After operating under the radar, the Israeli startup Bria.ai came out of stealth today (Tuesday), exclusively on Geektime, and announced an interesting collaboration with Getty Images, the largest image database in the world. Thanks to the Israeli startup’s development, Getty customers will be able to take any generic stock photo and change any object in it – all with the help of AI.
Combining the real and the synthesized
If 2021 was the year when you couldn't avoid NFTs, then 2022 is the year of AI-based image creators such as DALL-E 2. In recent months, the web has been flooded with a huge number of strange images, created with Text-To-Image mechanisms. But more often than not, if you don't have a prompt engineer available, you probably won't use the generated image for an advertisement or blog for your organization, or even for serious personal use.
During this explosive progress in the field of AI-based images, many wondered what would happen to those who earn their salary from images, such as photographers, editors and yes - also huge image databases, which seem to be left behind with rather generic images. After all, with AI, it is possible to produce unique images exactly according to the taste of the users. This is what the Israeli startup Bria.ai is trying to change; instead of just joining all kinds of existing models, it tries to come across the concept from a slightly different angle.
In an interview with Geektime, Yair Adato, CEO, and co-founder of Bria, says that although it is very easy to download images from the web, it is very difficult to change them to match exactly what you need: "Producing the image you want should be easier. I worked at Google, I worked at Trax, and 3 years ago I decided it was time that everyone could get the movie or photo they wanted... In the end, what you need is a stock photo and your photos.”
He further went on to explain that if you have, for example, a bag store, and you want to use a stock image to advertise your products, you will be faced with two choices: bring a professional photographer, models, and of course, everyone else needed to produce such a photo – or use Photoshop to place your bags, probably unconvincingly, within a stock photo you purchased. One option is expensive, the other option will most likely turn out poorly. The third option is to use DALL-E 2 which will probably not yield a result that you can use with your product.
Bria.ai 's system basically combines stock photos with your photos. You can search for an image in the image databases that the company works with and start playing with all kinds of settings: from changing the age of the models to changing the facial expression, erasing the background, defining colours, and also adjusting the image to your branding. Adato demonstrated the company's system to me, to show me an example. He searched for a picture of a person holding a smartphone. Then, I ask him to take the resulting stock image in the search engine and turn the person into a nerd. So, he just added a set of glasses, changed his hair, and even decided to place him in a server room because why not. Very interestingly, precisely in this demonstration, I could also see how the generative element of Bria’s system came into action: in this case, in Bria’s image database during the beta phase there was no image of a server room, so the system drew a "server room" with a product quite similar to the result of DALL-E 2 and the other engines. The system then planted the model in the foreground and applied a matching lighting effect that illuminates the model's face according to the background. Thus, the generic image of a person holding a smartphone became something that I could not find in another normal image bank.
Getty Images will allow you to change the photos you download from it
One of the first to implement the Israeli technology is Getty Images, the stock photo giant, which together with Bria announced a collaboration. As part of this collaboration, Getty will implement Bria's "respectable set of features" directly in iStock, its subsidiary. The collaboration between the two began two years ago when the company participated in Intel Ignite, Intel's accelerator program. When I ask Adato if they are essentially a sort of lifeline for stock image companies that began to fear for their businesses following the rise of the AI models that generate pictures upon request, he claims that he discussed this issue with the CEO of Getty: "He said that a company that understands how the market works, will have to embrace the technologies and go along with the market. That's why they made the announcement. To stay relevant in the world of AI, an image can no longer remain static but become dynamic, and now the user gets the image he or she wants."
At the same time as Bria announced this collaboration, which will go live in the coming months, their big competitor, Shutterstock, also announced its own collaboration with the giant, OpenAI, who are the creators of GPT-3 and DALL-E 2. As part of this collaboration, DALL-E 2 will be embedded directly into Shutterstock, where the model will appear to be trained on images from the company's image library, and photographers will be rewarded for purchases of AI-generated creations that use their images. It seems that the stock image companies understand that if you can't beat them, join them.
Coincidentally or not, the two giants announced these collaborations within a few hours of each other, but even though he is now competing against the titan of the field, Adato does not sound upset: "I am not surprised by Shutterstock's announcement. After all, it is an arms race, and I am very happy about the partnership of Bria with Getty. In fact, I'm happy that the field is waking up and the category is being built.” According to Adato, there are several significant differences between Bria’s platform and that of OpenAI. He claims that every actor in this field has taken a different approach. OpenAI became a kind of closed garden, while Stable Defusion decided to draw images for training from around the web, and in the meantime got into quite a bit of trouble with issues such as copyright and ethics. "We followed a completely different policy," Adato explains. "All the data used to train the models is our data, to which we have rights from other image banks. For example, I did a lot of training on people, and all I used are pictures that I have a release form on. For backgrounds, for example, I have agreements with image banks that allow me to use this data. I have a photo bank with 20 million photos, and some of them I'm allowed to sell and some of them I'm allowed to paint.” Thus, according to Adato, organizations can also use AI tools without fearing copyright issues.
Adato also repeats and emphasizes that photo banks like Getty ultimately also have a responsibility towards the photographers and creators who upload their work to the system as a means to make money: "For some people, this is their income; this is their work. They cannot disqualify the work of these people (using AI tools, etc.). I'm not trying to eliminate those creators; I'm just trying to make their pie bigger. I want a photographer who works with Getty to have more effective images. I am not replacing him. In the end, the system that is a combination of real things and synthetic things will be the best.”
Search engines without images
In addition to the photo editing system, part of Bria's technology lies precisely in its search engine, which Adato defines as Generative Search. According to him, his entire image database did not go through any labelling process. So, he demonstrated to me how a search for “I Love My Baby” leads to stock photos of babies, but a search for “I Love you baby” brings up images of couples. "It's a search based on natural conversation," he explains. "It's like when you do Text to Image you get a picture, I also do it the other way around. I'm looking for language, not pixels. It searches according to the meaning of the work.”
Bria was founded in 2020 by Yair Adato (CEO) and Gal Jacobi (President). Today the company employs a team of about 30 people, most of them in its offices in Tel Aviv and the rest in Seattle, Toronto, and Hong Kong. The company came out of stealth today (Tuesday) with a Seed round of $10.5 million led by Entrée Capital, IN Ventures, and several angel investors. In addition, the company was also granted a million dollars from Israel's Innovation Authority’s Chief Scientist’s office.