Getty Images has partnered with Nvidia to launch its AI text-to-image generating product called Generative AI by Getty Images.
Getty has a huge library of images and videos and says its AI image generator was trained exclusively on content that the company owns the license to. The advantage of this is that users can generate “commercially safe” images without any concern over possible copyright or license infringements.
In its terms and conditions, Getty says that its generative AI “services provide full protection and indemnification to use in a rapidly changing legal and regulatory environment.” That statement is becoming increasingly important as companies navigate the murky copyright issues surrounding generative AI.
Getty Images CEO Craig Peters said, “We’re excited to launch a tool that harnesses the power of generative AI to address our customers’ commercial needs while respecting the intellectual property of creators.”
The tool is based on Nvidia’s Edify model and its Picasso generative AI foundry. The training, optimization, and inference are all done on Nvidia’s DGX Cloud platform.
Images created with Getty’s tool will not be added to its library but will be used to retrain its model. Creators will also be financially compensated by “allocating both a pro-rata share in respect of every file and a share based on traditional licensing revenue,” if their images are used in training.
The image examples in Getty’s promo video look great but you’ll need to hop on the waiting list if you’d like to try out the tool yourself. What will it cost? Getty doesn’t say, except that it will be available on a separate subscription.
To keep it on the safe side of generative AI, you won’t be able to prompt the tool to generate an image of a known person or ask it to imitate the style of a specific artist.
While companies don’t need to worry about legal repercussions from using these images, they still can’t copyright the images they create. So once the image is published it’s not really clear what protections, if any, they have from other people using the image.
If you want to generate images using a model trained on Getty content you could just use Stable Diffusion. Getty is currently suing Stability AI over claims it scraped its content library to train its model. It’s a claim that’s hard to deny when Stable Diffusion generates images like this:
The weird arm isn’t the only awkward part of that image.
As generative models improve it seems that training datasets will eventually become the real differentiator between the different options. Using AI-generated content to train models leads to continuously degrading performance, so we still need some human-generated content.
With tools like DALL-E 3, Adobe Firefly, and now the Getty Images tool, the role of stock photographer will have to adapt. Photographers may need to transition to shooting stock for AI training while seeing less of their content actually published.