Fal.aia developer-focused platform for AI-generated audio, video, and image files, today announced that it has raised $23 million in funding from investors including Andreessen Horowitz (a16z), Black Forest Labs co-founder Robin Rombach, and Perplexity CEO Aravind Srinivas.
It's a two-round deal: $14 million of Fal's total got here from a Series A tranche led by Kindred Ventures; the remaining $9 million got here from a previously unannounced seed round led by a16z.
Burkay Gur and Gorkem Yurtseven co-launched Fal (short for “Features and Labels”) in 2021. Yurtseven previously worked at Amazon as a software developer, while Gur, a former Oracle engineer, led machine learning development at Coinbase for several years.
While tinkering with side projects in the course of the pandemic, Gur and Yurtseven, longtime friends, recognized the growing demand for AI cloud infrastructure—particularly infrastructure for running generative AI models.
“The big bet was that the emerging field of generative media was about to rework all media consumed,” Gur told TechCrunch. “The timing was perfect because right after Fal launched, some groundbreaking models were released.”
Fal offers two products: privately managed compute power and workflows for running models, and APIs for open source models that generate images, audio, and video. Fal was certainly one of the primary platforms to host Black Forest Labs' Flux, the model that permits image generation in Grok, X's controversial chatbot.
Many cloud competitors like CoreWeave offer similar resources, but what sets Fal apart is its scalability, Gur argues.
“Our platform can handle a whole lot of thousands and thousands of queries (and our) own inference engine is essentially the most powerful,” he said. “With Fal, you possibly can integrate models into your applications – the product is designed for firms where media is at the center of their work.”
Whether or not these claims delay under scrutiny, Fal has managed to construct a powerful customer base. In addition to Perplexity (which explains Srinivas' investment) and enterprise clients within the retail and e-commerce sectors, popular generative AI apps Photoroom, Freepik and PlayHT also pay for Fal's services, in keeping with Gur.
It's a profitable bunch. A source tells TechCrunch that Fal's annual revenue has risen to almost $10 million (about $800,000 per 30 days), up about tenfold since January. The Series A valued the startup at $80 million.
“Fal has reached 500,000 creators on the platform,” Gur said, “and generates 50 million images, videos or audio streams per day.”
Given the numerous deepfake and disinformation risks related to generative technologies, I asked Gur if Fal has moderation policies or filters for sensitive content. He said that Fal prefers to take a laissez-faire approach, leaving the choice of whether to implement security measures to the businesses developing models on Fal's platform.
“Lots of the moderation happens during training, so we leave that to the businesses that construct the models,” Gur said. “As you possibly can imagine, a really robust program requires more research and resources.”
That's a reasonably empty answer considering that Fal sponsors some open source training efforts as a part of his research grant program. One might assume that Fal has some say in the event of the models he funds.
However, Gur hinted that Fal would really like to eventually do some cleansing work of his own. “We have plans to do more of it in-house, counting on some vendors who focus on this sort of work,” he said.
I also asked about mental property liability. If the models on Fal's platform reproduce copyrighted data, will the corporate protect customers in the event that they are sued? Gur declined to reply that query. But the language in Fal's Terms of Service imply that customers are left to their very own devices.
This is in contrast to generative AI products from Adobe, Canva, Google, Microsoft and Shutterstock, all of which have indemnity clauses (albeit with some exceptions). Vendors like Getty Images, in addition to startups like Fairly Trained, have gone to date as to coach models only on “commercially secure” content to avoid the specter of copyright lawsuits altogether.
This implies that anyone who uses Fal takes a certain risk.
Fal plans to spend a lot of the capital raised to date on further developing its inference optimization product to make it self-serviceable. The company can be constructing a research team that can give attention to model optimization and complement Fal's 17-person team.
Fal’s other backers include Vercel founder Guillermo Rauch, entrepreneur and investor Balaji Srinivasan and Hugging Face CTO Julien Chaumond.