Topaz Sharpen AI and Topaz Gigapixel AI

If you're starting with a raw file, use Gigapixel to process and upscale. Then use the others if required.
Thanks Lan. There are probably many different ways folks use these filters. Different to how Topaz recommends their usage order.
My response was based on my own testing, but I was starting with dSLR images taken at base ISO. I think the best workflow will depend on the starting image.

Noisy images will probably be best taken via DeepPRIME first.

Do try my method though, I get better detail starting with Gigapixel than I do starting with DxO.

Unfortunately Gigapixel doesn't correct CA, or any other lens aberrations...
Thanks for the input Lan. I'm just starting out with those filters, so any experience from others is worth looking into...
I get the best results overall using PhotoLab for development (including DeepPRIME), followed by Gigapixel and finally Sharpen AI. Although Topaz recommends using its tools directly in raw images, they just aren't very good at raw development. However, it's really only worth using Gigapixel if you need more than 2x enlargement; up to that point, it's not noticeably better than simple bicubic.
Funnily enough I use it precisely for 2x enlargement, and my experience is that it does significantly better from raw files in that scenario. I am using APS-C or full-frame at base ISO though; if you're using a noisy source image I could see DxO being a better starting point. At the end of the day use whatever works best for you!
Certainly with a high ISO shot it is best to use DXO first. Here is a 50Mp file derived from the 12Mp Olympus Stylus 1 which has a 1/1.7" sized sensor. Excellent noise performance considering the source and 50Mp output size:



DXO pre process on left; Topaz raw conversion on right
DXO pre process on left; Topaz raw conversion on right

Ian
 
We the users have a decision to make, Adobe Camera Raw needs the raw file for max effect of Enhance, Super-Resoution, DevelopProfiles etc, DxO requires the raw file for their Deep Noise reduction, Topaz needs the raw file for Sharpen AI, etc, etc. Only one can have the first shot.

Do your tests to see what works for you.
ACR Enhance Details and Super resolution are much better than Gigapixel and about ten times faster. Gigapixel took about eight hours on my laptop to upres a D850 raw.

Topaz Denoise AI does an excellent job of cleaning up any noise.



Gigapixel
Gigapixel



ACR Enhance Details, Super Resolution
ACR Enhance Details, Super Resolution



Gigapixel 2X
Gigapixel 2X



ACR 2X
ACR 2X



D850
D850



DPR
DPR

I like and regularly use three Topaz plugins, but Gigapixel is not going to be one of them, especially at $99.

Adobe has set the bar very high indeed for fine detail.
 
We the users have a decision to make, Adobe Camera Raw needs the raw file for max effect of Enhance, Super-Resoution, DevelopProfiles etc, DxO requires the raw file for their Deep Noise reduction, Topaz needs the raw file for Sharpen AI, etc, etc. Only one can have the first shot.

Do your tests to see what works for you.
ACR Enhance Details and Super resolution are much better than Gigapixel and about ten times faster. Gigapixel took about eight hours on my laptop to upres a D850 raw.

Topaz Denoise AI does an excellent job of cleaning up any noise.

I like and regularly use three Topaz plugins, but Gigapixel is not going to be one of them, especially at $99.

Adobe has set the bar very high indeed for fine detail.
Some of the comments after this article are worth a read:

Photoshop Super Resolution vs Topaz Gigapixel AI: Upscaling Throwdown (petapixel.com)

I agree with Mark Smiths statement/comment

-Martin P
 
For anyone with a slow machine, it obviously makes sense to process the raw in a good raw processor (not Topaz), reduce noise (who wants to enlarge noise?) and crop before using Gigapixel. If you only want to enlarge a small part of the image, then only give Gigapixel that small crop. One of the problems with using Gigapixel on a raw file is that you have to enlarge the whole image.

For example, you might capture a distant bird, and it might only occupy 2mp of a 40mp image. If you want to grow it 4x in each dimension, Gigapixel will happily grow the 2mp to 32 fairly quickly, but it will actually fail to upscale a 40mp image 4x, as the max side is 32,000 pixels.

