Fal.ai, a platform focused on developing AI-generated audio, video, and imagery, today revealed 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.
This is a two-round deal: $14 million of Fal's total funding came from a Series A tranche led by Kindred Ventures. The remaining $9 million came from a previously unannounced seed round led by a16z.
Burkay Gur and Gorkem Yurtseven co-founded Fal (short for “Features and labels”) in 2021. Gur previously worked as a software developer at Amazon, while Yurtseven, a former Oracle engineer, led machine learning development at Coinbase for several years.
While working together on side projects during the pandemic, Gur and Yurtseven, longtime friends, noticed an increased demand for AI cloud infrastructure, particularly infrastructure to run generative AI models.
“The big bet was that the dawn of generative media was going to change all of the media we consume,” Gur told TechCrunch. “The timing was perfect, because shortly after Phal started, some groundbreaking models were released.”
Fal offers two products: privately managed compute and workflows for running models, and an API for open-source models that generate images, audio, and video. Fal was one of the first platforms to host Black Forest Labs' Flux, the model that powers image generation on X's controversial chatbot Grok.
Many cloud rivals, such as CoreWeave, offer similar resources, but Gur argues that what sets Fal apart is its scalability.
“Our platform can handle hundreds of millions of requests. [and our] “Our inference engine is the most powerful,” he says. “Fal allows you to integrate the models into your applications. This product is for companies that have media at the core of their business.”
Whether or not these claims stand up to scrutiny, Fal has managed to build an impressive client list: In addition to Perplexity, the reason for Srinivas' investment, and enterprise clients in the retail and e-commerce sectors, popular generative AI apps Photoroom, Freepik and PlayHT are all paying for Fal's services, Gur says.
Model from Fal.ai's model gallery. Image courtesy of Fal
It's a profitable business: A source told TechCrunch that Fal's annual revenue is approaching $10 million (about $800,000 per month), up roughly tenfold since January. In its Series A, the startup was valued at $80 million.
“Fal has 500,000 developers on its platform, generating 50 million images, videos and audio streams per day,” Gur said.
Given the many risks of deepfakes and misinformation with generative technology, I asked Ghar whether Fal has any moderation policies or filters for sensitive content. Ghar said Fal prefers a hands-off approach, leaving it up to the companies developing models on the Fal platform to decide whether to implement safety features.
“When it comes to moderation, a lot of the work is done during training, so we leave that to the companies that build the models,” Gurr said. “As you can imagine, it takes a lot more research and resources to build a very robust program.”
This is a bit of a hollow answer, given that Fal supports open source training activities as part of its research grants program, and one would assume that Fal has a say in the development of the models it funds.
But Mr. Gurr suggested that Phal might consider taking on the detoxification efforts itself at some point. “We have plans to do more of this work in-house and rely on contractors who specialize in this type of work,” Mr. Gurr said.
I also asked about intellectual property liability: If a model on Fal's platform spits out copyrighted material, would the company protect its customers if sued? Gur didn't answer, but the wording of Fal's terms of service suggests that customers are at their own risk.
This is in contrast to generative AI products from the likes of Adobe, Canva, Google, Microsoft, and Shutterstock, all of which come with disclaimers (with some exceptions), with vendors like Getty Images and startups like Fairly Trained going so far as to train their models only on “commercially safe” content to avoid the threat of copyright litigation altogether.
This means that anyone using Fal takes on a certain amount of risk.
Fal plans to use most of the funding it has raised so far to upgrade its inference optimization product to make it self-service. The company will also set up a research team focused on model optimization, which will join Fal's staff of 17.
Fal's other backers include Vercel founder Guillermo Rauch, entrepreneur and investor Balaji Srinivasan, and Hugging Face CTO Julien Chaumond.