Getty Images AI sounds promising but … 16 questions to ask

Getty Images AI supposedly delivers business-friendly, “commercially safe”, best-in-class image generation…

… what do you need to know before you consider it?

Illustration of concept behind Getty AI image generator. A white robot head with large glowing blue scanners for eyes, a mechanical neck, and a mechanical arm growing out from where its ear should be. The hand at the end of the arm is holding a paintbrush.

Getty Images AI: a unique text-to-image offer

Getty Images has jumped on the AI bandwagon with their own text-to-image generator that they call “Generative AI by Getty Images, powered by NVIDIA”. This new AI kid on the block represents a new quality-focused niche in the AI image generation market, along with Adobe’s Firefly.

But Getty Images Generative AI definitely isn’t for everyone. Anyone thinking about exploring it needs to ask themselves some important questions first.

What’s unique about Getty Images AI?

  • Getty trains their AI exclusively on their own extensive photo library. Their customers get Getty’s standard royalty-free license to use images they generate.
  • Customers’ AI-generated images won’t appear in Getty’s stock library.
  • The image creators whose work appears in the Getty library and is used by the AI to generate an image get a cut of the action.
  • Images the Getty AI generates are “commercially safe”.
  • The pricing and business model are certainly unusual.

Why does Getty use their own image library to train Getty Images AI?

Using only the Getty stock images library to train their AI has several advantages. Getty has permission to use their library’s images from the copyright owners. And, Getty has already curated the images in their stock library, so they’re training their AI on high quality content. Those two key points allow Getty to guarantee their AI output is “commercially safe”.

Why does Getty need permission to use images to train Getty’s AI?

Getty Images AI complies with Getty’s legal obligations to content creators. Except in very narrow circumstances (like fair use), everybody needs permission to use copyrighted images. That’s an obligation every image generation AI platform should comply with. Unfortunately, most don’t.

Using images for any purpose without permission is copyright infringement. Violating copyright isn’t just a moral or ethical problem. It also entails legal and financial penalties.

In fact, Getty is suing other AI companies for using images from the Getty image library without permission. As I mentioned in another blog post, the outcome from those legal battles will likely cause significant damage to the copyright infringers. Indeed, that unapproved misuse of copyrighted images could even put them out of business, if they can’t separate out copyright infringing images from their AI training and knowledge.

What does Getty mean when they say their AI is “commercially safe”?

Getty say that because they train their AI only on images in the Getty library, “you can rest assured that the images you generate, and license, are backed by our uncapped indemnification.” So Getty will help you defend yourself and pay any penalties that might arise if anyone challenges your use of an image generated by the Getty AI. That’s a bold claim. It’s also a good selling point for business customers who appreciate insurance against potential copyright or licensing issues. Getty is so confident of its legal position that it will put its money where its mouth is.

Why is curated content important for Getty’s AI?

Training the Getty Images AI on curated content—carefully selected and categorised by a human—increases the likelihood of good quality results. It’s the opposite of “garbage in, garbage out”. That labour-intensive curation at least partially justifies Getty’s prices.

Is Getty Images AI the only AI trained on curated content?

No, Getty’s AI isn’t the only text-to-image AI trained on curated content. Adobe’s Firefly AI is also trained on curated content, to ensure good quality and socially responsible output.

But Getty and Adobe are the exceptions. As far as I know, all other text-to-image generators (Midjourney, DALL-E, Stable Diffusion, etc) use automated processes to scrape images off the Internet and their own algorithms to categorise and use them. That automated (careless) approach can lead to very bad results—like racist, sexist, and stereotypical output.

Who can use images I create with Getty’s AI?

Getty won’t add customer-created images to their stock image library. Along with their standard royalty-free licensing, that means customers will control any images they create with the AI. However, the devil is in the details about ownership. Getty’s terms and conditions of use will clarify who can use their AI-generated images.

Will Getty use images customers create with Getty Images AI?

Yes, they’ll continue to train their AI using images customers create with the AI. That may be problematic for a couple of reasons.

