription because i think I’m getting better result from DxO. In most cases ( at least 95% ) I don’t need to go to Topaz.

Leaving Adobe wasn’t easy. Lightroom has some features that is quit unique.

Why not, an appropriate GPU can render the image in less than 10 seconds. All others will want to stay with a loupe window.

In that video I posted , there was a very high iso / noisy image.
Here is the iso lower:

Loss of details ?

Comparison of night image 1000 ISO…

Topaz…

DeepPRIME…

Sorry Topaz - no contest.

3 Likes