# fal

> fal is a generative-media inference cloud for running image, video, and audio diffusion models fast — 1,000+ models, a simple API, and pay-per-use pricing.

fal is a generative-media platform and inference cloud built for running diffusion models — image, video, audio, and 3D — fast. It exposes 1,000+ ready models through a single API with client SDKs, plus custom training and private deployments, billed pay-per-use. Its proprietary Inference Engine targets media generation rather than LLM text serving.

Website: https://fal.ai

fal is a **generative-media platform and inference cloud** for running diffusion and other media models fast. Instead of provisioning GPUs and building inference servers, you call a model — image generators like FLUX, video models, audio and text-to-speech, or 3D — through a single API and get results back. The platform hosts 1,000+ ready-to-run models in its gallery, with client SDKs for JavaScript, Python, and Swift.

What sets fal apart from LLM-text inference clouds is its **center of gravity: diffusion and generative-media models**, not language-model serving. It is aimed at developers building image, video, and audio generation features into applications, and its proprietary Inference Engine is tuned for that workload. In an AI-assisted workflow, fal is where a generated media feature actually runs, while your agent or app handles orchestration and prompts.

## Highlights

- **Optimized inference engine** — fal's proprietary Inference Engine is built for diffusion models and claims large speedups over generic serving, which matters for interactive, latency-sensitive media generation.
- **Large model gallery** — 1,000+ production-ready models across image, video, audio, and 3D, callable with no setup or fine-tuning required.
- **Simple API and client SDKs** — official clients for JavaScript, Python, and Swift, plus a server proxy so credentials stay off the client.
- **Pay-per-use with no subscription** — billed per output (e.g. per image) on prepaid credits; no charge for queue time or server errors, with hourly GPU and private deployment options for heavier use.

## In an AI-assisted workflow

fal slots in as the media-generation backend behind an app or agent. You call a model by id and pass inputs, then handle the returned asset URLs. For whether a hosted API or self-hosting fits your case, see [Self-host vs. API](/guides/mlops/self-host-vs-api-llm).

```ts
import { fal } from "@fal-ai/client";

fal.config({ credentials: process.env.FAL_KEY });

const result = await fal.subscribe("fal-ai/flux/dev", {
  input: { prompt: "a watercolor fox in a misty forest" },
});

console.log(result.data.images[0].url);
```

> [!TIP]
> Start in the model's interactive playground to dial in inputs, then copy the generated API call into your code — it keeps prompt and parameter tuning out of your edit-run loop.

## Good to know

fal is a proprietary, hosted platform; its client SDKs are open source, but the inference cloud itself is the product, not something you self-host. Pricing is pay-per-use and varies by model and billing unit, so confirm current rates and the exact unit on the model's gallery page and at fal.ai/pricing before you budget. fal targets media generation rather than vision understanding — for choosing models that interpret images, see [Vision-Language Models Compared (2026)](/guides/vision/vision-language-models-compared-2026).

---

_Source: https://agentscamp.com/tools/fal — Tool on AgentsCamp._