Generally impressed with Topaz Photo AI

Brueghel

Senior Member
Messages
1,685
Reaction score
1,211
I have DxO Pure Raw 2. It can do an absolutely great job in many situations.

But it left aside my stock of old JPG pictures, mostly coming from scanning old stuff.

Tried the Photo AI that (just?) came out. On many pics, it is a clear WOW ! On others, it tries so hard, even in the absence of critical info, that it invents things and irremediably disfigures some faces... Too bad.

Needs to be used with caution!
 
Last edited:
Yes, that's a good summary of PhotoAI, it can do great things, but you need to keep an eye on it!

It's not like DeepPRIME which you can use reliably 100% of the time.

I use PhotoAI for JPEGs and film scans myself too 🙂
 
Yes, that's a good summary of PhotoAI, it can do great things, but you need to keep an eye on it!

It's not like DeepPRIME which you can use reliably 100% of the time.
That's not been my experience.

Also, you can't use it at all onJpegs.

I use PhotoAI for JPEGs and film scans myself too 🙂

"It's good to be . . . . . . . . . Me!"
 
Yes, that's a good summary of PhotoAI, it can do great things, but you need to keep an eye on it!

It's not like DeepPRIME which you can use reliably 100% of the time.
That's not been my experience.
Do you mean that's not your experience with DeepPRIME or PhotoAI?

Note I said DeepPRIME not DeepPRIME XD, they are different 🙂
Also, you can't use it at all onJpegs.
Agreed, which is why I said:
I use PhotoAI for JPEGs and film scans myself too 🙂
 
Yes, that's a good summary of PhotoAI, it can do great things, but you need to keep an eye on it!

It's not like DeepPRIME which you can use reliably 100% of the time.
That's not been my experience.
Do you mean that's not your experience with DeepPRIME or PhotoAI?
Deep Prime, well, more specifically, all Photolab denoise options, although no noise reduction software can be used 'reliably, 100% of the time.'

Topaz; DeNoise AI and Photo AI, can also turn out some horrors but generally speaking I prefer it.
Note I said DeepPRIME not DeepPRIME XD, they are different 🙂
I know.
Also, you can't use it at all onJpegs.
Agreed, which is why I said:
I use PhotoAI for JPEGs and film scans myself too 🙂
"It's good to be . . . . . . . . . Me!"
 
Yes, that's a good summary of PhotoAI, it can do great things, but you need to keep an eye on it!

It's not like DeepPRIME which you can use reliably 100% of the time.

I use PhotoAI for JPEGs and film scans myself too 🙂
Yes, reflects my experience with DXO and Topaz.

Ian
 
On close up portraits where all main features are recognizable, it does great. Sometimes: amazing.

But if a face is missing some main feature (example: one cannot distinguish the mouth of a person), and if Photo AI is used with the autopilot settings and attempts to reconstruct the portrait. It will invent part of the mouth or the nose, or blank out an eye, etc... Which may lead to the transformation of a face we recognize into the face of a new person. With quite creepy results sometimes :-)

In this case, I have found that one has to personalize the settings and tone them down, reduce the agressivity and not shoot for the moon. Instead of an "amayzing" result, we will obtain a slight, modest, improvement and should be wise enough to accept this as the best possible

No AI engine will ever reconstruct reasonably a face unless it can use other, better, pictures of this face. Until this is possible, we have to be realistic and ask the engine to not even try to reconstruct the face.

Eg: I have only one photo where my grand father appears. Half of his face is really mushy, Then hoping that AI will recover his face is, obviously, a lost cause (whether I am certain that a miracle is possible or not). BUT! One of these days, we may (wild guess) be able to feed the engine better photos of the person, photos it can use to base its reconstruction process on, and hope that the result is nice
 
It keeps getting better. I use it and PureRaw 2. Every file is different and sometimes PAI does better and sometimes PR does. In the fall we will see where PAI is and how much they want for the upgrade.

Rumours about Adobe introducing AI and I'm confident it will be there one day. Hopefully for LrC by v13.
 
We use it quite a lot

Being able to batch process using its own judgement on each image is really helpful. The odd one I go back and reprocess but mostly it's throw them in, wait and job done.

I can't do that with DxO and also the lack of JPEG support can be a problem.
 
Yes, that's a good summary of PhotoAI, it can do great things, but you need to keep an eye on it!

It's not like DeepPRIME which you can use reliably 100% of the time.

I use PhotoAI for JPEGs and film scans myself too 🙂
Using it for slides I've scanned is my intention, though I haven't started yet.

Any tips? Do I just let it go, or do I need to fine tune it?

TIA
 
It keeps getting better. I use it and PureRaw 2. Every file is different and sometimes PAI does better and sometimes PR does. In the fall we will see where PAI is and how much they want for the upgrade.
Paying for a Topaz upgrade? That's not how Topaz works. If you pay the annual maintenance fee, you get all the upgrades that arrive in that year. That might be only one or two for DeNoise AI or 50 for Photo AI.
 
We use it quite a lot

Being able to batch process using its own judgement on each image is really helpful. The odd one I go back and reprocess but mostly it's throw them in, wait and job done.

