Are we thinking WRONG about AI Editing?

My last two videos had ChatGPT editing my photographs, first in Lightroom, where it went through a folder of my own edited pictures, worked out my style and then edited a raw file the way I would, and then in Photoshop, where it used Photoshop's own tools to remove a colour cast, add a long exposure effect to a sky and do a cutout, hair included.

In both of them it didn't generate a new picture. It moved the same sliders and used the same tools that you and I would, and every one of those adjustments was still there to change afterwards.

The comments on those videos have been brilliant and they've really made me think, but one kept coming up in different words, and the gist of it was this: so now every picture is going to look like yours.

When you watch what ChatGPT is doing in those videos, it's an understandable thing to think. I just don't think it's right, so rather than argue about it, I decided to test it, and the video above is the result.

Putting it to the test

I gave ChatGPT a folder of my portraits, with my edits on them, and asked it to work out how I edit. Then I gave it the raw file of a portrait I'd already edited myself and asked it to edit that in my style.

I'm not going to go through what I did to mine; the two results are side by side in the video, so have a look and see if you can tell which is which before I give it away.

Here's the thing, though. Whichever one you prefer, they both look like one of my portraits, and that's not because of the edit, it's because I took the picture. Everything that makes it look like one of mine happened before the shutter was pressed, and ChatGPT had none of that to work with; it just had my raw file.

The edit is the icing on the cake

The edit is the part of the process we can see on screen, so it's easy to think that's what makes a picture yours, but it's only one part of a much bigger process. I really enjoy editing, and it gives a picture its final look, but for me it's the icing on the cake.

The big part is what goes on between my ears: how I think about a picture before I take it and how I compose it. With a portrait, it's how the person looks in front of the camera and whether they're relaxed or not. With a seascape, it's what goes in the frame and what doesn't.

None of that is in the edit, so even if somebody had the exact colouring, contrast and tonality I use, their pictures still wouldn't look like mine. They'd look like theirs, with my colouring on them.

We can all buy the same kit, spend the same money on the same camera and lens, stand in exactly the same spot and point it at the same thing, and the pictures still won't look the same. There'll be something in each one that's uniquely yours, because of what you like and what you don't, how you were brought up and everything you've looked at over the years. All of that goes into a photograph, and none of it comes out of an edit.

An opinion, or a misunderstanding?

People have their own opinions about AI, and that's a good thing; having an opinion is part of being human, not something artificial. Sometimes an opinion just isn't right, though, and I know that sounds a strange thing to say. The idea that everybody's pictures are going to end up looking the same isn't really an opinion at all. It's a misunderstanding of what photography is.

Using the software, not replacing the picture

That's also why the way ChatGPT did those edits interests me so much. There's a difference between asking an AI to generate a new version of your picture and having it use the same software you and I would. When a picture's generated, you hand it over and get a new one back. When the AI works the sliders and the masks itself, you get your own raw file back at full quality, with every adjustment still there to change.

It's a bit like the difference between someone redrawing your picture and someone sitting next to you showing you what they did with it. Both have their place, but they're very different things.

Two very different photographers

Where it gets interesting is who all of this is for, because there are definitely two different audiences. Take Adobe as an example. On one side they've got working photographers, people with clients and deadlines, editing a shoot with hundreds of pictures that all need the same look. On the other they've got enthusiasts, people who take pictures because they just love doing it.

For the working photographer, imagine being able to hand an AI an actual part of your process. I don't mean something that makes a new picture; I mean something that applies the same look to a whole set of images inside the software, moving the sliders and making the masks without touching the originals, so you can go in and change whatever you want afterwards. That could be a massive time saver.

The hobbyist is a different story, because for them editing isn't a job to get through. They go out and take their pictures because they enjoy it, then come home and edit them with no rush at all, sitting in a room with a bit of music on and maybe a nice glass of something. For them the editing is part of the thing they love, and a one-click fix isn't for them.

However, some of what AI brings is. The Remove Tool in Photoshop has been a massive benefit; jobs that used to take an age now take moments. Anything that helps with selections, hair and fine cutouts especially, is going to be a huge help as well. None of that takes the enjoyment away. It takes away the tedious bits and leaves more time for the parts you actually enjoy.

How I use it

I use AI every single day, for the admin, the research and the writing that goes with running everything on my own. I'm a one-man band, so that's a lot of work it takes off my desk, and I'm not worried in the slightest that other people's pictures are going to look like mine.

