• Going to Hidden Earth??

    Come and visit the UKC stand and enter our FREE competitions!!

    £100's of fabulous prizes to be won, both on the day and by checking the forum the following week - get your FREE raffle ticket by visiting the UKC stand at Hidden Earth!

    Click here for prize details

Editing The Easy Way.

The images don't look better. They look like an AI has been told to make them look better.

I appreciate the value of it as a nostalgia exercise if they're your photos. I've used AI tools in the past to colourise and even add motion to photos of my parents as children, as a gift for my grandparents.

However the photos it produces look over saturated, over sharpened, and over contrasted. They fill in the gaps with artefacts and textures that stand out to those who have been in caves before.

However, it isn't helpful to make the primary argument against these tools to be their quality, because the quality is obviously going to get better in the future. What sticks out as obvious at the minute might be smoothed out in the future.
 
In a way I think that I have established a point. Its early days where AI editing is concerned and as we know it can only get " better ". We wait to see what the future holds. Any more from Chat I will keep on file. Always remember to save the originals.
 
It can get better at making an image look 'better', but it can never fill in that black space in your original photo accurately (unless it starts searching for other photos of the same passage, of course). The information of what it there simply doesn't exist (at least not in that photo). Fundamentally it's like a painter painting a new version of a photograph. There is artistic licence involved which means the image isn't really 'authentic'; it's not just adjusting what's already there but also wholesale creating elements and making assumptions not supported by the original data.

Doesn't mean it doesn't look 'good' of course, in the way that many truth-inspired fictions look good.
 
I have now experimented myself and can see that it can help considerably with archive shots by correcting serious colour cast and fungal attack. This image was from a Boots colour slide taken a week after the floods of '68 at the entrance of Swildon's Hole. It has done a far better job than my attempt with photoshop and another programme. Colour is correct now.
 

Attachments

  • ChatGPT Image Aug 11, 2026, 02_59_44 PM.jpg
    ChatGPT Image Aug 11, 2026, 02_59_44 PM.jpg
    353.7 KB · Views: 57
I do wish Mr O' Doc would use a hosting service but then he has inspired me to return to the subject primarily in regard to ChatGPT. OK. So how many of you current cave photographers have an archive of their own material dating back to the mid 1960's ? Is it just me and old Doc ? Yes by today’s standards all of our 1960's stuff is compromised in one way or another . Try doing that today with no editing as it was on slide film and now as he says degraded in some way. Photoshop editing which I am sure most of you do is no different than Chat. Its just that Chat thinks for you. I would like to bet a non edited image has not featured on the cover of Descent for donkey's years.
We wanted to re start the North Choke dig beyond Jonathan's Chamber in Withyhill. The Management Committee turned it down. Jonathan's Chamber is special. ( to me at least ). Rightly its now off limits to digging traffic. You might think that its off limits for everyone for ever. Perhaps thats how the Management Committee now see it. The top of The Frozen Deep is off limits for ever. Oh Mr M C I want to take photos in there. Are there not enough done already by photographers of the past ? Why should there be more of the same ? So out come the archive unedited shots from a long while ago and if Chat can make them more " acceptable " why not. Same with my Shatter Helectite images. They aint there any more as they got trodden on in the early 1970's. Doc's archive shots.
Yesterday was a trip into Reservoir Hole for somebody who had not seen The Frozen Deep before. They all said " oh you are not taking photos are you. You have enough already." ( maybe 2,000 ! ) Valid point but these photos were for a keepsake and memory for somebody unlikely to ever go there again. So my remit is not technical perfection its about creating a memory the best way that I can. Apart from him the group were totally unsupportive. And why not they have seen it all before including faffing photographers with a ton of kit ? We did The One Show filming and even I was pissed off .
I wanted the arch. Maybe the widest natural arch anywhere in the UK .They would not allow me time to set it up. The camera compromised horribly. f2 for no depth of field and peripheral lack of focus. 1/8 second with a hand held camera and ISO 800. I show here my slightly Photoshopped version ( haze filter ). And Chats answer to my request " make it better ". I know which one our visitor will like the best.

Original .



ChatGPT.



You might well say horses for courses.
 
