fal

fal is a pay-per-use serverless GPU platform for generative media APIs, starting at $1.89/hour for H100. It provides access to over 1,000 production-ready models for image, video, audio, and 3D generation via a unified API. Serverless inference runs on H100, H200, B200, and other GPUs, with no cold starts and global distribution. Model APIs offer per-output pricing: video models like Wan 2.5 cost $0.05 per second, Kling 2.5 Turbo Pro $0.07 per second; image models bill per image or megapixel. Dedicated clusters are available for training and fine-tuning at hourly GPU rates from $1.89 to $8.50. The platform is SOC 2 compliant, supports private endpoints, and offers SDKs for rapid integration. Trusted by over 1.5 million developers, fal is ideal for building generative media products, personalizing models, and scaling inference workloads without managing infrastructure.