Where some time ago the new big thing was 360° photos, VR, and 3D visuals, in 2022 it’s all about AI-generated images: pictures that are entirely created by artificial intelligence software.
The latest visual technology lets anyone create incredibly realistic photos that were not shot with a camera or technically flawless paintings that were not painted by anyone but rather computer-generated from text descriptions.
As an entirely new genre of imagery, AI-generated art is gaining traction in the stock media industry, posing new challenges and opening gaps for new software tools to fill.
Today, we’ll have a look at what’s the state of things in the field of AI photos and AI-generated artwork. It’s exciting!
The novel thing about AI-generated images is that they introduce a completely new concept in visual creation: synthetic media, aka generating images.
Synthetic images are those created not from a camera and not from a manned digital illustration/painting program, but from intelligent text-to-image creation software.
Instead of capturing or drawing an image, you are introducing a brief description of a vision, and AI is creating a visual representation of that vision from the text prompt, from scratch.
vAIsual, a pioneer company in this field that is excelling in AI-generated imagery, calls this technology an “algorithmic camera”, and it’s a very accurate description.
Note: this technology is not to be mixed with the AI visual similarity technology that makes for great reverse image search tools, for example. AI generative models do not compare or match images to one another but create new images.
The core value of AI photos is in the potential it opens for creatives. You could create a perfect oil painting, even if you don’t know how to hold a brush. You could create realistic images of the pyramids of Giza, even if you never left your small town in the US.
More AI Tools for Visual Creatives
Besides that AI art app, there is a growing number of AI photo tools out there that can truly shift your creative process.
VanceAI, for example, is a platform packed with AI-powered features that are a great help in graphic design: image enhancing, portrait retouching, background removal, automatic cartoonization, and so much more can be done using their software. It's quite a well-rounded offer.
And the creative resources platform Picsart is diving deep into the AI field as well, as they recently launched their own AI image generator, built within their image editing application, that lets you generate images from text prompts, and further edit them using AI-driven photo editing features. Very cool.
There are also now several free AI image generators available online.
We know you were wondering when we will mention DALL-E, the AI image generator from Elon Musk's OpenAI consortium that lets you generate images from text descriptions. A tool that has recently taken more relevance in this space, as it opened to the public (via invite) and with full commercial use rights.
Much like the other pioneers we mention here, OpenAI developed DALL-E as an image generation system, based on their proprietary CLIP technology, which used a neural network to extract comprehensive descriptions from images. The original DALL E had relative effectiveness to generate images from text (combining objects in collage-style images, for example.) but it was low-quality in terms of pixel count.
DALL-E 2 –the latest version of the software– can generate very accurate images of higher quality, from descriptions using natural language. Be they photorealistic, illustrations, or painting-styled, the software can come up with all kinds of crazy and funny images, like an astronaut riding a horse, a bowl of soup that is a portal to a new dimension, and so on. It can also make edits to images using written instructions only, as well as generate variations from an original image.
Right now, the tool brought on board one million users from its waitlist and gives them an allotment of free credits to generate, edit or create variations of images, as well as the chance to buy more credits if needed. The core value, though, is in the fact all images created with DALL-E come with full usage rights, including the rights to reprint, sell, and merchandise the content. In layman's terms, it lets users make money from their generated visuals.
Of course, this is not the only player in the field of AI image generators. Several other apps are equally popular amongst artists right now, such as Stable Diffusion and Midjourney, which also focus on AI image generation and creating imagery from text descriptions.
Shutterstock Generate: A Legally-Backed AI Image Generator Powered by OpenAI
In a more recent (and exciting!) development, Shutterstock launched a new AI image generator branded Shutterstock Generate in early 2023. This tool results from a partnership between the agency and OpenAI that brings Dall-E functionality onto the agency's platform.
Generate is a text-to-image generator based on the Dall-E model but trained exclusively with Shutterstock content, which adds transparency to the whole process in terms of legalities. Images generated with this app are licensed by Shutterstock when you download them and are fully editable with preset filters and the whole Shutterstock Creative Flow toolkit. And also, the agency created a Contributor Fund that compensates its artists for using their work and metadata in training datasets and the resulting AI-generated images.
You can read all about it in our full coverage of Shutterstock Generate.