So my recommendations are:
  • Don't upscale images that aren't sharp — they'll look bad, however much you sharpen them.
  • Don't upscale noisy images — Gigapixel tends to turn noise into ugly artefacts.
  • You don't really need Gigapixel for a 2x upscale. Other methods produce very similar results, without artefacts, more quickly. Even bimode is often good enough.
  • Don't process raw images in Gigapixel; process them in a better raw processor, adjust shadows, etc, reduce noise, crop, and only then feed the TIFF to Gigapixel.
  • If the image is noisy, it's best to use DeepPRIME when processing the raws. DeNoise AI isn't nearly as good.
  • Gigapixel does a good job with 4x upscaling, and even 6x works well with some images. But don't try to go higher. If you have to go higher, do it in stages, sharpening in between.
  • Gigapixel still does better with bird feathers and animal fur than buildings and human faces, despite supposedly now being trained to deal with man-made objects and human faces.
  • It's usually worth using Sharpen AI on the final result.
 
For anyone with a slow machine, it obviously makes sense to process the raw in a good raw processor (not Topaz), reduce noise (who wants to enlarge noise?) and crop before using Gigapixel. If you only want to enlarge a small part of the image, then only give Gigapixel that small crop. One of the problems with using Gigapixel on a raw file is that you have to enlarge the whole image.

For example, you might capture a distant bird, and it might only occupy 2mp of a 40mp image. If you want to grow it 4x in each dimension, Gigapixel will happily grow the 2mp to 32 fairly quickly, but it will actually fail to upscale a 40mp image 4x, as the max side is 32,000 pixels.

So my recommendations are:
  • Don't upscale images that aren't sharp — they'll look bad, however much you sharpen them.
  • Don't upscale noisy images — Gigapixel tends to turn noise into ugly artefacts.
  • You don't really need Gigapixel for a 2x upscale. Other methods produce very similar results, without artefacts, more quickly. Even bimode is often good enough.
  • Don't process raw images in Gigapixel; process them in a better raw processor, adjust shadows, etc, reduce noise, crop, and only then feed the TIFF to Gigapixel.
  • If the image is noisy, it's best to use DeepPRIME when processing the raws. DeNoise AI isn't nearly as good.
  • Gigapixel does a good job with 4x upscaling, and even 6x works well with some images. But don't try to go higher. If you have to go higher, do it in stages, sharpening in between.
  • Gigapixel still does better with bird feathers and animal fur than buildings and human faces, despite supposedly now being trained to deal with man-made objects and human faces.
  • It's usually worth using Sharpen AI on the final result.
Thanks Nigel. I've been playing around with the four newly acquired programs from Topaz with varying results. I've been processing in the order that they recommend here:

The Ultimate Workflow for Topaz Labs AI Software – Topaz Labs

Still early days yet for me, but I will try other ways and compare results. I'm actually finding gigapixel over sharpens somewhat.

-Martin P
 
For anyone with a slow machine, it obviously makes sense to process the raw in a good raw processor (not Topaz), reduce noise (who wants to enlarge noise?) and crop before using Gigapixel. If you only want to enlarge a small part of the image, then only give Gigapixel that small crop. One of the problems with using Gigapixel on a raw file is that you have to enlarge the whole image.

For example, you might capture a distant bird, and it might only occupy 2mp of a 40mp image. If you want to grow it 4x in each dimension, Gigapixel will happily grow the 2mp to 32 fairly quickly, but it will actually fail to upscale a 40mp image 4x, as the max side is 32,000 pixels.

So my recommendations are:
  • Don't upscale images that aren't sharp — they'll look bad, however much you sharpen them.
  • Don't upscale noisy images — Gigapixel tends to turn noise into ugly artefacts.
  • You don't really need Gigapixel for a 2x upscale. Other methods produce very similar results, without artefacts, more quickly. Even bimode is often good enough.
  • Don't process raw images in Gigapixel; process them in a better raw processor, adjust shadows, etc, reduce noise, crop, and only then feed the TIFF to Gigapixel.
  • If the image is noisy, it's best to use DeepPRIME when processing the raws. DeNoise AI isn't nearly as good.
  • Gigapixel does a good job with 4x upscaling, and even 6x works well with some images. But don't try to go higher. If you have to go higher, do it in stages, sharpening in between.
  • Gigapixel still does better with bird feathers and animal fur than buildings and human faces, despite supposedly now being trained to deal with man-made objects and human faces.
  • It's usually worth using Sharpen AI on the final result.
Thanks Nigel. I've been playing around with the four newly acquired programs from Topaz with varying results. I've been processing in the order that they recommend here:

