fal

fal is a serverless inference platform for generative media, founded in 2021 by Burkay Gur and Gorkem Yurtseven. It hosts more than 1,000 production models — including FLUX, Veo 3, Kling, Wan, Seedream, and Nano Banana — behind a single queue-based REST API, with realtime WebSocket and SSE streaming for interactive use. A proprietary inference engine claims up to 10x lower latency, and fal Serverless lets developers deploy their own models while fal Compute offers dedicated H100/H200/A100/B200 instances.

The honest picture: fal is a middleman — you pay per generation on top of the model's own compute cost, so high-volume workloads can get expensive compared to self-hosting. Reported pricing sits around $0.03 per image and $0.10–$0.62 per video depending on model and resolution. Still, with roughly $587M raised including a $140M Series D led by Sequoia (Dec 2025) at a $4.5B valuation, 2.5M+ developers, and enterprise customers like Canva, Quora, Perplexity, Adobe, and Shopify, it has become the default way to ship generative media features without GPU DevOps.

Sign up at fal.ai, grab an API key, and call the queue API or use the fal client libraries. Serverless functions deploy with @fal.function decorators.