7 things I learned from using AI-generated images for the first time
World-renowned media repository Getty Images made an AI image generator. It has complemented its massive library with impactful generative AI technology for over a year. We sat down with Grant Farhall and Bill Bon, Getty's Chief Product Officer and Director of Content Operations, for the inside scoop on how the company is leveraging its industry presence to streamline content production while protecting individual creators' rights.
Proper licensing drives Getty's push to "save consumers time, money, and risk, while creating powerful visuals," with a different kind of AI image generator, as Mr. Farhall explained. "We find it vital to respect the rights of IP holders and creators. The key element is making a responsible service that is commercially safe." But what exactly is that?
Related 7 things you should never share with AI chatbots
Keep these details private
Posts 2 Four aspects of commercially safe AI What's different about Getty's image generator Source: Getty
The tool's focus on IP rights doesn't noticeably limit its effectiveness or variety.
"First," Mr. Farhall began, "the model was trained on transparent, clean data sets from our creative library. Nothing from editorial libraries, web scrapers, public domain sources, or generative AI, because we don't allow synthetic images into our libraries. We have permission to use all the content, and the tool can't generate intellectual property like copyright-protected characters."
Mr. Bon (who demonstrated the tool thoroughly) expounded, "It can't create protected IP, not just on legal terms, but also because the data contains no commercial content. We require creators to remove logos, trade dress, and other IP before submitting images. Our responsibly trained, ethical tool also verifies rights ownership when users upload reference images."
The fact that some AI services will replicate a named artist's style, when the artist hasn't given permission or been compensated, that's offensive. Anyone who cares about the creative community's long-term viability should care about that.
-- Grant Farhall, Getty CPO
Mr. Farhall continued, "Second, we provide a standard commercial, royalty-free license. Customers can use our generated images with full permissions. Third, based on our confidence in the model, we offer legal protection for any imagery our AI generates. We know it won't cause any problems, so we guarantee indemnification."
Rounding out the pillars of responsible AI, Mr. Farhall finished, "We return a portion of revenue to creators responsible for the training data, thereby sustaining that ecosystem. That creator economy underpins our strategy, and we think the broader world should care, too. We need original thought, we need imagination to stand, and we need rights respected. Because we can't forego amazing, human-made visuals, and models depend on new content moving forward."
What Getty AI generation can do With a focus on a user-friendly interface
This time, there's a clear separation. This image was upscaled (somewhat poorly) by AI.
You can start with a simple prompt or use Getty's prompt builder to specify subjects, actions, surroundings, and visual aesthetics. If a prompt includes protected content, like "a Nike shoe," generation won't occur until it's fixed.
Clearly labeled parameters like photo and illustration mode, aspect ratio, lens aesthetics, and mood filters line the screen. Prompts only generate generic versions of identifiable objects like cars and celebrities. "We have a rich history of pre-shot images, and a legal team ensuring customers don't generate something that gets them in hot water," Mr. Bon explained.
Each prompt returns four images. Uploading reference images (if you own the rights) can influence color or composition. One click retrieves new, subtler variations of an almost-there image. Locking the seed prompt to add stipulations without affecting the original parameters also sees frequent use, Mr. Bon explained.
Related 8 ways AI is changing the game for artists, musicians, and other creators
Creativity is no longer a solo human gig, AI is crashing the party
Posts
Getty's inpainting and outpainting tools, Refine and Extend, simplify crafting bespoke backgrounds and adjusting individual elements. In the demonstration, a brush tool swipe and a few words spawned a realistic scarf on a woman's neck.
Close
The simple process of uploading product imagery for perfect placement in Getty's responsibly generated AI images.
Despite almost no learning curve, the interface provides the tools many professionals need. It's simple to tailor color profiles, create detailed landscapes, and test custom product placement. Mr. Farhall tied the wide-ranging potential to the company serving "global agencies and large enterprise brands through Getty Images, down to small businesses through iStock."
How Getty's AI generation stands on its own
When asked, Mr. Farhall explained the tool isn't integrated within Getty Images' standard search portal. "Right now, you're either searching the catalog or generating images," Mr. Farhall said. "Your instinct is sound, though, considering how this could bring search and generation together for a unified content discovery process."
In a world of on-demand images where you can create anything, what do you create? We call that the 'blank page problem,' and our creative library makes a great starting point.
Mr. Farhall eagerly explained how Getty developed the software. "It's all in-house, in partnership with Nvidia. Some modification capabilities were built with Bria, our other AI partner. Bria also uses commercially safe imagery -- nothing scraped, synthetic, or public domain. Getty's tool is all home-grown. We don't use any existing products, like Stable Diffusion or open source models." It's hard to argue with that kind of vertical integration.
Why real photos aren't going anywhere AI is only one part of the future Source: Getty
Several examples of what Getty's custom image modification can do to subjects, backgrounds, and color profiles.
Farhall doesn't anticipate generative AI replacing real photography. "Feedback shows customers experimenting with how to gain the biggest AI efficiency boost. Some customers use AI early in the process, for brainstorming and quick ideation, but not final content production, for two reasons. Some worry about legal risks, which we've solved.
"Customers also observe brands using AI, then seeing backlash. Brands that foster close relationships with audiences, grounded in authenticity, show concern and hesitation." Actual photos still matter, Mr. Farhall reassured, "because they're real people in real settings doing real things."
If dogs with Pokémon accessories are wrong, I don't want to be right.
"We don't envision AI displacing pre-shot images. We ask, 'How do these work side by side?' So far, much of the tool's adoption includes modifying pre-shot images from our catalog to suit specific needs."
Mr. Farhall's ground-up consideration: "In a world of on-demand images where you can create anything," he asked, "What do you create? We call that the 'blank page problem,' and our creative library makes a great starting point."
Next, a realistic observation. "It's still more efficient to search the library and find hundreds of images almost instantly. AI generation is fast, ours especially, but five seconds to generate, then five more for each change, adds up. Comparatively, AI isn't necessarily more efficient. For us, it's about how these work together to give customers more creative control."
Do individual creators have a future? The team at Getty thinks so Source: Android Police
Clearly, nothing can replace human-made art.
Finally, I asked if creative careers can withstand AI's onslaught. Mr. Farhall answered with aplomb. "There is absolutely a path forward. We built this to prove it could be done. We want to disprove the notion that enhancing AI capabilities requires broad web-scraping. It's happening, again, with video. We want to exemplify building on content that respects creators' rights.
"The biggest threat is AI companies choosing to trample on those rights. They choose not to compensate creators. The fact that some AI services will replicate a named artist's style, when the artist hasn't given permission or been compensated, that's offensive. Anyone who cares about the creative community's long-term viability should care about that.
"Getty stands against that, strongly advocating AI development that respects ownership, and allows creators to thrive. It's vital to us, and more companies should agree. Consumers need to hold all of us to a high standard.
Related 7 things I learned from using AI-generated images for the first time
What you may not know about AI image generators
Posts
"Consider transparent data sets -- something simple and easy. A customer should be able to ask their vendor, 'How is your model trained?' and get a detailed, accurate response. It isn't sufficient for a company to say it doesn't know or can't share that. We can and do, because we know every asset we use." While some parts of the industry increase their focus on open source development, Getty is utilizing its market strength to push for fairness moving forward.
Who's responsible for AI?
Ultimately (and fairly), Getty places some responsibility on consumers, as AI takes over increasing roles in the industry. As Mr. Farhall closed, "Customers must demand commercial safety, clean models, and good tools. And we should all be held to those standards."