The Ultimate Workflow for Topaz Labs AI Software – Topaz Labs

Still early days yet for me, but I will try other ways and compare results. I'm actually finding gigapixel over sharpens somewhat.
Perhaps in some cases, but it can't do the 'motion blur' type sharpening that's Sharpen AI's forte. That's what I normally apply in the final stage.
 
For anyone with a slow machine, it obviously makes sense to process the raw in a good raw processor (not Topaz), reduce noise (who wants to enlarge noise?) and crop before using Gigapixel. If you only want to enlarge a small part of the image, then only give Gigapixel that small crop. One of the problems with using Gigapixel on a raw file is that you have to enlarge the whole image.

For example, you might capture a distant bird, and it might only occupy 2mp of a 40mp image. If you want to grow it 4x in each dimension, Gigapixel will happily grow the 2mp to 32 fairly quickly, but it will actually fail to upscale a 40mp image 4x, as the max side is 32,000 pixels.

So my recommendations are:
  • Don't upscale images that aren't sharp — they'll look bad, however much you sharpen them.
  • Don't upscale noisy images — Gigapixel tends to turn noise into ugly artefacts.
  • You don't really need Gigapixel for a 2x upscale. Other methods produce very similar results, without artefacts, more quickly. Even bimode is often good enough.
  • Don't process raw images in Gigapixel; process them in a better raw processor, adjust shadows, etc, reduce noise, crop, and only then feed the TIFF to Gigapixel.
  • If the image is noisy, it's best to use DeepPRIME when processing the raws. DeNoise AI isn't nearly as good.
  • Gigapixel does a good job with 4x upscaling, and even 6x works well with some images. But don't try to go higher. If you have to go higher, do it in stages, sharpening in between.
  • Gigapixel still does better with bird feathers and animal fur than buildings and human faces, despite supposedly now being trained to deal with man-made objects and human faces.
  • It's usually worth using Sharpen AI on the final result.
Thanks Nigel. I've been playing around with the four newly acquired programs from Topaz with varying results. I've been processing in the order that they recommend here:

The Ultimate Workflow for Topaz Labs AI Software – Topaz Labs

Still early days yet for me, but I will try other ways and compare results. I'm actually finding gigapixel over sharpens somewhat.
Perhaps in some cases, but it can't do the 'motion blur' type sharpening that's Sharpen AI's forte. That's what I normally apply in the final stage.
Here's an example with Sharpen AI after the other steps were taken. The Gigapixel is too high res to post. [ this is a heavy crop]

c1e45c13f6974877947e3e431ee6d592.jpg

-Martin P
 
Last edited:
For anyone with a slow machine, it obviously makes sense to process the raw in a good raw processor (not Topaz), reduce noise (who wants to enlarge noise?) and crop before using Gigapixel. If you only want to enlarge a small part of the image, then only give Gigapixel that small crop. One of the problems with using Gigapixel on a raw file is that you have to enlarge the whole image.

For example, you might capture a distant bird, and it might only occupy 2mp of a 40mp image. If you want to grow it 4x in each dimension, Gigapixel will happily grow the 2mp to 32 fairly quickly, but it will actually fail to upscale a 40mp image 4x, as the max side is 32,000 pixels.

