Bria Secures $40M Funding for Licensed Data-Trained AI Models

Bria lands new funding for AI models trained on licensed data | TechCrunch

AI-powered image generators, which are at the center of a number of copyright lawsuits against AI companies, are frequently trained on massive amounts of data from public websites. Most of these companies argue that fair use doctrine shields their data scraping and training practices. But many copyright holders disagree. That's why some startups and firms developing image generators are trying a different tack: Training generators exclusively on licensed content. New York and Tel Aviv-based Bria, founded in 2023 by entrepreneurs Yair Adato and Assa Eldar, is one of these. Bria pays for images from around 20 partners, including Getty Images, and uses these to train image-generating models with content guardrails. Adato, Bria's CEO, said that the platform "programmatically" compensates image owners according to their "overall influence." "Bria foundation models house one billion visuals and millions of videos," Adato told TechCruch. "Bria has mitigated biases that can sometimes emerge in AI-generated visual content by training its models on globally representative data sets. The company's models consistently produce visuals that reflect diversity, making them suited for various creative applications." Bria offers plug-ins for image editing and design apps including Photoshop and Figma, as well as a fine-tuning API that allows customers to customize the company's models for specific applications. Users can run Bria's models on the company's platform or an outside computing environment, like a public cloud. In either case, customers own the data and outputs, Adato said. "Enterprise customers can pay for access to source code and [models]," Adato said. "We provide over 30 specialized APIs for creating and modifying visuals, which customers access through subscription and usage-based pricing. Companies can pay to fine-tune our generative AI models with their brand assets, creating custom engines that maintain their visual identity." Bria's plans are ambitious. Adato tells TechCrunch that the 40-person company seeks to foster an "IP ecosystem" where businesses can access licensed images from media conglomerates for use in commercial creations, with "built-in compliance." Bria also plans to expand its platform and models to support additional media types, including music, video, and text, as well as on-device applications. "Bria continues to thrive despite broader tech industry challenges," Adato said. "While the sector faces headwinds from central tech company market maturation, macroeconomic pressures causing budget constraints, and oversaturation of basic AI wrapper applications, these factors strengthen Bria's position." While a growing number of ventures are trying to build businesses around licensed media generators, including Adobe, Spawning AI, and Shutterstock, Bria has managed to gain a foothold in the nascent market. On Thursday, the company announced that it raised $40 million in a Series B funding round led by Red Dot Capital with participation from Maor Investment, Entrée Capital, GFT Ventures, Intel Capital, and In-Venture. Bringing Bria's total raised to around $65 million, the majority of the new cash will be put toward product development, Adato said. "We are growing fast with our 40 customers, demonstrating significant annual recurring revenue growth of more than 400% last year," Adato said. "We're also expanding our team with additional expertise in several key areas: generative AI researchers and engineers in music and video, global sales and marketing leaders, IP and copyright experts, and generative AI consultants. We expect to double our team size by the end of the year."

Bria raises $40M to develop generative AI models trained on licensed data - SiliconANGLE

Bria raises $40M to develop generative AI models trained on licensed data Bria, a generative artificial intelligence startup platform for developers that uses only licensed data, today announced that it raised $40 million in early funding to scale its technology to help enterprise customers create on-brand images. Red Dot Capital led the Series B funding round with participation from Major Investment, Entrée Capital, GFT Ventures, Intel Capital and In-Venture, which brings the total capital raised by Bria to $65 million. Since text-to-image generative AI models began to wow users with the ability to create vivid pictures of cats and people based on natural language prompts, one major issue has been that AI must be trained on existing artwork. A problem comes when AI developers and companies scrape the internet for content, often without regard for the copyright ownership or original authorship of the images used to train the AI model. Training an AI model on copyrighted sources taints its outputs with the potential of producing content that might use a source that could belong to an unwilling creator. "Since its inception, Bria's mission has been to bridge the accountability gap between organizations and Gen AI through transparency, accessibility, and control," said Dr. Yair Adato, Bria founder and chief executive. "While our patented attribution technology has instilled trust in AI systems, facilitating responsible, controlled access to coveted IP content will change the game." Bria said that it exclusively trains its generative AI models exclusively on licensed data. This is possible because every piece of content that goes into the training of its model is attributed to a source that can be mapped to the output of any generation. According to the company, this allows data owners to be programmatically compensated any time their images are used in relation to the overall influence of their data on the output. To make this happen, Bria has brought on over 30 data partners, including Getty Images, Envato, Alamy, Freepix and Depositphoto. Together, the startup and its partners ensure that the training data is fully licensed and attributed so that no copyright infringement occurs during the training process and rights holders are kept in the loop. Adato said that he intends to broaden the attribution engine to cover even more types of content, including music, video and text, to provide AI with a more transparent and sustainable economy for developers and creators. Bria's platform provides source-available models to developers and powerful application programming interfaces to interface them with their applications. Developers can also integrate with AWS SageMaker, Azure AI Foundry, AI marketplaces, Comfy UI and use plug-ins for Adobe Photoshop and Figma.