What it gives me is time, and that matters to me more than anything, because I absolutely love doing what I do. I got into this creative world because I love being creative, and I love the freedom of it and the escapism it gives me. I don't want anything to replace that. If anything I want more of it, which is exactly how I'm using AI; it takes care of the jobs I'd rather not be doing and gives me back more time with a camera in my hand.

The mood of the moment

When I gave ChatGPT my Lightroom and Photoshop to use, it didn't edit my pictures exactly the way I would. It can pick up the things I tend to do, but when I'm editing I'm very much in the mood of the moment, chopping and changing as I go, and that freedom of just going with it is a big part of why I enjoy editing at all. AI doesn't know the mood or the feeling I want to give a particular picture, and creating that is what I've always loved.

The Photoshop side was different, because those were tasks. The colour cast, the sky and the hair cutout were jobs where I knew what the finished result needed to be, and what I found really interesting was the thought of handing those kinds of tasks to an AI to do by hand, using the software the same way I would, so I've got more time for the creative side. It was an experiment, but it's certainly one that's got me thinking.

ChatGPT Can Now ACTUALLY Use Photoshop, and It's Impressive

ChatGPT can now take control of your computer, moving the mouse, clicking and typing its way through a job the way you would, and the question I've been asked the most about it is whether it can use Photoshop. I've been testing it, and yes, it can.

What makes it interesting is that it isn't generating a new picture. It's working inside Photoshop with the same tools you and I use, which means everything it does is left as layers and masks you can go back into and adjust yourself, at full quality.

To see how far it could go, I gave it three jobs on three different pictures: fixing a colour cast, creating a long exposure sky, and cutting out hair, which is the thing I get asked about more than anything else in Photoshop.

How it works

I'm using ChatGPT 6 Astra on the high setting, in the Work section of the app. I type out what I want in plain language, send it, and it replies with what it intends to do. The first time it needs to use the computer it asks for permission, I click allow, and from then on it's hands off; I just sit back and watch it work.

With each job I asked it to keep everything non-destructive and editable, and not to save the file, just leave it as layers. You can see everything it does as it does it, and you can stop it at any point.

Fixing a colour cast

The first picture was a portrait of a biker with a definite colour cast, a slightly green and magenta look to it. I asked ChatGPT to look at the image, decide for itself which tools to use, and give me a neutral result.

It added a Curves adjustment layer, worked on the individual red, green and blue channels, and set the layer's blend mode to Color so that it only changed the colour and left the brightness and contrast alone, which is a genuinely sensible way to do it. The original background layer was untouched.

Turning the layer off and on, the cast was gone. It's not too bad at all, and I was quite impressed with that.

A long exposure sky

For the second one I wanted to step it up and see whether it could create an effect rather than make a correction. The picture was of my car on a very wet, stormy day, and I asked for the look of a ten second long exposure, with fast moving clouds coming over the top of us.

What really caught my attention was that it used the same technique I use for this myself. It selected the sky, put it onto its own layer, then converted that layer to a Smart Object before applying a Radial Blur, so the filter stays editable afterwards, which I thought was really clever.

Better still, it moved the centre of the blur. By default a Radial Blur spreads out from the middle of the picture, but ChatGPT pulled that point right down to the bottom, so the blur looks as if it's coming from the horizon and the clouds are streaming towards you. That's exactly the thinking you'd want.

I did have to stop it at the very end, because it was trying to save the file, but the effect itself was done.

Cutting out hair

The last job was the hardest, and the one I was most intrigued by. It's a picture of my friend Simon from a Viking shoot, on a grey background, but with lots of fine, loose flyaway hairs and fur cuffs on what he's wearing. I asked for the best cutout it could manage, body, hair and everything as one, output as a layer mask so it stays editable.

Watching this one was fascinating. It went into Select and Mask and used the Refine Edge Brush around the hair, and it kept switching the view mode to check its own work. At one point it reported that the fine strands were coming through, but that the black preview was showing grey left between the looser locks, so it went back to work on those and the fur before checking the final mask. Right at the end it even renamed the layer, which is maybe something all of us could learn from.

The result is a pretty good job. It picked up a lot of the fine hairs, it's a little smudgy in places, and there's some transparency through the fur. To judge it properly I put a solid colour fill layer behind Simon, a sky blue rather than anything too different from the grey he was photographed against.

The transparency in the fur is easily fixed. Duplicate the layer with its mask (Command or Ctrl + J) and the density builds up, then brush out any areas where the background starts to show through.

Doing the heavy lifting

It isn't perfect, and I don't expect this to be the end point, but as a way of doing the heavy lifting so you can go in and do the fine tuning yourself, it's pretty impressive. The colour correction was very good, the long exposure was really well done, and the hair cutout, the hardest of the three by some distance, was far better than I expected.

