Fake war images and the death of stock photography
On stock images, conflict footage and job loss
Above is an image from Adobe Stock, a portal for stock images. If you’ve ever needed a cheap, or even free, image for a work project you will be familiar with sites like Adobe Stock, Shutterstock and Unsplash, as well as its more upmarket cousin Getty Images. The image is made with generative AI. Adobe has been dealing with the fallout since Australian news site Crikey reported AI images portraying the war in Israel-Gaza were available on the portal and were being used by some online outlets without mention of its gen AI origins.
Adobe released a statement that came nowhere near an apology or a commitment to remove the images:
“Adobe Stock is a marketplace that requires all generative AI content to be labeled as such when submitted for licensing. These specific images were labeled as generative AI when they were both submitted and made available for license in line with these requirements. We believe it’s important for customers to know what Adobe Stock images were created using generative AI tools.
Adobe is committed to fighting misinformation, and via the Content Authenticity Initiative, we are working with publishers, camera manufacturers and other stakeholders to advance the adoption of Content Credentials, including in our own products. Content Credentials allows people to see vital context about how a piece of digital content was captured, created or edited including whether AI tools were used in the creation or editing of the digital content.”
I did a quick search this morning, it resulted in this image from an uploader labeled “Israel-Hamas war, military operations near the Gaza Strip. Powerful Israel army modern main battle Merkava Mk 4 tank. 3D heavy military vehicle IDF Merkava tank weapon, Israeli-Palestinian conflict”. It’s from an account that otherwise has uploaded bog-standard stock images that all appear AI-generated. There is no labeling that the image is generative AI. And this isn’t just an Adobe problem, the same image is available from the same ‘studio’ on Shutterstock.
The uploader of the image up top is a prolific uploader. The uploads are all labeled as gen AI and are within Adobe Stock terms. Most are AI versions of standard stock images, though AI images depicting the Ukranian conflict are also in their portfolio. This single uploader has added at least 10,000 images to the platform, which appears to be common for most AI stock image uploaders.
Those two examples are shocking because they could pass as genuine images to the untrained eye. The portal is also full of more obviously fake content of explosions or war-affected children, often from uploaders who add AI war footage alongside AI stock images of cats wearing headphones. If one image doesn’t sell, don’t worry, there’s another completely unrelated image coming in 20 minutes.
There are a few things going on here:
1) a cottage industry is profiteering off war imagery and Adobe’s statement is a shoulder shrug
2) generative AI content is not always properly labeled on the platform
3) many, many people are churning out AI content for stock image sites
The first two points are, hopefully, getting well covered in the mainstream this week. The third isn’t exactly breaking news, but it’s moving us towards a different way we have the world presented to us. News photographers are probably safe from this particular threat, if not safe from multiple other news industry threats. There’s still, thankfully, a value put on bearing witness. And the high-end campaigns are safe. Even when AI image generation is indistinguishable from the real thing there will be value in proving the real thing, on being able to point to a real-life production process.
But what about the stock image raison d’etre, the necessary images that no one really cares about? If your bank used this smiling AI lady in promotional material for a new type of mortgage, would you notice? If you noticed, would you care? I would likely care more if there was no image, it would look wrong, half-assed. That AI lady though? I’m not sure I’ve ever given bank marketing material enough attention that I would notice [cue joke about noticing things from my wife as she copy-edits this].
The slow creep of AI images into our lives will get headlines when it’s something horrific, like profiting off death, or when an artist you love pushes back against their work appearing in datasets. But the disappearance of everyday stuff, the stock image of a toddler tantrum or of women laughing with salad, will barely be noticed. The stuff you barely register when it illustrates a leaflet or an ad or an article, but which involved a photographer and models and a studio getting paid actual money. That’s where people are feeling the squeeze. It’s hard to get up in arms about that when it’s so slow. Slow but relentless.
Of the many, many, uploads of generative AI stock images that appeared in searches today, some were from actual professional photographers. Opening up another revenue stream in an if-you-can’t-beat-them move that’s hard to fault. In five years will all stock images be AI? Will it seem crazy to a younger generation that people paid hundreds for a photo of a model pretending to laugh with her pretend children? And will those currently earning from AI stock photos already be out of a job? Even now it feels like most of these stock AI images could be easily replicated through the simple measure of having an intern who’s into AI.
There will be dramatic moments over the coming years when entire industries rise up against the intrusion of AI into their profession. News cycles where the future of labor laws are battled over. But the sudden death of the stock photographer seems to already be happening and a mundane creative job could be disappearing with a whimper.
Small Bits #1: New weird internet thing dropped
Tiktok failed to load.Enable 3rd party cookies or use another browser
This one is Very Online and another example of the previously discussed ways AI is essentially just changing shitposting a little. It’s an AI version of Homer Simpson/Dan Castellaneta’s voice singing 1996 dance classic Born Slippy and has done numbers on TikTok and on X. It was made with Voicify AI, it’s very dumb and surprisingly catchy.
Small bits #2: Custom GPT but it’s an animal just dropped
OpenAI had its first dev day this week and announced the rollout of tailored ChatGPTs. This allows Plus and Enterprise users to create their own customized GPTs. Next week I’ll be doing a deeper dive on what this looks like in practice, but on first use, it feels like a more marketing-friendly version of the existing custom instructions options.
Small bits #3: Cool and very funny chat tool just dropped
“Grok is an AI modeled after the Hitchhiker’s Guide to the Galaxy, so intended to answer almost anything and, far harder, even suggest what questions to ask!
Grok is designed to answer questions with a bit of wit and has a rebellious streak, so please don’t use it if you hate humor!”
That’s a direct quote from the release of Elon Musk’s new AI chat tool. Musk also posted on X that the tool is “based & loves sarcasm. I have no idea who could have guided it this way”. In a crowded field of chat tools, we now have an option for people who unironically like nu-metal and still argue about the female Ghostbusters reboot. So that’s nice.