Training the AI on its own output creates a feedback loop. Potentially, that could distort future output. One hopes that Getty’s developers have taken that into account. Even so, it would be an important question for any customer to ask them.

And it somewhat contradicts Getty’s commitment that they won’t add customer-generated AI output to their stock library. If they use that output the same way as they use their stock library contents, essentially they’re adding customers’ content to the mix.

Should photographers be worried about Getty Images AI?

Absolutely not! In fact, they should welcome Getty’s commitment to compensate the creators of images used in Getty AI output. How that will work in practice remains to be seen. But as of September 2023, Getty is the first and apparently only AI vendor to compensate the people who create the images their AI uses to make the vendor money.

What does NVIDIA have to do with Getty Images AI?

Getty says “Our technology partner, NVIDIA, and their proven computing infrastructure provides speed and stability to generate incredible visuals even at commercial scale.” So NVIDIA are providing the technology (hardware and software) that will enable Getty to deliver image generation quickly to their customers.

How much does Getty Images AI cost?

A LOT! But exactly how much? Apparently it’s a secret.

The minimum Getty price for a single low-resolution image from their stock library is EUR150. Large images can cost upwards of EUR400 each depending on whether you buy in bulk.

That gives you an idea of how pricey Getty AI images might be. But when you look for pricing on Getty Images AI output, they ask you to get in touch to negotiate.

In other words, if you have to ask, you probably can’t afford it.

How does Getty Images AI pricing compare to other generative AI image solutions?

Getty’s pricing appears to be on a completely other level. Just using their standard pricing, at a minimum a low-resolution stock image costs USD150.

If you’re a Midjourney subscriber, low-resolution images can cost anywhere from USD0.015 to USD0.05.

So Getty stock images can be up to 10,000 times more expensive than Midjourney images.

It wouldn’t be surprising if a business subscription for Getty Images AI would cost a lot less per image than Midjourney or others. But Getty’s pricing is almost certainly premium level. So their output will have to be amazingly great to justify the premium pricing: no or very little human time and effort spent on pre-generation prompt engineering and post-generation editing required.

Can I try Getty’s AI before I buy?

Maybe. But apparently you have to ask nicely first, and only if you’re a business. When you click on the Demo button on their site you open a form to request that someone from their sales team will “reach out”.

Company name and company type are required fields on that demo request form. There’s no option for individuals.

Why don’t Getty allow unsupervised demos of Getty Images AI?

Good question. My guess is that they want to control any demos to ensure the best presentation, and avoid any unhelpful output. That’s a red flag, in my eyes. If a software vendor wants to control demos, that might mean they have misgivings about how the software performs: they know that the software does some things well, but glitches on others.

Getty would presumably allow a prospective business customer to have a trial period to test the AI in the business’ own environment. They’d also need to agree an expected level of service. They’d better, at those prices!). But you can test Adobe’s Firefly and other paid text-to-image generators online for free. Getty is very much in the minority by restricting who can see their AI in action.

Does Getty’s AI make business sense?

Getty Images Generative AI might be a good fit for your business if you can afford it. The premium pricing and business-only orientation imply that Getty promises quality output, quickly. However, any return on investment calculation has to include the time humans may need to spend on generating versions, prompt engineering, and post-generation polishing.

Is Getty Images AI right for you?

That depends on who you are. If you’re a business user—agency, media, marketing, PR—then maybe Getty’s AI might make business sense. If you’re an individual, then no, Getty Images AI is definitely not for you. Why? Price. Oh, and Getty doesn’t want to waste their time on individuals. They’ll only talk to businesses about their image generation AI.

Screenshot of Getty Images AI pricing structure
Screenshot of Getty Images AI pricing structure
Screenshot of Getty Images contact sales about plans and pricing online form. Note that company name, company type, and job title are all required fields.
Screenshot of Getty Images contact sales about plans and pricing online form – note that company name, company type, and job title are all required fields

Do you have any questions or thoughts you want to share? Comment below!


Comments

Leave a Reply

Your email address will not be published. Required fields are marked *