One client. Your Brotu key. Their models.
Video, image, speech, text, motion control, avatars and omni refine.
Bring a vendor key and generate on the vendor. Otherwise Brotu runs it.

Built by @andersonbrdev · @brotuApp

What this is

Vendors do not agree on anything. Kling wants a task id. BytePlus Seedance 400s if you send a field the model does not accept. Qwen spells resolutions in uppercase. Google Veo is another host again.

@brotu/ai is one TypeScript client for all of them. Get your key at brotu.app. Pass a vendor key you already have and that model hits the vendor. Everything else generates on Brotu.

146 models today: 74 video, 41 image, 12 speech, 19 text. Switching vendor is a model id, not a rewrite.

import { brotu } from "@brotu/ai";

const ai = brotu({
  apiKey: process.env.BROTU_API_KEY!, // brotu_sk_… from https://brotu.app
  providers: {
    kling: { apiKey: process.env.KLING_API_KEY! },
    byteplus: { apiKey: process.env.ARK_API_KEY! },
    google: { apiKey: process.env.GEMINI_API_KEY! },
    openai: { apiKey: process.env.OPENAI_API_KEY! },
    topaz: { apiKey: process.env.TOPAZ_API_KEY! },
  },
  webhook: {
    url: "https://my.app/hooks/brotu",
    secret: process.env.BROTU_WEBHOOK_SECRET,
  },
});

Your Brotu key opens the catalog. A vendor key, when you have one, generates on that vendor.

The portable surface is the same on every vendor:

Call	What it does
ai.video	text-to-video, image-to-video, edit. upscale for a source file
ai.image	text-to-image, image-to-image. upscale for a source still
ai.text	chat / completions
ai.audio	text-to-speech
ai.jobs	poll or wait on a stored handle
ai.webhook	register a URL the client POSTs when a generation settles
hooks	run your own code in-process when a generation starts, settles or fails
ai.estimateCost	units (and USD, when the catalog has a verified rate)
brotu CLI	same client in the terminal: models, video, image, job wait

Vendor-only work sits under a namespace that only exists when that key is set:

Call	What it does
ai.kling.motionControl	your character, someone else's movement
ai.kling.omniVideo	multimodal video with @id references
ai.kling.avatar	portrait + audio = talking head
ai.kling.outpainting	expand an image canvas
ai.kling.imageOmni	compose / edit across references
ai.google.omniVideo	generate a video, then refine it by talking to it

Every public method returns { data, error }. data is unusable until you narrow on error.

Install

bun add @brotu/ai

npm i @brotu/ai

pnpm add @brotu/ai

Video

submit queues the work and returns a job handle. generate is submit + wait in one call. Video takes minutes, so submit is the one you want in a request handler.

const { data: job, error } = await ai.video.submit({
  model: "kling/v2-6", // or seedance-1-0-pro-fast-251015, wan2.7-t2v, veo-3.1-fast-generate-preview
  prompt: "a cat wearing sunglasses, cinematic, slow dolly in",
  duration: 5,
  aspectRatio: "16:9",
  resolution: "720p",
});

if (error) return console.error(error.code, error.message);

const { data } = await ai.jobs.wait(job);
data.outputs[0].url;

Image-to-video is the same call with a first frame:

await ai.video.submit({
  model: "seedance-1-0-pro-fast-251015",
  prompt: "the camera pushes in",
  imageUrl: "https://example.com/frame.png",
  duration: 5,
  resolution: "480p",
});

Topaz takes a source file. Use upscale, not generate. Proteus (prob-4) is the default video model.

const { data: job } = await ai.video.upscale({
  model: "prob-4",
  videoUrl: "https://example.com/clip.mp4",
  resolution: "4k",
});

const { data } = await ai.image.upscale({
  model: "standard-v2",
  imageUrl: "https://example.com/still.jpg",
  resolution: "2K",
});

Image

Synchronous on most vendors. submit still returns a job, already settled.

const { data, error } = await ai.image.generate({
  model: "gpt-image-2",
  prompt: "a ginger cat on a windowsill, soft light",
  aspectRatio: "1:1",
  quality: "medium",
});

if (error) return console.error(error.message);
data.outputs[0].url; // data URI on OpenAI / Gemini, URL on others

Text

OpenAI goes through the Responses API (gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna). Gemini goes through the Interactions API (gemini-3.7-flash and the rest of the current ladder). Qwen uses its OpenAI-compatible chat surface.

const { data, error } = await ai.text.generate({
  model: "gpt-5.6-luna",
  prompt: "summarize this in two sentences",
  systemPrompt: "be brief",
  maxTokens: 200,
});

if (error) return console.error(error.message);
data.outputs[0].raw?.text;

Vision is the same call with referenceImages.

Text has no file, so the output is a data:text/plain URI plus raw.text. The bill is per token: estimateCost reports the rate and refuses to guess the total.