This is also the worst and the slowest it's ever going to be, given how quickly all of this is moving. I'll be doing a lot more testing, and one of the things I want to try is whether it can write down the steps it takes and grab screenshots as it goes, so you could almost create your own tutorials from it.

I Let ChatGPT Edit My Photos in Lightroom

The photograph in the video above was edited from start to finish in Lightroom, and I didn't make a single one of the adjustments. ChatGPT did. It used my Lightroom, moved the sliders, made the masks and took the picture from a slightly underexposed raw file to a finished edit, in my style.

ChatGPT has been able to use your computer for a while now. You give it permission and it takes over the mouse and keyboard and works through a job the way you would, opening apps, clicking and typing, and you can see everything it does as it does it and stop it at any point. With the recent update to Astra it's taken a real step up, which got me wondering whether it could actually edit a photograph. I don't mean generate a new picture; I mean sit inside Lightroom and edit a raw file with the same tools I use, and learn to do it the way I would.

Teaching it my style

The first job was getting it to understand how I edit. In Lightroom I set up an album called My Seascape Style with 11 of my images in it. Some still have their adjustments visible, masks included, so it could see exactly what I'd done. Others have the edit baked in, because I'd sent them over to Photoshop for some heavy lifting and brought them back, so there's nothing in the panels to look at, only the finished picture.

I used GPT-6 Astra with the effort level set to high. Day to day I find medium is plenty, but for studying my edits I wanted it to do an extra good job, even though it takes longer.

Then I told it what I wanted, dictated rather than typed because it's quicker. Look at every image in the album, study the adjustments, the masking, the detail and the colour grading, and try to understand not only what I'd done but why. For the images with nothing visible in the panels, study the mood and the atmosphere and the tonality instead, and work out what I'd have done to get there.

As soon as I sent it, ChatGPT asked for permission to use Adobe Lightroom, and I allowed it for that conversation. From then on you can watch it working away on its own, opening images and panels and looking around.

Watching it work

What really interested me was how far it went. It dived into the masking and looked at how I'd masked the rocks in the foreground and the adjustments I'd used to bring them out, which is exactly the kind of fiddly detail I'd half expected it to skip.

It posts updates as it goes, and part way through it had noticed that I recover the highlights and open up the shadows, then bring the depth back with control over the black point and selective detail. It had even picked up that I use one film-style profile for the warmer, pastel pictures and a different one for the cooler surf shots.

The whole thing took 11 minutes and 30 seconds. At the end it said it had reviewed the adjustments across all 11 images, including 27 local masks, saved an image-by-image reference of my style, and, thankfully, left my images unchanged.

Handing it a raw file

For the edit itself I picked an unedited raw file from the Local tab in Lightroom, a seascape that was slightly underexposed, which felt like a fair test. I told it to edit the image in my style using everything available in Lightroom, including the masking, the colour grading and denoise, then sat back with my arms folded to prove I wasn't touching anything.

When it finished, it described what it had done as my pastel seascape treatment: pink and lavender grading, cool water, deep textured rocks and luminous foam. The edit included four separate masks, denoise, selective sharpening, a custom tone curve, subtle grain and a vignette, plus a horizon correction.

What it actually did

Going through the Edit panel, it had chosen a profile and made adjustments right through the Light section. In the curve it had done one of the things I love to do, which is lift the blacks so they aren't a deep zero black, and it did that on a point curve. It used the colour mixer, and in the colour grading it worked on the shadows and midtones as well as the blending and balance. It used denoise too, and it built all those masks.

Hand on heart, that's pretty much how I would have edited it. I wouldn't leave it exactly as it is, I'd go in and tweak a few things, but it got me about 90% of the way there in roughly 10 minutes.

Try it with your own photos

If you'd like to give this a go, the detailed prompt I used is free to download. Head over to the classroom in my free Photography Community on Skool and you'll find it under YouTube download files:

https://www.skool.com/photography-community/about

Where this leaves us

There's been a lot of resistance to letting AI edit our photographs, and for good reason. When it generates the picture for you the quality drops, at the moment anyway, and you can't go back in and adjust what it's done. This is very different; it's using my software to make the same manual adjustments I'd make to my own raw files, at full quality, and every one of them can still be changed.

It's not something I'd do myself, because editing is a big part of how I make a picture mine and I just love the process. However, if you're new to editing and don't know where to start, or you know what you want but not how to get there, this is a bit like having someone who's really good with Lightroom sit next to you and show you what they'd do. You can watch every move it makes and undo any of them, and I think there's real potential in that as a way of learning.