Neo GenErase vs Affinity InPaint

Digital Nigel

Forum Pro
Messages
22,406
Solutions
37
Reaction score
10,977
Location
London, UK
Luminar Neo's new AI-based GenErase tool does essentially the same job as Affinity's InPaint tool. In both, you paint an area of the image, and it fills it with generated pixels inspired by the surrounding area. Affinity's tool has been around for years, makes no AI claims, and is fast and easy to use. So I thought I'd compare them:

This is the original image, taken at Kjosfossen waterfall on the Flåm Railway in Norway.



Original image — I wanted to removed the concrete reinforcement section in the bottom right
Original image — I wanted to removed the concrete reinforcement section in the bottom right

First I tried Neo's GenErase. Despite multiple attempts, i couldn't get it to entirely remove the grey band where it failed:



Neo GenErase. Note the thin grey strip and blurring along the bottom left. I couldn't get rid of this despite numerous attempts.
Neo GenErase. Note the thin grey strip and blurring along the bottom left. I couldn't get rid of this despite numerous attempts.

Then I tried Affinity's InPaint brush:



Affinity InPaint. Much quicker and easier, worked on the first attempt with no failed grey bits.
Affinity InPaint. Much quicker and easier, worked on the first attempt with no failed grey bits.

If you look carefully, you can see where it's duplicated existing areas, so I'd probably do some cloning to cover them, but I've left them here.



Comparing them:



Comparing the original with the twoe generated versions
Comparing the original with the twoe generated versions





They're obviously both fake, but I can't decide which is the more natural looking. Which do you prefer?
 
Comparing them:

Comparing the original with the twoe generated versions
Comparing the original with the twoe generated versions

They're obviously both fake, but I can't decide which is the more natural looking. Which do you prefer?
I probably wouldn't have noticed the fix in either one if I wasn't told about it.

One thing about AI-based tools is that they produce a different result every time. It looks like the Affinity Photo mechanism essentially does a clone from a nearby area. The cloning is very apparent once you realize it's there.
 
Last edited:
Just for comparison:

96e18ab416c84530b0ac8eb06cd16a00.jpg



--
———————————————————————------
"View their gallery before accepting their comments."
———————————————————————------
My photos: http://www.gordonpritchard.blogspot.com/
 
Comparing them:

Comparing the original with the twoe generated versions
Comparing the original with the twoe generated versions

They're obviously both fake, but I can't decide which is the more natural looking. Which do you prefer?
I probably wouldn't have noticed the fix in either one if I wasn't told about it.

One thing about AI-based tools is that they produce a different result every time. It looks like the Affinity Photo mechanism essentially does a clone from a nearby area. The cloning is very apparent once you realize it's there.
Yes, I already knew it used cloning, so I went looking for it, and it wasn't hard to find. I'm not sure I'd have noticed it otherwise. But Affinity was so much quicker and pain-free to use that I'd be tempted to use it anyway rather than this first release of the GenErase tool (which loses all exif data).
 
In V1 of Gen Erase, Neo has a problem with edges. It puts a gray rectangle along the edge.

This video shows the problem at 6:40:



fc4db524e6a04b6cbf3299b3e5dbfc2f.jpg





After a few attempts, I was able to get one with no gray bar:



 No gray rectangle...
No gray rectangle...



--
**** REDACTED ****
 
Last edited:
I find that the AP results depend quite a lot on the brush size v. the area being edited.

It works great for "spotting" however if the surrounding area is patterned the results can be frustrating.

It is quick and easy to use and mostly provides pleasing results.

I'm very happy with AP. I'd have to upgrade my hardware to take advantage of newer "AI" tools and I don't often find that I >need< to do AI-based edits.

I'm also very happy with Neat Image's noise handling which is also old-school programming and which does an amazing job really fast.