Adobe's AI Generative Model Coming to Photoshop and Adobe Express
During Adobe Max 2022 conference, Adobe announced that its proprietary AI generative model would soon land on two of its most popular image editing products, Photoshop and Adobe Express.
Using text-to-image technology, this feature will let users generate their own custom images from a text prompt, and then edit them to their liking using Photoshop's rich editing functions, as well as summon the ideal design templates on Express based on a text description of the design's purpose.
You see, while photos of real people are an all-time best-seller, they are also considerably more difficult to create. You have to have models –professional or not–, and model releases. Licensing photos of people involves respecting sensitive use clauses, too. And even then, no model release is ironclad, and image misuse can still happen.
Enter AI photos of people. These are, simply put, fake people photos: realistic photos of people that aren’t real and don’t exist in the world.
vAIsual, a company specializing in licensable, synthetic stock media –the brainchild of stock media legends Michael Osterrieder and Nico Menijes, together with business industry veterans Mark Milstein and Istvan Novak– is truly ahead of the game as earlier this year they launched their collection of AI-generated faces: generate images and headshots of people that aren’t real. They achieve this by using GAN (generative adversarial network) and text-to-image tech, and their own training data.
Not only that, but they also released their datasets for you to generate your own, high-quality fake faces images. This solves not only the issue of hiring models but also gives you complete control over how the new faces in your images look (age, gender, ethnicity, emotional expression, hairstyle, everything to your specs). Not to mention, it speeds up the production times.
And equally valuable, realistic faces of fake people eliminate the need for model releases, concerns about sensitive use, and limitations regarding how these images can be used.
Furthermore, vAIsual has partnered with PantherMedia agency to bring their unique library of legally safe, AI headshots of people into Smarterpix –PantherMedia’s free photo site–, and they’re already working full-speed to make full-body AI people photos happen.
More AI People and Fake Face Generator Services
vAIsual isn't the only company realizing the value of fake people photos. There are a few other players developing similar services.
Generated.Photos, for example, started with synthetic images, then added a collection of fake faces, and later included a face generator where you can create your custom “fake face”. Right now, they have Generated Humans, a whole dataset with over 100,000 full-body, fake people photos ready to use.
One interesting development in the industry regarding AI-generated images is how the legal aspect of image licensing is changing to cover and regulate AI photos.
Biometric Data: Authorizations and Protection
AI has to be trained in a neural network before it can successfully turn text input into images. The way most companies are training machine learning algorithms is through real-life photos, including those of existing images of humans, most specifically their facial features – what is called biometric data.
This brings up the issue of biometric data security, which has been of increasing relevance ever since the first AI-generated images reached mass media –think of Deepfake, and the concerns several celebrities expressed over software that created realistic, animated molds of their faces that could be used in potentially any kind of way. While not as famous, lots of other people have brought up their objection to having their facial features used to train AI software, without their consent.
Again, we have to mention vAIsual’s innovative approach. They pitch themselves as the “clean data guys”, and they truly are: their AI people images are created on the base of neural networks trained with their own, in-house shot portraits. Furthermore, all real-life models featured in those portraits signed biometrical release forms, and vAIsual owns all copyright to the synthetic images they produce. This means both their datasets –which you can license– and the AI generated images created with them, are legally clean and safely licensable.
Not too behind, the household name in stock photography Getty Images has taken steps towards the future, too, by recently updating their model release to include AI, machine learning, and biometric data. With their new Enhanced Model Release form –developed with input from the DMLA–, the stock photo powerhouse aims at offering data privacy for models regarding the use of their likeness to train machine learning algorithms used for AI photo creation. While the legal framework for biometric data protection is still in the works, the fact a major agency like this one is already adapting its legal forms to include it is very telling that this is where the industry is moving towards.
DALL-E‘s recent opening with full commercial use rights for content generated on their platform is another interesting point regarding the legal frame. The software's content policy and safety measures have done a lot to protect itself, its content, and everyone's rights, with protocols that block attempts from creating photos of public figures (like celebrities) and real-life individuals, or from generating content considered harmful –on concepts like violence, adult content, politics, conspirations, etc.–. In fact, from early on their system simply removed “tags” and content involving celebrities to make sure the software wouldn't create content related to them. However, there is not a lot of information regarding biometric data and copyright data from images used to train the software. Legally speaking, at this point, images from DALL-E aren't entirely cleared for legal use, which is a must when using visuals commercially.
