When fal launched three and a half years ago, the founders saw an opportunity emerging at the intersection of AI and creativity. “We had both quit our jobs and were looking for an idea in the machine learning space,” recalls Cofounder and CTO Gorkem Yurtseven. The breakthrough came with the release of Stable Diffusion, the first major open-source image model. “[It was] a revolution,” Yurtseven says, “because for the first time you didn’t rely on your own data to train models. You could just find a model off the shelf and start using it.”

That innovation inspired the team to build infrastructure to support developers experimenting with generative media. “We decided someone had to provide infrastructure for this new type of AI application,” he explains. “We got started by hosting Stable Diffusion as an API, and we kept improving that API. Now we have over 300 to 400 different image and video models on the platform.”

Today, fal is a generative media platform for developers, offering APIs that make it simple to integrate high-performance image and video models into applications. “We provide image and video models through easy-to-use APIs powered by our inference engine,” says Yurtseven. “Developers can build products on top of our infrastructure without needing to manage the complexity of model hosting or scaling.”

Meeting Infrastructure Demands with DigitalOcean

Running hundreds of large AI models requires massive GPU capacity—and reliability. That’s where DigitalOcean comes in. “We maintain a really big GPU fleet to run these video and image workloads,” Yurtseven explains. “We have to work with reliable infrastructure providers like DigitalOcean to host these models.” Having known DigitalOcean’s reputation for reliability, the team was quick to test its new GPU offering. “I knew DigitalOcean from a long time ago for reliable CPU-based application hosting,” he says. “When I saw that they also had support for GPUs, we decided to give it a try—and it’s been very reliable so far.”

Now fal leverages DigitalOcean Droplets, GPUs, and Storage to run and serve their AI workloads. “We connect them to our system and use them to run image and video models,” Yurtseven says. Behind the scenes, fal’s APIs route user requests to models deployed on DigitalOcean GPUs, powering creative applications across industries.

A Partnership Built on Performance and Support

Finding GPU capacity can be a constant challenge for AI companies—but fal’s experience with DigitalOcean has been refreshingly smooth. “Finding GPU capacity is always a challenge,” says Yurtseven. “With DigitalOcean, the sales team, the support team, even the executives were really nice to us from the beginning and were able to accommodate our needs.”

The teams stay in close contact through a shared Slack channel. “We talk almost daily,” he adds. “They communicate really well, especially when they need to take certain machines into maintenance. It’s been a really pleasant experience for the whole team.”

Beyond strong collaboration, fal values DigitalOcean’s simplicity, cost efficiency, and reliability. “Simplicity and unit cost were very attractive in the beginning—that maybe opened the doors,” says Yurtseven. “But once we proved that everything was reliable and easy to use, we moved a lot more capacity to the platform.”

He also highlights DigitalOcean’s uptime and stability as critical advantages. “GPU workloads usually tend to have a lot of failures,” he says. “DigitalOcean was definitely one of the most reliable providers we worked with. As we tested reliability in the early days, we kept moving more of our workloads to DigitalOcean—and we’ll keep doing that as long as we get capacity.”

Looking Ahead: Scaling Generative Media for the Future

For fal, the generative media space is still in its infancy. “Image and video generation hasn’t hit mainstream yet,” Yurtseven notes. “There’s a lot more work to be done in terms of model quality, image and video editing capabilities, and larger models that unlock new use cases.”

As the field evolves, fal aims to maintain its leadership at the intersection of open-source AI and enterprise applications. “We want to keep our leadership position and continue delivering value,” says Yurtseven. “Whenever a new capability is introduced, we see more use cases open up—and different types of customers start taking advantage of these models.”

That future includes a deeper partnership with DigitalOcean. Through the DigitalOcean AI Platform, fal will soon make its high-performance image and video models accessible to DigitalOcean users. “With our partnership, we’re making some of the models hosted by fal available on the DigitalOcean AI Platform,” Yurtseven says. “We believe we’re the most performant image and video inference provider in the market, and everyone doing business with DigitalOcean will have access to these models powered by our inference engine.”

Interested in powering your application with DigitalOcean and fal? Right this way!