Audio (text-to-speech)

prompt is the text to speak. Gemini TTS has 30 official voices (default Kore). Qwen and ElevenLabs each have their own list.

const { data, error } = await ai.audio.generate({
  model: "gemini-3.1-flash-tts-preview",
  prompt: "Say cheerfully: good morning",
  voice: "Puck",
});

if (error) return console.error(error.message);
data.outputs[0].url;

ElevenLabs needs an explicit voice. Gemini and Qwen pick a default if you omit it.

Live / realtime WebSocket APIs are not on this surface. They do not fit generate.

CLI

One line, no Node required once a release binary exists:

curl -fsSL https://raw.githubusercontent.com/Zorbi-Tech/brotu/main/install.sh | bash

That drops brotu in /usr/local/bin (or ~/.local/bin). The CLI lives in cli/ (@brotu/cli), not inside @brotu/ai. If there is no binary for your machine it falls back to npm/bun.

Or, with Node already installed:

npx @brotu/cli --help
npm i -g @brotu/cli

brotu models
brotu video "a cat, cinematic" -m kling/v2-6 --duration 5
brotu job wait brotu-job.json --save out.mp4

BROTU_API_KEY comes from brotu.app. Vendor keys (KLING_API_KEY, ARK_API_KEY, GEMINI_API_KEY, OPENAI_API_KEY, QWEN_API_KEY, ELEVENLABS_API_KEY, TOPAZ_API_KEY) generate on the vendor when you have them. Video submits and writes a job file. --wait blocks.

Jobs

A job handle is plain JSON. Store it, hand it to another process, poll it tomorrow.

const { data: snapshot } = await ai.jobs.poll(job); // one check
const { data: result } = await ai.jobs.wait(job, { timeoutMs: 420_000 });

generate() holds the connection open for the whole run. Fine in a script. Wrong behind HTTP.

Webhook

Register a URL on the client. When generate, jobs.wait or a terminal jobs.poll settles, the SDK POSTs the result there. A down hook never fails the generation.

const ai = brotu({
  apiKey: process.env.BROTU_API_KEY!,
  providers: { kling: { apiKey: process.env.KLING_API_KEY! } },
  webhook: {
    url: "https://my.app/hooks/brotu",
    secret: process.env.BROTU_WEBHOOK_SECRET,
  },
});

// later, or instead of the constructor option
ai.webhook.set("https://my.app/hooks/brotu");
ai.webhook.clear();

Per request, if one generation should go somewhere else:

await ai.video.submit({
  model: "kling/v2-6",
  prompt: "a cat",
  webhook: "https://my.app/hooks/this-one",
});

The POST is JSON. Check x-brotu-event and, if you set a secret, x-brotu-webhook-secret.

{
  "event": "generation.succeeded",
  "jobId": "task-1",
  "provider": "kling",
  "model": "kling/v2-6",
  "kind": "video",
  "outputs": [{ "url": "https://…", "mimeType": "video/mp4" }],
  "completedAt": "2026-08-16T12:00:00.000Z"
}

generation.failed carries error: { code, message } and no outputs. Timeouts and routing errors (unknown model, missing key) do not fire the hook — nothing came back.

Error codes: unknown_model, missing_key, unsupported_provider, invalid_request, provider_error, timeout.

Hooks

A webhook needs an endpoint. A hook is a function, so it runs inside your process — send the email, write the row, push to the queue, no HTTP round trip. Both fire at the same moments, and a hook that throws never fails the generation.

const ai = brotu({
  apiKey: process.env.BROTU_API_KEY!,
  providers: { kling: { apiKey: process.env.KLING_API_KEY! } },
  hooks: {
    onVideoLoading: (e) => console.log("generating", e.model),
    onVideoSuccess: (e) => sendEmail("your video is ready", e.outputs?.[0]?.url),
    onVideoError: (e) => sendEmail("your video failed", e.error?.message),
  },
});

One optional callback per kind and stage — twelve names, all typed:

	Loading	Success	Error
image	onImageLoading	onImageSuccess	onImageError
video	onVideoLoading	onVideoSuccess	onVideoError
audio	onAudioLoading	onAudioSuccess	onAudioError
text	onTextLoading	onTextSuccess	onTextError

Loading fires once the model is routed, before the provider is called. Success and Error fire wherever the webhook fires — generate, jobs.wait, a terminal jobs.poll — deduped by job id, so a job settles once no matter how often you poll it.

Every hook receives the same event:

{
  kind: "video",
  stage: "Success",
  provider: "kling",
  model: "kling/v2-6",
  jobId: "task-1",
  outputs: [{ url: "https://…", mimeType: "video/mp4" }],
  error: undefined,          // { code, message } on Error
  metadata: { userId: "42" }, // whatever you passed on the request
  processingTimeMs: 8100,
  at: "2026-08-16T12:00:00.000Z",
}

Motion control, avatars, omni

These have no portable equivalent, so they are not on ai.video. The namespace is missing unless that key is configured.