Copyright: Who Owns What
And last but not least, there is the issue that the majority of images used to train AI algorithms are the result of other artists work, and are very likely copyrighted. There are a number of open questions regarding whether you can create art with an AI image generator and copyright it as your own, given it's technically made from bits of other artists' images. Not to mention, the problems that surge from having AI software trained with copyrighted content that wasn't specifically authorized for said use.
The State of AI-Generated Images in the Stock Photo Industry
Our observations about legal aspects are also those made by big stock photo agencies such as Getty Images and Shutterstock, who banned the submission of AI images to their platforms due to unaddressed or unclear copyright issues. However, Shutterstock later announced that they would accept AI-generated images, but only those created with their native Shutterstock Generate tool.
Adobe Stock was one of the first main services to open for AI-generated image submissions, although with very detailed and strict submission guidelines to ensure legal and technical quality.
Dreamstime is another player in welcoming AI-generated content into their library. And also PantherMedia has more recently begun accepting AI image submissions, too.
In connection with biometric data privacy, another tech field that seems to be developing is that of AI pictures policing: tools that can detect/flag photos as AI generated.
Here people's images also are the main focus. In a world where almost every platform, social media and beyond, demands a user profile picture, it might become a necessity to be able to identify fake people's photos from real-life ones – how sci-fi does that sound.
And some such tools are already available. V7 Labs recently released an AI-based Chrome extension that identifies AI-generated people pictures (intended to flag fake profiles online). It claims to have a 99.2% accuracy. However, we would take that number with a grain of salt, because we tested it with vAIsual via Smarterpix AI people images… and it did not pick up a single one of them.
Accuracy of this specific tool aside, it is undeniable that with the perspective of AI people images spreading, it will be important to be able to tell them apart from real people photos in certain contexts. If you want to use an AI person as your video game avatar profile, that might be harmless. But if someone is using an AI photo in a fake LinkedIn profile to scam people, then it’s a problem.
Another important aspect of photo policing is the one that concerns content creators –photographers, illustrators, and visual artists– whose work is included in training datasets for AI software.
Since not all AI image generators are disclosing the datasets they used, and even if they do, they haven't always sought permission from all the copyright owners for the millions or billions of images included in them, having tools to identify whether your work was used to train AI tools or not, can be useful.
Spawning.ai is one company that wants to address this issue, and for that, they've made available a feature appropriately named Have I Been Trained?, where you can enter text or an image to check against a database of 5.8 million pictures that have been used to train AI photo tools. If you find your photo there, you can be sure it has been used in AI training.
As an emerging field, AI generated pictures are still unknown in a lot of aspects. Here we will quickly address the most frequently asked questions about AI generated photos.
What are generated photos? What are AI images?
The terms generated photos and AI images are pretty much interchangeable today. Generated photos refer to photography that is generated, not captured by a camera.
AI images are pictures generated through artificial intelligence software. Basically, digital images are created out of thin air from a text prompt.
Can AI generate images?
Yes. A few years ago we would have to say no (or rather, not yet), but today's AI technology is capable of creating images with very impressive results, parting from text instructions and a visual database.
What is an AI art generator?
An AI art generator is a software tool that creates visual, digital art –paintings, illustrations, etc.– using AI technology. These tools let you create technically superb visual art even if you are not an artist.
Can I sell AI generated art?
Cautiously, yes. The legal frame for AI generated images to be licensable is still in the works, but as this article explains there are several companies pioneering in this aspect, particularly regarding AI people photos and biometric data management, in a way that makes it possible to sell AI generated pictures safely.
And that’s what’s up with AI generated images right now. It’s quite significant if you ask us.
To recap, vAIsual and other groundbreaking companies are making it possible to create pictures of everything, including people, out of thin air. Getty Images –and potentially the whole stock photo industry– are already updating their legal forms and data security policies to cover biometric data in connection with AI software training. And other developers are coming up with tools to tell an AI generated photo from a “traditional” one.
And the field is only in its infancy! We have to wait and see what else is to come. Are you as excited as we are?