- Gary
 
Luminar Neo's new AI-based GenErase tool does essentially the same job as Affinity's InPaint tool. In both, you paint an area of the image, and it fills it with generated pixels inspired by the surrounding area. Affinity's tool has been around for years, makes no AI claims, and is fast and easy to use. So I thought I'd compare them:

This is the original image, taken at Kjosfossen waterfall on the Flåm Railway in Norway.

Original image — I wanted to removed the concrete reinforcement section in the bottom right
Original image — I wanted to removed the concrete reinforcement section in the bottom right

First I tried Neo's GenErase. Despite multiple attempts, i couldn't get it to entirely remove the grey band where it failed:

Neo GenErase. Note the thin grey strip and blurring along the bottom left. I couldn't get rid of this despite numerous attempts.
Neo GenErase. Note the thin grey strip and blurring along the bottom left. I couldn't get rid of this despite numerous attempts.

Then I tried Affinity's InPaint brush:

Affinity InPaint. Much quicker and easier, worked on the first attempt with no failed grey bits.
Affinity InPaint. Much quicker and easier, worked on the first attempt with no failed grey bits.

If you look carefully, you can see where it's duplicated existing areas, so I'd probably do some cloning to cover them, but I've left them here.

Comparing them:

Comparing the original with the twoe generated versions
Comparing the original with the twoe generated versions

They're obviously both fake, but I can't decide which is the more natural looking. Which do you prefer?
Digital Nigel,

Anytime we do a retouch, we always know what we changed. There is nothing we can do about that. The question is, whether it is good enough that others would not see it?

In the two cases you have shown, I would find it hard to have noticed either one as being altered, and that means they both pass the test in my opinion. If I was seeing either of those photos for the first time, I simply would not have realized what you changed.

Just for comparison purposes, here is a version from Photoshop with Generative Fill used on that area.



d28021e8c9af4bc9a1cf7adca80ab517.jpg





--
Major Jack
"You are welcome to retouch any photograph I post in these forums without prior consent from me, just please respect my copyright."
 
  • Like
Reactions: Lan
Luminar Neo's new AI-based GenErase tool does essentially the same job as Affinity's InPaint tool. In both, you paint an area of the image, and it fills it with generated pixels inspired by the surrounding area. Affinity's tool has been around for years, makes no AI claims, and is fast and easy to use. So I thought I'd compare them:

This is the original image, taken at Kjosfossen waterfall on the Flåm Railway in Norway.

Original image — I wanted to removed the concrete reinforcement section in the bottom right
Original image — I wanted to removed the concrete reinforcement section in the bottom right

First I tried Neo's GenErase. Despite multiple attempts, i couldn't get it to entirely remove the grey band where it failed:

Neo GenErase. Note the thin grey strip and blurring along the bottom left. I couldn't get rid of this despite numerous attempts.
Neo GenErase. Note the thin grey strip and blurring along the bottom left. I couldn't get rid of this despite numerous attempts.

Then I tried Affinity's InPaint brush:

Affinity InPaint. Much quicker and easier, worked on the first attempt with no failed grey bits.
Affinity InPaint. Much quicker and easier, worked on the first attempt with no failed grey bits.

If you look carefully, you can see where it's duplicated existing areas, so I'd probably do some cloning to cover them, but I've left them here.

Comparing them:

Comparing the original with the twoe generated versions
Comparing the original with the twoe generated versions

They're obviously both fake, but I can't decide which is the more natural looking. Which do you prefer?
Digital Nigel,

Anytime we do a retouch, we always know what we changed. There is nothing we can do about that. The question is, whether it is good enough that others would not see it?

In the two cases you have shown, I would find it hard to have noticed either one as being altered, and that means they both pass the test in my opinion. If I was seeing either of those photos for the first time, I simply would not have realized what you changed.
Yes, you're right, I'm probably being too picky.
Just for comparison purposes, here is a version from Photoshop with Generative Fill used on that area.

d28021e8c9af4bc9a1cf7adca80ab517.jpg
That one is better.
 

Keyboard shortcuts

Back
Top