Remembering the events of early January 2024, when Lacock Village was flooded and only a few of us could reach the Village Hall (and fully expecting the same to happen this year) we took the decision to devote this evening to a zoom call from our homes.
The call was hosted by Joe Houghton with a talk entitled Photography in an AI world
He started by discussing the results Leonardo AI had given him using a prompt of "A pensioner photographer taking a photo using a futuristic looking camera of a high-tech urban city".
 |
| "pensioner photographer taking a photo..." as generated by Leonardo AI |
He then showed images generated by CoPilot AI using exactly the same prompt:
 |
| "pensioner photographer taking a photo..." as generated by CoPilot AI |
This prompted a discussion of how biased this particular AI was towards white, male, bearded photographers which is, as Joe explained, a symptom of the images used to train the AI.
By way of comparison, I tried exactly the same prompt in the beta version of Photoshop, which employs Adobe's much vaunted Firefly. There are dozens of options, and choosing some fairly randomly, it came up with these:
 |
| "pensioner photographer taking a photo..." as generated by Firefly via Photoshop |
 |
| "pensioner photographer taking a photo..." as generated by Firefly via Photoshop |
 |
| "pensioner photographer taking a photo..." as generated by Firefly via Photoshop |
The cities aren't particularly futuristic looking, but Firefly's bias doesn't seem to be towards male bearded photographers.
Joe then quoted from an article he wrote for the RPS in June 2023 saying “there's only one way to resolve the issue of AI in photography... ban any image being submitted to any competition shot or processed on a digital camera”.
He then outlined a timeline:
- Between 15,000 and 17,000 years ago: The Lascaux cave paintings appeared showing life, colour and movement
- 1st Century BC: Frescos at Pompeii show realism
- 1130: "Cloudy Mountains" by Mi Youren in China shows a landscape with depth
- 1645: St Antoniuskapel, Utrecht, painted by Pieter Jansz, shows full perspective and converging lines
- 1827: Heliograph on pewter, the first photograph ever taken
- 1859: Twilight by Camille Silvy, a seamless composite of four negatives
- 1998: First AI generated image
- 2022: Jason Allen's AI-generated work wins the digital category with the image below at the Colorado State Fair
 |
| Théâtre_D’opéra_Spatial |
The image was originally generated by Midjourney, it had a further 625 edits in Photoshop. Although there was a great shout-out by "real" photographers at the time, there was nothing in the rules to disallow the submission.
So the question arises as to who cares whether an image is AI generated. Clients don't - as far as they're concerned it's results that count... if they don't have to pay a photographer, then that's even better. So who does care? The answer seems to be only those judging image competitions, of which wildlife is the worst, having the most restrictive rules where zero edits/alterations are allowed.
The LENSA app can create photorealistic images from an uploaded phone image. It has a darker side as it (and other apps) can produce sexualised images.
 |
| An image enhanced using LENSA |
A recent survey revealed that websites allowing users to create explicit content were visited more than 18 million times from Japan,
and Japan is in 3rd place behind India and the USA in terms of website visits.
We then saw:
- A deepfake real-time video of someone morphing into various well-known people
- The facebook page of Emin Kuliyev whose images of people backlit by the setting sun were all AI generated
- The winning image from an Australian competition which was later admitted to be completely AI generated
- The "outing" of a popular Instagram photographer revealed as an "AI fraud"
- The Sony World Photography Award winner who refused the award, admitting his image was AI generated - the competition judges didn't even notice how the hands were deformed.
Joe then discussed how AI is making its presence known more and more in digital cameras themselves. The new Olympus OM cameras will have AI controlled:
- Autofocus
- Subject detection
- Focus stacking
- Exposure bracketing
- Blended high resolution (pixel shifted) images
- Neutral density filters, up to 10 stops
- Live view graduated filters (hard or soft)
In addition, phone manufacturers are lauding the ability of their products to blend images in camera to create composite shots. Camera manufacturers are introducing AI to analyse a scene and automatically adjust exposure, focus, white balance and ISO for an optimal image. Smartphones add features like bokeh, night mode and object recognition.
Websites like My Heritage use AI to enhance and colour, old black and white images:
then, as if that wasn't already enough, the site can use deepfake technology to make ancestors speak.
HEDRA can be used with an uploaded image and a voice track to create a lip-synced video.
Both CANVA and Photoshop can extend images. Topaz Gigapixel can upscale images and add detail that wasn't there in the original. You can design faces with AI.
With those thoughts buzzing around our heads, Joe ran out of time, leaving us to ponder the future.
Our next meeting will be on January 16th.
%20Moth.jpg)
.jpg)
