fal.ai
Generative media infrastructure with one API for image, video, audio, and 3D models.
- A unified API for a large catalog of image, video, audio, and 3D models.
- Serverless inference with an optimized engine, queueing, and autoscaling.
- Private deployments for custom, fine-tuned, or bring-your-own models.
- On-demand GPU compute across RTX, H100, H200, B200, GB200, and B300 hardware.
Best for: Developers building image, video, or audio features into an app. · AI product teams that need to test and switch between models. · Creative tool builders running high-volume generation.
Introduction
fal is infrastructure for generative media. Developers use one API and a set of SDKs to call image, video, audio, and 3D models, then run production workloads on serverless GPUs or private deployments. The platform also supports fine-tuning and custom model hosting, so a team can move from a prototype to a scaled endpoint without rebuilding its stack.
Key Features
- A unified API for a large catalog of image, video, audio, and 3D models.
- Serverless inference with an optimized engine, queueing, and autoscaling.
- Private deployments for custom, fine-tuned, or bring-your-own models.
- On-demand GPU compute across RTX, H100, H200, B200, GB200, and B300 hardware.
- SDKs and documentation for common languages and frameworks.
- Enterprise controls including SSO, private endpoints, usage analytics, and SOC 2 compliance.
- Output-based pricing for models and hourly pricing for GPU compute.
Who is it for
- Developers building image, video, or audio features into an app.
- AI product teams that need to test and switch between models.
- Creative tool builders running high-volume generation.
- Enterprises that need private endpoints and procurement support.
What can you do with fal.ai
- Generate images or video from prompts through a single API.
- Compare models for a task and switch without changing providers.
- Deploy a fine-tuned model or a custom LoRA as an endpoint.
- Run batch or real-time inference on serverless GPUs.
- Train or fine-tune a model on managed compute.
- Add audio, speech, or 3D generation to a workflow.
How does fal.ai work
- Create an account and generate an API key.
- Pick a model from the catalog and review its input and output options.
- Call the model through the API or an SDK, or use a queue for longer jobs.
- Deploy a custom or fine-tuned model when you need your own weights.
- Choose serverless or GPU compute based on the workload.
- Monitor usage and costs, then scale or reserve capacity as needed.
Pricing
fal uses usage-based pricing. GPU compute is listed at $4.50 per hour for an H100 with prices as low as $2.49 per hour, $7.99 per hour for a B200, and $12.99 per hour for a B300. Model APIs are billed per output: image models from $0.003 per megapixel, video from $0.015 per second, and audio from $0.0002 per second, depending on the model.
FAQ
Do I need my own GPU?
No. Serverless endpoints run on fal's GPU fleet, and you can also rent GPU compute by the hour for custom deployments or training.
Can I deploy my own model?
Yes. fal supports private deployments, fine-tuned models, LoRAs, and bring-your-own weights on dedicated endpoints.
Which models are available?
The catalog covers image, video, audio, 3D, and world models from multiple providers, including models from Black Forest Labs, Google, OpenAI, ByteDance, and others.
What does fal cost?
Model APIs are billed per output unit, such as per image, second, or megapixel, while GPU compute is billed hourly. The pricing page lists current rates for each model and GPU.
Does fal support enterprises?
Yes. Enterprise options include private model endpoints, SSO, SOC 2 compliance, usage analytics, priority support, and reserved capacity.
How do I start building on fal?
Create an account, generate an API key, and call a model through the API or an SDK. Documentation covers authentication, queues, deployments, and SDK examples.
