@apicity/fal

v0.7.3

Published

2 days ago

Fal Platform API provider for model management, pricing, usage, and analytics.

0High
0Medium
0Low

justintanner

ai fal fal-ai platform-api models pricing usage analytics apicity

@apicity/fal

Fal Platform API provider for model management, pricing, usage, and analytics.

Runtime dependencies:

zod@^4.4.3 — request schemas attached to every POST endpoint as .schema

Installation

npm install @apicity/fal
# or
pnpm add @apicity/fal

Quick Start

import { createFal } from "@apicity/fal";

const fal = createFal({ apiKey: process.env.FAL_API_KEY! });

Real-world example: upload a portrait, render a Sora 2 video

fal's signature flow is upload-once, reuse-everywhere — drop bytes onto fal's CDN via a presigned PUT, then thread the resulting https://*.fal.media/ URL through any model endpoint. The two-step snippet below combines tests/integration/fal-storage-upload-initiate.test.ts (POST initiate → PUT bytes) with tests/integration/fal-sora-2-image-to-video.test.ts (image-to-video generation), so every URL, byte count, and asset id below comes from real recorded HARs.

import { readFile } from "node:fs/promises";
import { createFal } from "@apicity/fal";

const fal = createFal({ apiKey: process.env.FAL_API_KEY! });

// 1. Reserve a signed upload slot. `initiate` returns two URLs: a
//    permanent `file_url` you'll feed to downstream models, and a
//    presigned `upload_url` you PUT the bytes to. Both point at the
//    same fal CDN — no third-party hosting needed.
const slot = await fal.storage.upload.initiate({
  file_name: "man.jpg",
  content_type: "image/jpeg",
});
console.log(slot.file_url);
// → "https://v3b.fal.media/files/b/0a96d564/QR9a1l-E0UuoR6zOHUMlX_man.jpg"
//   (the `cat1.jpg` recording shows the same URL shape with a
//    cat1 suffix; the suffix tracks `file_name` you passed in.)

// 2. PUT the bytes to the presigned URL. fal storage is plain HTTP —
//    no SDK call needed, just `fetch` with a matching Content-Type.
//    The signature on `upload_url` expires after a short window;
//    upload promptly. The resulting `file_url` is durable and
//    fetchable by every fal model endpoint.
const bytes = await readFile("./man.jpg");
const put = await fetch(slot.upload_url, {
  method: "PUT",
  headers: { "Content-Type": "image/jpeg" },
  body: bytes,
});
if (!put.ok) throw new Error(`upload failed: ${put.status}`);

// 3. Hand the now-permanent `file_url` to OpenAI's Sora 2 image-to-
//    video model. fal returns a typed bundle: the MP4, a webp
//    thumbnail, and a horizontal spritesheet — all hosted on the
//    same fal CDN. `duration` accepts 4 | 8 | 12 | 16 | 20 (seconds);
//    `aspect_ratio` is "auto" | "9:16" | "16:9".
const result = await fal.sora2.imageToVideo({
  prompt: "the man waves at the camera as the wind blows his hair",
  image_url: slot.file_url,
  aspect_ratio: "16:9",
  duration: 4,
});

console.log(result.video.url);
// → "https://v3b.fal.media/files/b/0a96bf3c/8U5wwkg9EC_eK0Jr3XyiR_Vgq1ZZPm.mp4"
console.log(result.video.file_size);
// → 2009236   // ~2 MB MP4 for a 4-second 720p clip
console.log(result.video_id);
// → "video_69e37804033c8191959194ea8aa8fc6e08bf9f3eb453b1b1"
console.log(result.thumbnail?.url);
// → "https://v3b.fal.media/files/b/0a96bf3c/bsgsaBd5IqdwOuufu_qSx_2yOP4u34.webp"
console.log(result.spritesheet?.url);
// → "https://v3b.fal.media/files/b/0a96bf3c/_9tqG1dEuRCEeegOulGrk_pWsHbiNB.bin"

Notes

The recorded sora-2 HAR inlines the image as a data:image/jpeg;base64,… URL — fal accepts both inline data URLs and any https:// URL it can reach. Uploading via fal storage first keeps request bodies tiny (350 KB → <1 KB) and lets you reuse the asset across multiple model calls without re-encoding.
The package re-exports a one-call uploadFile(provider, { data, filename, contentType }) helper that wraps the initiate-then-PUT dance and returns the file_url directly — use it when you don't need granular control over the lifecycle or signed URL.
Every POST endpoint exposes a Zod schema: call fal.sora2.imageToVideo.schema.safeParse(input) to validate a payload before paying for inference.
WAN 2.7 reference-to-video validates generated duration as 2–10 seconds. duration: 0 remains limited to source-clip edit-video flows where it means keeping the original clip length.
Long-running calls accept an AbortSignal second argument and compose with the package's middleware, e.g. withRetry(fal.sora2.imageToVideo, { retries: 3 }) from @apicity/fal to ride out transient queue / 429s.
Errors throw FalError with status, type, request_id, and the parsed body attached: try { ... } catch (e) { if (e instanceof FalError) console.error(e.status, e.body); }.

API Reference

72 endpoints across 19 groups. Each method mirrors an upstream URL path.

bytedance

POST https://api.fal.ai/v1/fal-ai/bytedance/seed-speech/tts/v2

Cost tier: prohibitive

Published

Vulnerabilities

Links

Maintainers

Keywords

Readme

@apicity/fal

Installation

Quick Start

Real-world example: upload a portrait, render a Sora 2 video

API Reference

bytedance

falAi

gptImage1p5

hunyuan

klingVideo

models

nanoBanana

nanoBanana2

nanoBanana2Lite

nanoBananaPro

queue

qwenImage

serverless

sora2

storage

veo3p1

wan

workflows

xai

Middleware

License