I can't do that with DxO and also the lack of JPEG support can be a problem.
You can batch process with DxO — the default settings for DeepPRIME generally work pretty well for any image. But, yes, only with raw images. DxO does have HQ NR for JPEGs, but that doesn't do anything special.
 
It keeps getting better. I use it and PureRaw 2. Every file is different and sometimes PAI does better and sometimes PR does. In the fall we will see where PAI is and how much they want for the upgrade.
Paying for a Topaz upgrade? That's not how Topaz works. If you pay the annual maintenance fee, you get all the upgrades that arrive in that year. That might be only one or two for DeNoise AI or 50 for Photo AI.
That was what I meant. The annual fee like Adobe.,
 
We use it quite a lot

Being able to batch process using its own judgement on each image is really helpful. The odd one I go back and reprocess but mostly it's throw them in, wait and job done.

I can't do that with DxO and also the lack of JPEG support can be a problem.
You can batch process with DxO — the default settings for DeepPRIME generally work pretty well for any image. But, yes, only with raw images. DxO does have HQ NR for JPEGs, but that doesn't do anything special.
But it doesn't appear to choose each image on an image by image basis and only process with one set of settings across all images.
 
We use it quite a lot

Being able to batch process using its own judgement on each image is really helpful. The odd one I go back and reprocess but mostly it's throw them in, wait and job done.

I can't do that with DxO and also the lack of JPEG support can be a problem.
You can batch process with DxO — the default settings for DeepPRIME generally work pretty well for any image. But, yes, only with raw images. DxO does have HQ NR for JPEGs, but that doesn't do anything special.
But it doesn't appear to choose each image on an image by image basis and only process with one set of settings across all images.
DeepPRIME automatically adapts to each image. Low noise images will see little or no NR effect, high noise images will see a big effect. You don't need to adjust the settings for each image. Just leave it at 40 or 50 for every image, and it will work well for all of them.

However, I don't advise using XD like that.
 
Last edited:
It keeps getting better. I use it and PureRaw 2. Every file is different and sometimes PAI does better and sometimes PR does. In the fall we will see where PAI is and how much they want for the upgrade.
Paying for a Topaz upgrade? That's not how Topaz works. If you pay the annual maintenance fee, you get all the upgrades that arrive in that year. That might be only one or two for DeNoise AI or 50 for Photo AI.
That was what I meant. The annual fee like Adobe.,
It's not like Adobe. You can choose whether to buy a Topaz maintenance contract or not, or have gaps between contracts. Your perpetual licence remains valid whether or not you renew maintenance.
 
We use it quite a lot

Being able to batch process using its own judgement on each image is really helpful. The odd one I go back and reprocess but mostly it's throw them in, wait and job done.

I can't do that with DxO and also the lack of JPEG support can be a problem.
You can batch process with DxO — the default settings for DeepPRIME generally work pretty well for any image. But, yes, only with raw images. DxO does have HQ NR for JPEGs, but that doesn't do anything special.
But it doesn't appear to choose each image on an image by image basis and only process with one set of settings across all images.
DeepPRIME automatically adapts to each image. Low noise images will see little or no NR effect, high noise images will see a big effect. You don't need to adjust the settings for each image. Just leave it at 40 or 50 for every image, and it will work well for all of them.
Ah okay thanks (ai didn't know that) I don't suppose it sharpens, scales and so on all at once?
However, I don't advise using XD like that.
 
We use it quite a lot

Being able to batch process using its own judgement on each image is really helpful. The odd one I go back and reprocess but mostly it's throw them in, wait and job done.

I can't do that with DxO and also the lack of JPEG support can be a problem.
You can batch process with DxO — the default settings for DeepPRIME generally work pretty well for any image. But, yes, only with raw images. DxO does have HQ NR for JPEGs, but that doesn't do anything special.
But it doesn't appear to choose each image on an image by image basis and only process with one set of settings across all images.
DeepPRIME automatically adapts to each image. Low noise images will see little or no NR effect, high noise images will see a big effect. You don't need to adjust the settings for each image. Just leave it at 40 or 50 for every image, and it will work well for all of them.
Ah okay thanks (ai didn't know that) I don't suppose it sharpens, scales and so on all at once?
Or at all. PhotoLab has fine-tuned lens sharpening, but doesn't do anything for motion blur or OOF images. It can scale images, but only using the basic bicubic algorithm, not AI. It also doesn't do any facial reconstruction.
However, I don't advise using XD like that.
 
We are trying DXO Photolab 6 now. Deep prime works well, the color modules have user controls. Use it mainly for raw conversions vs DPP. We were thinking of adding Topaz Sharpen AI for sharpening. Any experience with the latter vs Photo AI?
 
We are trying DXO Photolab 6 now. Deep prime works well, the color modules have user controls. Use it mainly for raw conversions vs DPP. We were thinking of adding Topaz Sharpen AI for sharpening. Any experience with the latter vs Photo AI?
Sharpen AI is now at the end-of-life stage. All the development work is going into Photo AI, so even if Sharpen AI may still be better for some images, Photo AI is likely to overtake it soon.
 

Keyboard shortcuts

Back
Top