Photography in the Age of AI, with Stephen Shankland

02/29/2024Link0

How much can you edit a photo before it stops becoming true? That’s the question CNET tech reporter Stephen Shankland recently asked in the opening lines of his story, How Close is that Photo to the Truth: What to Know in the Age of AI.

The article examines digital photography and advanced smartphone image processing in the era of AI. It reaches beyond the polarizing visual minefield of generative AI by delving into aspects of this technology that’s been quietly pre-baked into almost every camera on the market these days.

The sophisticated processing under the hood of your digital camera is our jumping-off point for a wide-ranging discussion with Shankland that touches on many aspects of the digital workflow, before scaling the slippery slopes of generative AI.

A few of the many points we cover include: Comparing the three primary generative AI platforms and discussing their differences, an assessment of AI manipulations and deep fakes, the ways in which a proliferation of camera phones can serve as a buttress against fakery, and the social contract inherent in weighing the veracity of an image.

Today’s AI landscape seems to be morphing by the minute, a reality that’s reflected here with bonus content! Barely a week after our original discussion, Open AI’s new text-to-video application, Sora, was released to a tidal wave of interest, so we got Shankland back on mic. Stay to the end to hear our first impressions of this new technology and listen closely to discover how an AI bot got the last word in our chat.

Guest: Stephen Shankland

Above photograph © Allan Weitz, https://www.allanweitzdesign.com

Sponsored by

Original Source: Working with photographs from his personal library, Weitz cherry-picked images featuring color palates, textures, and forms to serve as the ‘DNA’ of prompt-inspired AI camera illustrations. The starting point for his first imaginary camera is the steering system of a 1930s Hackercraft mahogany runabout. © Allan Weitz
Original Source Working with photographs from his personal library, Weitz cherry-picked images featuring color palates, textures, and forms to serve as the ‘DNA’ of prompt-inspired AI camera illustrations. The starting point for his first imaginary camera is the steering system of a 1930s Hackercraft mahogany runabout. © Allan Weitz
AI Result: The resulting promptograph is a Victorian-style SLR with hints of wood, leather, and brass, the basis of which can be traced to the original source photograph. © Allan Weitz
AI Result: The resulting promptograph is a Victorian-style SLR with hints of wood, leather, and brass, the basis of which can be traced to the original source photograph. © Allan Weitz
Original Source: A stylized photograph of a toy taxi is the basis of a camera produced by using the term ‘mobile photography’ as a prompt. © Allan Weitz
Original Source: A stylized photograph of a toy taxi is the basis of a camera produced by using the term ‘mobile photography’ as a prompt. © Allan Weitz
AI Result: The ‘mobile photography’ prompt generated some wonderfully silly cameras on wheels, treads, and other rolling devices. The selected result has a definite military feel about it. © Allan Weitz
AI Result: The ‘mobile photography’ prompt generated some wonderfully silly cameras on wheels, treads, and other rolling devices. The selected result has a definite military feel about it. © Allan Weitz
Original Source: Weitz used a sunset photograph of a lighthouse as source for more than a dozen interpretations featuring cameras floating on water or riding waves. © Allan Weitz
Original Source: Weitz used a sunset photograph of a lighthouse as source for more than a dozen interpretations featuring cameras floating on water or riding waves. © Allan Weitz
AI Result: Weitz’s chosen result was one of several he truly liked. © Allan Weitz
AI Result: Weitz’s chosen result was one of several he truly liked. © Allan Weitz
Original Source: Weitz paired a photograph of a tool shed in an apple orchard with prompts featuring camera terminologies such as ‘single lens reflex’ and ‘field camera.’ © Allan Weitz
Original Source: Weitz paired a photograph of a tool shed in an apple orchard with prompts featuring camera terminologies such as ‘single lens reflex’ and ‘field camera.’ © Allan Weitz
AI Result: How much does Weitz like the results? These two images now hang side-by-side in his home, and they make him smile every time he looks at them. © Allan Weitz
AI Result: How much does Weitz like the results? These two images now hang side-by-side in his home, and they make him smile every time he looks at them. © Allan Weitz
Topaz Photo AI software can dramatically reduce noise in images shot at high ISO. It generates its own detail in some parts of the image to try and improve realism. © Stephen Shankland/CNET
Topaz Photo AI software can dramatically reduce noise in images shot at high ISO. It generates its own detail in some parts of the image to try and improve realism. © Stephen Shankland/CNET
When viewing portraits on the small screen of a phone, it can be tough to distinguish computationally generated bokeh from the real background blur available with traditional cameras and higher-end lenses. © Stephen Shankland/CNET
When viewing portraits on the small screen of a phone, it can be tough to distinguish computationally generated bokeh from the real background blur available with traditional cameras and higher-end lenses. © Stephen Shankland/CNET
Samsung Galaxy phones recognize when you're taking a photo of the moon. Even with blurry shots, such as at left, heavy AI processing is applied that can fabricate moon-like texture. © Stephen Shankland/CNET
Samsung Galaxy phones recognize when you're taking a photo of the moon. Even with blurry shots, such as at left, heavy AI processing is applied that can fabricate moon-like texture. © Stephen Shankland/CNET
Stephen Shankland headshot, photo © James Martin/CNET
Stephen Shankland headshot, photo © James Martin/CNET
Previous Pause Next
Original Source: Working with photographs from his personal library, Weitz cherry-picked images featuring color palates, textures, and forms to serve as the ‘DNA’ of prompt-inspired AI camera illustrations. The starting point for his first imaginary camera is the steering system of a 1930s Hackercraft mahogany runabout. © Allan Weitz
AI Result: The resulting promptograph is a Victorian-style SLR with hints of wood, leather, and brass, the basis of which can be traced to the original source photograph. © Allan Weitz
Original Source: A stylized photograph of a toy taxi is the basis of a camera produced by using the term ‘mobile photography’ as a prompt. © Allan Weitz
AI Result: The ‘mobile photography’ prompt generated some wonderfully silly cameras on wheels, treads, and other rolling devices. The selected result has a definite military feel about it. © Allan Weitz
Original Source: Weitz used a sunset photograph of a lighthouse as source for more than a dozen interpretations featuring cameras floating on water or riding waves. © Allan Weitz
AI Result: Weitz’s chosen result was one of several he truly liked. © Allan Weitz
Original Source: Weitz paired a photograph of a tool shed in an apple orchard with prompts featuring camera terminologies such as ‘single lens reflex’ and ‘field camera.’ © Allan Weitz
AI Result: How much does Weitz like the results? These two images now hang side-by-side in his home, and they make him smile every time he looks at them. © Allan Weitz
Topaz Photo AI software can dramatically reduce noise in images shot at high ISO. It generates its own detail in some parts of the image to try and improve realism. © Stephen Shankland/CNET
When viewing portraits on the small screen of a phone, it can be tough to distinguish computationally generated bokeh from the real background blur available with traditional cameras and higher-end lenses. © Stephen Shankland/CNET
Samsung Galaxy phones recognize when you're taking a photo of the moon. Even with blurry shots, such as at left, heavy AI processing is applied that can fabricate moon-like texture. © Stephen Shankland/CNET
Stephen Shankland headshot, photo © James Martin/CNET

