@modelslab/ai-sdk-provider

ModelsLab provider for the Vercel AI SDK. It adds image generation through the ModelsLab v7 text-to-image API, with access to Flux, SDXL, Imagen and thousands of other ModelsLab image models.

• Implements the AI SDK ImageModelV4 specification (AI SDK 7).
• Polls queued generations until the images are ready.
• Returns the images as bytes, plus the hosted image URLs in providerMetadata.

Install

npm install @modelslab/ai-sdk-provider ai

Requires Node.js 22 or later and ai 7.x.

API key

Create an API key in the ModelsLab dashboard and set it as MODELSLAB_API_KEY. API access requires a paid ModelsLab plan; plans start at $21/month (pricing).

export MODELSLAB_API_KEY=your_api_key

Usage

import { modelslab } from '@modelslab/ai-sdk-provider';
import { generateImage } from 'ai';
import { writeFileSync } from 'node:fs';

const { image } = await generateImage({
  model: modelslab.image('flux'),
  prompt: 'A red apple on a white table, studio lighting',
  size: '1024x1024',
});

writeFileSync('apple.png', image.uint8Array);

Custom provider instance

import { createModelsLab } from '@modelslab/ai-sdk-provider';

const modelslab = createModelsLab({
  apiKey: process.env.MODELSLAB_API_KEY, // default
  baseURL: 'https://modelslab.com/api/v7', // default
  headers: { 'x-my-header': 'value' },
  // fetch: customFetch,
});

Models

Pass any ModelsLab image model_id to modelslab.image(), for example flux, sdxl, midjourney, imagen-3 or imagen-4. Browse models at modelslab.com/models.

Settings

AI SDK setting	Sent to ModelsLab as	Notes
prompt	prompt
n	samples	Up to 4 per call. The AI SDK splits larger requests.
size ("WxH")	width, height
aspectRatio	aspect_ratio
seed	—	Not supported; returns an unsupported warning.
files, mask	—	Not supported yet; return an unsupported warning.

Provider options

const { images, providerMetadata } = await generateImage({
  model: modelslab.image('flux'),
  prompt: 'A lighthouse at dusk',
  n: 2,
  providerOptions: {
    modelslab: {
      track_id: 'order-123',
      pollIntervalMillis: 2000, // default 2000
      pollTimeoutMillis: 300000, // default 300000 (5 minutes)
    },
  },
});

console.log(providerMetadata.modelslab); // { images: [{ url }], id }

pollIntervalMillis and pollTimeoutMillis control the provider. All other options (width, height, size, resolution, colors, background_color, track_id, webhook, ...) go to the API unchanged and override the values derived from the AI SDK settings.

How it works

1. POST {baseURL}/images/text-to-image with your API key in the JSON body (key).
2. If the response status is processing, poll POST {baseURL}/images/fetch/{id} until it is success, failed/error, or the poll timeout expires. The AI SDK abortSignal stops the polling.
3. Download each output URL and return the bytes.

ModelsLab returns HTTP 200 with "status": "error" for most failures. The provider converts these to an APICallError with the API message. The API key is removed from the error's requestBodyValues.

Development

npm install
npm run typecheck
npm test
npm run build

Links

• ModelsLab API documentation
• AI SDK image generation

License

MIT