Is there an official SDK?

Not yet — the API is plain REST and an OpenAPI document is published, so you can generate a client in your language today. The quickstart shows curl, JavaScript and Python by hand.

Can I bring my own LoRA?

Not on the image models. The API never accepts weights or filenames. Train a character from photos instead and reference it by id.

Do you support image-to-video?

Yes — Wan 2.2 image-to-video, 5 or 8 second clips at 32 fps, billed per second of output.

FastGen Cloud vs fal.ai

fal.ai is built around fast inference across a wide generative catalogue, with its own client libraries and queue. FastGen Cloud covers a narrower set of image and video tasks behind one plain REST contract and a prepaid balance.

Get an API key Quickstart

Side by side

Dimension	FastGen Cloud	fal.ai
Catalogue	A curated image and video lineup: text-to-image, editing, identity, character training, image-to-video.	A broad generative catalogue spanning image, video, audio and more.
API surface	One REST contract. Every model is a string in the same request body; the job envelope never changes.	Per-model endpoints and schemas, with first-party JS and Python clients.
Integration	Plain HTTP — no SDK required. An OpenAPI document is published for codegen.	Client libraries are the primary integration path, with HTTP underneath.
Payment model	Prepaid balance, debited at submit, auto-refunded on failure. Spend is capped by the balance.	Usage-based against a funded account.
Custom weights	Not accepted by design. Custom subjects go through server-side character training.	Several models accept LoRA weights supplied by URL.
Prompt handling	Verbatim. No injected negatives, no added rating or style tags.	Varies per model.
Output retention	Up to 7 days, then deleted. Download what you need to keep.	Set by their platform.

Comparison of platform design, not of prices — fal.ai sets its own rates and changes them independently, so check its current pricing page before deciding. Last reviewed 20 August 2026.

Where the two differ most

The clearest difference is surface area. A broad catalogue is worth a lot when you are exploring what to build. Once you have shipped, breadth stops being the thing you are paying attention to and consistency starts — one auth model, one job envelope, one retry story across every operation you call.

The second difference is billing direction. A prepaid balance cannot be overrun, which matters if you are exposing generation to end users and worried about a runaway loop or an abusive account.

What you give up

A narrower catalogue is a real constraint, and it is worth stating plainly rather than talking around.

No audio, no music, no speech — this is an image and video platform only
No bring-your-own LoRA weights by URL on the image models
Fewer model variants: one strong option per task rather than a dozen to benchmark

What migrating actually involves

The shape is the same everywhere: authenticate, submit a job, wait for a callback, download the output. In practice a migration is one client module and a model-name mapping.

submit + wait

const res = await fetch("https://api.fastgencloud.com/v1/images/generations", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.FASTGEN_API_KEY}`,
    "Content-Type": "application/json",
    "Idempotency-Key": requestId,
  },
  body: JSON.stringify({ model: "general-image", prompt, size: "1024x1024" }),
});
const job = await res.json();          // 202 → { id, status: "queued" }

// then either poll GET /v1/jobs/{id}, or pass webhook_url and get a signed callback

How billing works here

Money is a prepaid balance in your account. You top it up with a card, each job debits it at submit time, and a failed job is refunded automatically — there is no invoice at the end of the month and no way to run up a bill you did not intend.

Rates are per image or per second of video, published on the pricing page. Images above 1 megapixel bill pro rata by pixel count. Outputs are retained for up to 7 days.

Frequently asked questions

Is there an official SDK?: Not yet — the API is plain REST and an OpenAPI document is published, so you can generate a client in your language today. The quickstart shows curl, JavaScript and Python by hand.
Can I bring my own LoRA?: Not on the image models. The API never accepts weights or filenames. Train a character from photos instead and reference it by id.
Do you support image-to-video?: Yes — Wan 2.2 image-to-video, 5 or 8 second clips at 32 fps, billed per second of output.

Start generating

Create an account, add a prepaid balance, and call the API with a key from the console. No subscription, no minimum, no per-seat pricing.

Get an API key