Bria Unveils Industry-First Open-Source Text-to-Image AI Trained Only on Licensed Data

Bria, the responsible AI platform for generative visuals, today announced the release of its third-generation text-to-image model, marking a first in the industry: a fully open-source model trained exclusively on licensed data. This milestone underscores Bria’s ethical commitment to creator rights while achieving state-of-the-art performance and efficiency.

Smaller, Smarter, and Ethically Trained

Bria’s new model delivers performance on par with other leading open-source alternatives—with 66% fewer parameters. Internal benchmarks show it requires half the compute and training data to achieve comparable results, making it highly fine-tuneable and efficient for real-world applications.

“We’re proving that ethical AI can be just as powerful—if not more efficient—than models trained on scraped data,” said Yair Adato, CEO of Bria. “Our approach shows that building smaller, responsible models doesn't mean compromising on quality.”

Head-to-Head with Industry Leaders

Bria evaluated its model alongside top open-source and licensed models using:

• Blind preference testing with thousands of design professionals

• Metrics for aesthetic quality, prompt adherence, and text rendering accuracy

• Fine-tuning efficiency and compute performance

The outcome: Bria’s model matched or exceeded competitors across key creative and technical criteria—and did so with a dramatically more sustainable and transparent training approach.

Open-Source + Full Developer Stack

The third-gen model is available now via Hugging Face, complete with:

• ControlNets

• IP Adaptors

• Auxiliary models

• Enterprise-ready platform integrations

Bria’s ecosystem is tailored for fast developer adoption, with built-in content moderation, enterprise-grade security, and seamless access through:

• MCP (Model Context Protocol) servers

• Figma and Adobe Creative Suite plugins

Solving the Creator Compensation Gap

Unlike competitors who rely on web-scraped data, Bria’s model is built entirely on licensed content, sourced through direct partnerships with artists, photographers, and content creators. This not only eliminates copyright and trademark risks but also ensures fair compensation for contributors.

“Every Bria-generated image is a vote for a sustainable creative ecosystem,” said Vered Horesh, CSO at Bria. “We’re proving that AI doesn’t have to exploit creators—it can empower them.”