Fal

fal is a generative media inference platform for developers. It provides a single API for calling more than 1,000 image, video, audio and 3D models, and it also runs custom models on serverless GPUs. Customers who need larger capacity can use dedicated compute clusters for fine tuning and training. The platform has three parts. Model APIs give hosted access to current models, including video models such as Kling and Seedance, image models such as Nano Banana, Flux and GPT Image, and others, with no GPU setup. Serverless lets teams deploy their own models and apps, with private deployments, bring your own weights and observability tooling, scaling from zero to large GPU counts. Compute offers on demand H100, H200 and B200 class machines and clusters. The vendor states its inference engine is built to run diffusion models faster than alternatives. It is aimed at engineering teams building products on generative media, and at research labs that need training capacity. Pricing is usage based. Model APIs are priced per output unit, for example $0.0398 per image for Nano Banana, $0.14 per second for Kling Video v3 Pro and $0.005 per megapixel for Z Image Turbo. Serverless GPU list prices include H100 at $4.50 per hour, H200 at $6.00, B200 at $7.99 and B300 at $12.99, with lower rates shown as low as $2.49 per hour for H100. Enterprise terms are available through sales. Prices vary by model, and settings such as resolution or duration can change the final cost.