Kling: motion control

Your character, someone else's movement.

if (!ai.kling) throw new Error("needs a kling key");

const { data: job, error } = await ai.kling.motionControl({
  model: "kling-2.6",
  imageUrl: "https://example.com/character.png",
  videoUrl: "https://example.com/dance.mp4",
  characterOrientation: "video", // `video` allows 30s; `image` caps at 10s
  resolution: "1080p",
});

if (!error) {
  const { data } = await ai.jobs.wait(job);
  data.outputs[0].url;
}

Kling: talking head

const { data: job } = await ai.kling.avatar({
  imageUrl: "https://example.com/portrait.png",
  soundFileUrl: "https://example.com/voice.mp3",
  prompt: "warm, speaking to camera, slight head movement",
  mode: "pro",
});

Kling: omni video

Everything in the prompt is addressed by @id.

const { data: job } = await ai.kling.omniVideo({
  model: "kling-3.0-omni",
  prompt: "@hero walks through the scene from @backdrop, cinematic",
  references: [
    { type: "refer_image", url: "https://example.com/hero.png", id: "hero" },
    { type: "refer_image", url: "https://example.com/street.png", id: "backdrop" },
  ],
  resolution: "1080p",
  duration: 10,
  aspectRatio: "9:16",
});

Kling: outpainting and image omni

await ai.kling.outpainting({
  imageUrl: "https://example.com/square.png",
  up: 0,
  down: 0,
  left: 0.5,
  right: 0.5,
  prompt: "continue the street scene naturally",
});

await ai.kling.imageOmni({
  prompt: "<<<image_1>>> on marble, studio light",
  imageUrls: ["https://example.com/can.png"],
});

Google: conversational Omni

Generate a video, then keep editing it by talking to the result. Pass interactionId back as previousInteractionId.

if (!ai.google) throw new Error("needs a google key");

const first = await ai.google.omniVideo({
  model: "gemini-omni-flash-preview",
  prompt: "a cat walks through rain at night",
});

const edited = await ai.google.omniVideo({
  model: "gemini-omni-flash-preview",
  prompt: "make it dawn, keep the camera",
  previousInteractionId: first.data?.interactionId,
});

Cost

Always reports the billable units. Reports USD only where the catalog has a verified rate. Token models return usd: null on purpose: the total depends on how much the model writes.

const { data } = await ai.estimateCost("video", {
  model: "kling/v3",
  prompt: "x",
  duration: 5,
  resolution: "720p",
});
data.units;
data.usd; // number, or null

Storage

Provider result URLs expire. Give the client a bucket and finished outputs land there. outputs[].url points at your copy; the original stays in sourceUrl.

const ai = brotu({
  apiKey: process.env.BROTU_API_KEY!,
  providers: { kling: { apiKey: process.env.KLING_API_KEY! } },
  storage: {
    bucket: "my-bucket",
    region: "us-east-2",
    accessKeyId: process.env.S3_KEY!,
    secretAccessKey: process.env.S3_SECRET!,
    endpoint: "https://....r2.cloudflarestorage.com",
    publicUrl: "https://cdn.example.com",
  },
});

@aws-sdk/client-s3 is an optional peer. If one output fails to copy, that output keeps the provider URL instead of failing the whole generation.

Providers

Provider	Video	Image	Speech	Text	Extra
Kling	✓	✓	✓		motion, avatar, omni, outpainting
BytePlus (Seedance, Seedream)	✓	✓
Qwen (Wan, HappyHorse)	✓	✓	✓	✓
Google (Veo, Gemini, Omni)	✓	✓	✓	✓	conversational omni
OpenAI		✓		✓
ElevenLabs			✓
Topaz	✓	✓			video + image upscale, interpolate

Every model, duration, resolution and price: sdks/node/CATALOG.md. Generated from the catalog, so it cannot drift from what the code runs.

Override a host per client when you are on another region:

providers: {
  kling: { apiKey, baseUrl: "https://api-beijing.klingai.com" },
}

Register extra models, or patch a built-in one, by id:

import { registerModels } from "@brotu/ai";

registerModels([{ id: "kling/v3", provider: "kling" /* ... */ }]);

Repo

sdks/node/        @brotu/ai
cli/              @brotu/cli
catalog/          shared model catalog as JSON

Python and Go clients are planned. They will read catalog/catalog.json.

cd sdks/node && bun install && bun run gate
cd ../../cli && bun install && bun test

Open an issue before a PR. See CONTRIBUTING.md.

Not in here

Credits, quotas, accounts and job persistence are product concerns. Metering belongs on a server.