I have now experimented myself and can see that it can help considerably with archive shots by correcting serious colour cast and fungal attack

My preferred approach would be to perform deterministic corrections that you can record alongside the corrected image so that it's clear what you've done to "correct" the image. You could ask your globe-burning LLM of choice to prepare one of these for you rather than doing the correction work yourself.

A tool like GMIC would be ideal for this, as you can store .gmic files which contain the transforms you've applied. This looks like:

Code:
my_edit :
  -exposure 0.5,0,0
  -sharpen 80
  -contrast 10
  -saturate 1.2
  -hald_clut lut.png

Most of those are self explanatory. The last one is a CLUTs (colour look-up tables) which is a tiny PNG which you'd also need to archive which maps a colour in your orignal image to a colour in your corrected one.

If you don't like working with command line tools then Darktable has a similar system (.dtstyle), as does Lightroom (.xmp).

You should also be aware that it's very likely that any image "corrected" using an AI tool will contain a fingerprint that makes it clear that you've used AI to prepare it. I would imagine that many organisations, like Descent, will start to check images for these fingerprints and reject them before too long (I certainly would).
 
To test how well LLMs can build these deterministic correction files I prompted Claude (Sonnet 5, "Medium" effort) to prepare one for TOR's photo above. Here's how it did (click to expand):

Original: original.webp

"Corrected": cave_photo_corrected.jpg

Horrendously over-sharpened to my eyes, but because it's a deterministic correction there's no hallucination and we know exactly what it did, which is:
  • White balance: reduces reds and increases greens and blues a bit.
  • Levels (black/white point): clips the extreme shadows/highlights and stretches the image across the full range.
  • Gamma: brightens the shadows and midtones without increasing the highlights, in order to account for the under-exposure.
  • Denoise: to account for the increased shadow brightness.
  • Contrast & saturation: shove these up like all amateur photographers do to make the image more striking.
  • Clarity: I had to google this, but apparently it blurs a copy of the image, subtracts it from the original to get a "detail" layer, then adds a scaled version of that back in. This adds more depth without a global contrast change.
  • Sharpen: adds small-radius unsharp mask.
It didn't use a separate CLUT as it did all its colour correction in the white balance, levels, gamma etc. steps. The resultant .gmic file (full of comments and slop) is <5kb so very easy to archive alongside the "corrected" image. And no AI watermark.
 
Thanks for the memories Chat. My son Jonathan in Jonathan s Chamber a long time ago. A poor original at 174KB re rendered by Chat at 1.29 Mb.
A photo never to be repeated as the chamber is now off limits and Jonathan is in Australia. He was one year old the month I discovered it.

 
Much as we might regret the increasing intrusion of Artificial Intelligence, it nevertheless can be a wonderful tool for certain things. I've been revisiting some of my old caving and family photos from my early days, many of folks who have since left us. I've been using the Photoshop "Partner" AI bolt-ons for editing, and although not all the results are acceptable (it can make people look completely different, avoided by masking off their features), some are spectacularly good, and a few have brought tears to my eyes as I see an old friend or loved one suddenly emerge from the fog and detritus of an old print and become spookily real.

Away from photography, I enjoy building DIY hi-fi, and the Google AI engine has guided me through my incompetent fumblings with amazing results. After a while, you start to almost believe you're communicating with a real person.
 
Seems to work best with damaged slides or severe colour casts. I suspect it is a case of being skilled at using Photoshop to get the same results.
 
A fifteen minute mess around using Chat and Photoshop.

My original set up which was entirely wrong. Top left hidden the flood light which is fine. Bottom right the torch which made the horrible shadow. The stal right over exposed as I did not ask him to turn his light down.



Chat shadow reduction but Photoshop cloning to erase shadow areas. Dont look too hard for the artifacts.



Back to Chat to reduce shadow even more thus creating a different image type altogether.



All simple stuff and I could ask Chat to add more colour back. Where do you want to take editing ?

For this one none of it matters as the subject just wanted a memory of his trip.
 
It's interesting that you refer to 'the horrible shadow'; I quite like dark shadows in cave photographs because . . . well, generally speaking, that's what I see when I'm there.

And you seem to have got an image of the eclipse in the second version (just to the right of the big stal in the middle).🤣
 
Last edited:
Back
Top