Anvia

Image generation models ​

Image generation models create image bytes from a text prompt through a provider-neutral Anvia request. Run generation on a server, worker, or internal tool so provider credentials remain private.

1. Create an image model ​

The provider client owns the provider-specific model implementation.

ts
import { OpenAIClient } from '@anvia/openai'

const client = new OpenAIClient({ apiKey })

export const imageModel = client.imageGenerationModel({ modelId: 'gpt-image-2' })

OpenAI, Gemini, and Grok provide v1 image-generation adapters. Each provider may support different dimensions, formats, and additional parameters.

2. Generate an image ​

Pass the model, prompt, dimensions, and provider options in one object. Pass retries to override retry behavior and abortSignal to cancel generation.

ts
import { writeFile } from 'node:fs/promises'
import { generateImage } from '@anvia/core/image-generation'
import { imageModel } from './models'

const response = await generateImage({
    prompt: 'A clean product illustration of a document ingestion pipeline',
    model: imageModel,
    width: 1024,
    height: 1024,
    providerOptions: {
        output_format: 'png',
    }
})

await writeFile('document-pipeline.png', response.images[0].data)

Width and height default to 1024 and must be positive integers. Provider-specific options remain explicit in providerOptions.

3. Read the normalized response ​

The response exposes the first image directly and preserves every returned image.

ts
console.log({
  firstImageBytes: response.images[0].data.byteLength,
  imageCount: response.images.length,
  mediaType: response.images[0].mediaType,
})

for (const image of response.images) {
  console.log(image.data.byteLength, image.mediaType)
}

rawResponse remains available when application code needs provider-specific metadata, but avoid returning it directly to clients.

4. Design the production boundary ​

Validate prompts, dimensions, formats, and account-specific limits before invoking the model. Generated media can be slow and expensive, so long-running or bulk work usually belongs in a queue.

Store output in application-owned object or media storage with the provider, model, format, safe prompt metadata, and ownership scope. Do not place raw image bytes in agent memory or routine logs.

Continue with Audio generation models.

Built for Anvia.