So my recommendations are:
  • Don't upscale images that aren't sharp — they'll look bad, however much you sharpen them.
  • Don't upscale noisy images — Gigapixel tends to turn noise into ugly artefacts.
  • You don't really need Gigapixel for a 2x upscale. Other methods produce very similar results, without artefacts, more quickly. Even bimode is often good enough.
  • Don't process raw images in Gigapixel; process them in a better raw processor, adjust shadows, etc, reduce noise, crop, and only then feed the TIFF to Gigapixel.
  • If the image is noisy, it's best to use DeepPRIME when processing the raws. DeNoise AI isn't nearly as good.
  • Gigapixel does a good job with 4x upscaling, and even 6x works well with some images. But don't try to go higher. If you have to go higher, do it in stages, sharpening in between.
  • Gigapixel still does better with bird feathers and animal fur than buildings and human faces, despite supposedly now being trained to deal with man-made objects and human faces.
  • It's usually worth using Sharpen AI on the final result.
Thanks Nigel. I've been playing around with the four newly acquired programs from Topaz with varying results. I've been processing in the order that they recommend here:

The Ultimate Workflow for Topaz Labs AI Software – Topaz Labs

Still early days yet for me, but I will try other ways and compare results. I'm actually finding gigapixel over sharpens somewhat.
Perhaps in some cases, but it can't do the 'motion blur' type sharpening that's Sharpen AI's forte. That's what I normally apply in the final stage.
Here's an example with Sharpen AI after the other steps were taken. The Gigapixel is too high res to post. [ this is a heavy crop]

c1e45c13f6974877947e3e431ee6d592.jpg

-Martin P
What settings did you use in Gigapixel auto or manual?

Can you put the fike in Dropbox etc?

Ian
 
For anyone with a slow machine, it obviously makes sense to process the raw in a good raw processor (not Topaz), reduce noise (who wants to enlarge noise?) and crop before using Gigapixel. If you only want to enlarge a small part of the image, then only give Gigapixel that small crop. One of the problems with using Gigapixel on a raw file is that you have to enlarge the whole image.

For example, you might capture a distant bird, and it might only occupy 2mp of a 40mp image. If you want to grow it 4x in each dimension, Gigapixel will happily grow the 2mp to 32 fairly quickly, but it will actually fail to upscale a 40mp image 4x, as the max side is 32,000 pixels.

So my recommendations are:
  • Don't upscale images that aren't sharp — they'll look bad, however much you sharpen them.
  • Don't upscale noisy images — Gigapixel tends to turn noise into ugly artefacts.
  • You don't really need Gigapixel for a 2x upscale. Other methods produce very similar results, without artefacts, more quickly. Even bimode is often good enough.
  • Don't process raw images in Gigapixel; process them in a better raw processor, adjust shadows, etc, reduce noise, crop, and only then feed the TIFF to Gigapixel.
  • If the image is noisy, it's best to use DeepPRIME when processing the raws. DeNoise AI isn't nearly as good.
  • Gigapixel does a good job with 4x upscaling, and even 6x works well with some images. But don't try to go higher. If you have to go higher, do it in stages, sharpening in between.
  • Gigapixel still does better with bird feathers and animal fur than buildings and human faces, despite supposedly now being trained to deal with man-made objects and human faces.
  • It's usually worth using Sharpen AI on the final result.
Thanks Nigel. I've been playing around with the four newly acquired programs from Topaz with varying results. I've been processing in the order that they recommend here:

The Ultimate Workflow for Topaz Labs AI Software – Topaz Labs

Still early days yet for me, but I will try other ways and compare results. I'm actually finding gigapixel over sharpens somewhat.
Perhaps in some cases, but it can't do the 'motion blur' type sharpening that's Sharpen AI's forte. That's what I normally apply in the final stage.
Here's an example with Sharpen AI after the other steps were taken. The Gigapixel is too high res to post. [ this is a heavy crop]

-Martin P
What settings did you use in Gigapixel auto or manual?

Can you put the fike in Dropbox etc?

Ian
Here's a larger version Ian

 gigapixel- auto mode usage
gigapixel- auto mode usage

-Martin P
 
L
Adobe has set the bar very high indeed for fine detail.
Some of the comments after this article are worth a read:

Photoshop Super Resolution vs Topaz Gigapixel AI: Upscaling Throwdown (petapixel.com)

I agree with Mark Smiths statement/comment

-Martin P
Unfortunately, Mark Smith, whose work is generally excellent didn't do the Enhance Details first, which is about 50% of the difference I've clearly shown above. It's about a three minute addition to the process on a slow computer.