Episode Timeline:

  • 2:22: How much can a photo be edited before it stops “becoming” true? Plus, the digital processing that goes on under the hood of your digital camera.
  • 7:06: The sophisticated processing in your camera phone and how the resulting images compare to pictures made with a 35mm digital camera.
  • 13:02: How much digital editing is too much and what’s the least number of image adjustments possible before a photograph stops “being true.”
  • 18:22: The matter of generative AI manipulations and deep fakes, the democratization of altering images, and how the proliferation of camera phones can serve as a buttress against fakery.
  • 23:24: Comparing the three big generative AI platforms Shankland has worked with—Open AI’s Dall-E, Google’s ImageFX, and Adobe’s Firefly—and discussing how they differ, plus Allan Weitz’s impressions about working with Adobe Firefly, and how much of an AI-generated image is truly one’s own.
  • 31:58: Prompt engineering, the bias of training data, the role of having fun when assessing the creative aspects of generative AI, and factoring in a social contract when gauging the veracity of an image.
  • 40:22: Episode Break
  • 41:30: The potential for career opportunities in prompt engineering, new educational programs to arise from these new technologies, plus reasons why illustration is the creative area most threatened by AI.
  • 48:27: The democratization of creative tasks due to computer technology, and the value of having a unique style or vision to creative success, plus the advantages of AI for stylistic purposes.
  • 52:08: Ethical considerations, intellectual property rights, and copyright concerns in relation to AI generation.
  • 57:03: In-camera authentication, content credentialing, and following the provenance of an image to be assured of its trustworthiness, plus whether this technology will ever show up in camera phones.
  • 1:04:24: Episode bonus: Shankland’s first impressions of Open AI’s new text-to-video application, Sora.

Guest Bio:

Stephen Shankland has covered technology, computing, and digital imaging as a principal writer and reporter for CNET, since 1998. He is also a professional photographer who is particularly intrigued by new trends in AI. Shankland stumbled into journalism as a fledgling science reporter covering the Los Alamos National Laboratory. His first, and biggest, scoop was about radioactive kitty litter discovered at the town dump.

Stay Connected:

Stephen Shankland’s Instagram: https://www.instagram.com/stshank/

Stephen Shankland’s Twitter: https://twitter.com/stshank/

Stephen Shankland’s Facebook: https://www.facebook.com/people/Stephen-Shankland/

Stephen Shankland’s Flickr: https://www.flickr.com/photos/shankrad/

Stephen Shankland’s MuckRack: https://muckrack.com/stshank

Stephen Shankland’s CNET profile: https://www.cnet.com/profiles/shankland/

Stephen Shankland’s CNET article on AI: https://www.cnet.com/tech/

Open AI’s Chat GPT: https://chat.openai.com/

Open AI’s Dall-E: https://openai.com/dall-e-2

Open AI’s Sora: https://openai.com/sora

Host: Allan Weitz

Senior Creative Producer: Jill Waterman

Senior Producer: Mike Weinstein

Executive Producer: Shawn C. Steiner

0 Comments