Enhance Details takes out roughly 80% of false color and moire while providing a much better decoding of the fine details. Both the Super Resolution and the Gigapixel crops of the eagle shot in the article are very blurry and not a good candidate for uprezzing. I also question the use of bird pics of indeterminate quality to evaluate sharpness. Test charts work a lot better, because we know to a certainty what the target is supposed to look like, e.i. legible. Which is why we have them.



Gigapixel D850
Gigapixel D850



Enhance Details plus Super Res
Enhance Details plus Super Res

If there was a lens which could get that much more detail (there isn't) we would all be swinging from the chandeliers and backordering it 'til doomsday. But it's a throw in now with the photography plan. Thanks, Adobe!
 
What settings did you use in Gigapixel auto or manual?

Can you put the fike in Dropbox etc?

Ian
Here's a larger version Ian

gigapixel- auto mode usage
gigapixel- auto mode usage

-Martin P
Sorry, Martin, but that rendition has horrible artifacts. Dropbox the raw so we can have a decent shot at it.
 
L
Adobe has set the bar very high indeed for fine detail.
Some of the comments after this article are worth a read:

Photoshop Super Resolution vs Topaz Gigapixel AI: Upscaling Throwdown (petapixel.com)

I agree with Mark Smiths statement/comment

-Martin P
Unfortunately, Mark Smith, whose work is generally excellent didn't do the Enhance Details first, which is about 50% of the difference I've clearly shown above. It's about a three minute addition to the process on a slow computer.

Enhance Details takes out roughly 80% of false color and moire while providing a much better decoding of the fine details. Both the Super Resolution and the Gigapixel crops of the eagle shot in the article are very blurry and not a good candidate for uprezzing. I also question the use of bird pics of indeterminate quality to evaluate sharpness. Test charts work a lot better, because we know to a certainty what the target is supposed to look like, e.i. legible. Which is why we have them.

If there was a lens which could get that much more detail (there isn't) we would all be swinging from the chandeliers and backordering it 'til doomsday. But it's a throw in now with the photography plan. Thanks, Adobe!
Here's a throw a way shot [poor sharpness] straight from the camera to have a crack at for anyone who wishes to do so...I'm not a test chart testing type- kinda prefer the real world scenario.

I'd like to see how others would work/crop it:



sooc j-peg
sooc j-peg

-Martin P
 
What settings did you use in Gigapixel auto or manual?

Can you put the fike in Dropbox etc?

Ian
Here's a larger version Ian

-Martin P
Sorry, Martin, but that rendition has horrible artifacts. Dropbox the raw so we can have a decent shot at it.
Depends what you are looking at...there were a lot of branches/leaves blowing about the place when taking the image, so some of it may be interpreted differently.

I don't use dropbox!

-Martin P
 
What settings did you use in Gigapixel auto or manual?

Can you put the fike in Dropbox etc?

Ian
Here's a larger version Ian

-Martin P
Sorry, Martin, but that rendition has horrible artifacts. Dropbox the raw so we can have a decent shot at it.
Depends what you are looking at...
I'm looking close up at what you posted:



GHO
GHO

there were a lot of branches/leaves blowing about the place when taking the image, so some of it may be interpreted differently.
Mmm no. It's an oversharpened mess. I'm pretty sure a much better result is possible.
I don't use dropbox!
If it's going to be that big of a deal to post the raw to your favorite storage, which is the only way we can evaluate the differences, then never mind...
 
What settings did you use in Gigapixel auto or manual?

Can you put the fike in Dropbox etc?

Ian
Here's a larger version Ian

-Martin P
Sorry, Martin, but that rendition has horrible artifacts. Dropbox the raw so we can have a decent shot at it.
Depends what you are looking at...
I'm looking close up at what you posted:
there were a lot of branches/leaves blowing about the place when taking the image, so some of it may be interpreted differently.
Mmm no. It's an oversharpened mess. I'm pretty sure a much better result is possible.
I don't use dropbox!
If it's going to be that big of a deal to post the raw to your favorite storage, which is the only way we can evaluate the differences, then never mind...


dab23312a9e9426e9baad1ad7cfa96be.jpg

Here's the original J-peg I worked from...feel free to give it a go...

-Martin P
 

Keyboard shortcuts

Back
Top