Image Generation

Generate images from text prompts using the generateImage() function with support for various AI image models.

Basic Usage

Generate an image from a text prompt:

import { generateImage } from '@core-ai/core-ai';
import { createOpenAI } from '@core-ai/openai';

const openai = createOpenAI({ apiKey: process.env.OPENAI_API_KEY });
const model = openai.imageModel('gpt-image-1');

const result = await generateImage({
    model,
    prompt: 'A watercolor painting of a robot coding in a quiet mountain cabin at sunrise',
    n: 1,
    size: '1024x1024',
});

for (const [index, image] of result.images.entries()) {
    console.log(`Image ${index + 1}:`);
    if (image.url) {
        console.log('URL:', image.url);
    } else if (image.base64) {
        console.log('Base64 preview:', image.base64.slice(0, 80), '...');
    } else {
        console.log('No image payload found.');
    }
    if (image.revisedPrompt) {
        console.log('Revised prompt:', image.revisedPrompt);
    }
}

Google image models

Google providers expose Gemini native image models through the same API:

Google GenAI
Google on Vertex AI

import { generateImage } from '@core-ai/core-ai';
import { createGoogleGenAI } from '@core-ai/google-genai';

const google = createGoogleGenAI({
  apiKey: process.env.GOOGLE_API_KEY,
});

const result = await generateImage({
  model: google.imageModel('gemini-nano-banana-2.1'),
  prompt: 'A watercolor painting of a small orange robot watering a cactus on a windowsill',
  size: '1024x1024',
});

import { generateImage } from '@core-ai/core-ai';
import { createGoogleVertex } from '@core-ai/google-vertex';

const googleVertex = createGoogleVertex({
  projectId: process.env.GOOGLE_VERTEX_PROJECT,
  region: 'europe-west1',
});

const result = await generateImage({
  model: googleVertex.imageModel('gemini-2.5-flash-image'),
  prompt: 'A watercolor painting of a small orange robot watering a cactus on a windowsill',
  size: '1024x1024',
});

Gemini image model IDs use native multimodal generation and return base64 image data. They do not support requesting multiple outputs with n. Model availability on Vertex AI depends on the selected project and region.

Configuration Options

Customize image generation:

const result = await generateImage({
    model,
    prompt: 'A futuristic cityscape at sunset',
    n: 3, // Generate 3 images
    size: '1024x1024', // Image dimensions
});

Available sizes depend on the provider and model. Common sizes include: - 256x256 - 512x512 - 1024x1024 - 1792x1024 (landscape) - 1024x1792 (portrait)

Image Response Format

The generateImage() function returns:

type ImageGenerateResult = {
    images: GeneratedImage[];
};

type GeneratedImage = {
    base64?: string; // Base64-encoded image data
    url?: string; // URL to the generated image
    revisedPrompt?: string; // AI-revised prompt (if modified)
};

Multiple Images

Generate multiple variations:

const result = await generateImage({
    model,
    prompt: 'A minimalist logo for a tech startup',
    n: 4, // Generate 4 variations
    size: '512x512',
});

console.log(`Generated ${result.images.length} images`);

result.images.forEach((image, index) => {
    console.log(`\nImage ${index + 1}:`);
    if (image.url) {
        console.log('URL:', image.url);
    }
});

Handling Image Data

Save to File
Display in Browser
Upload to Storage

import { writeFile } from 'fs/promises';
import { generateImage } from '@core-ai/core-ai';

const result = await generateImage({
  model,
  prompt: 'A beautiful landscape',
});

const image = result.images[0];

if (image.base64) {
  // Save base64 data
  const buffer = Buffer.from(image.base64, 'base64');
  await writeFile('generated-image.png', buffer);
  console.log('Saved image to generated-image.png');
} else if (image.url) {
  // Download from URL
  const response = await fetch(image.url);
  const buffer = await response.arrayBuffer();
  await writeFile('generated-image.png', Buffer.from(buffer));
  console.log('Downloaded and saved image');
}

import { generateImage } from '@core-ai/core-ai';

async function displayImage() {
  const result = await generateImage({
    model,
    prompt: 'A modern office space',
  });

  const image = result.images[0];

  if (image.url) {
    // Display URL directly
    document.getElementById('output').innerHTML =
      `<img src="${image.url}" alt="Generated image" />`;
  } else if (image.base64) {
    // Display base64 as data URL
    const dataUrl = `data:image/png;base64,${image.base64}`;
    document.getElementById('output').innerHTML =
      `<img src="${dataUrl}" alt="Generated image" />`;
  }
}

import { generateImage } from '@core-ai/core-ai';
import { S3Client, PutObjectCommand } from '@aws-sdk/client-s3';

const s3 = new S3Client({ region: 'us-east-1' });

const result = await generateImage({
  model,
  prompt: 'A product photo',
});

const image = result.images[0];
let buffer: Buffer;

if (image.base64) {
  buffer = Buffer.from(image.base64, 'base64');
} else if (image.url) {
  const response = await fetch(image.url);
  buffer = Buffer.from(await response.arrayBuffer());
} else {
  throw new Error('No image data');
}

await s3.send(
  new PutObjectCommand({
    Bucket: 'my-bucket',
    Key: 'generated-image.png',
    Body: buffer,
    ContentType: 'image/png',
  })
);

console.log('Uploaded to S3');

Revised Prompts

Some providers modify your prompt for better results:

const result = await generateImage({
    model,
    prompt: 'a robot',
});

const image = result.images[0];

if (image.revisedPrompt) {
    console.log('Original prompt:', 'a robot');
    console.log('Revised prompt:', image.revisedPrompt);
    // Revised prompt: "A humanoid robot with sleek metallic surfaces..."
}

Image Sizes

Different models support different sizes:

Square
Landscape
Portrait

const result = await generateImage({
  model,
  prompt: 'A centered logo design',
  size: '1024x1024', // Square format
});

const result = await generateImage({
  model,
  prompt: 'A wide panoramic landscape',
  size: '1792x1024', // Landscape format
});

const result = await generateImage({
  model,
  prompt: 'A tall building from ground level',
  size: '1024x1792', // Portrait format
});

Prompt Engineering Tips

Write effective prompts for better results:

Be specific and descriptive

// Vague
const result = await generateImage({
  model,
  prompt: 'a dog',
});

// Better
const result = await generateImage({
  model,
  prompt: 'a golden retriever puppy sitting on grass in a sunny park, ' +
          'photorealistic, 8k, natural lighting',
});

Include style and quality keywords

const result = await generateImage({
  model,
  prompt: 'a mountain landscape at sunset, ' +
          'oil painting style, impressionist, ' +
          'vibrant colors, detailed brushstrokes',
});

Specify composition and perspective

const result = await generateImage({
  model,
  prompt: 'a modern kitchen, ' +
          'wide angle shot, ' +
          'centered composition, ' +
          'eye-level perspective, ' +
          'well-lit interior',
});

Add technical details for realism

const result = await generateImage({
  model,
  prompt: 'portrait of a woman, ' +
          'Canon EOS R5, 85mm f/1.4, ' +
          'shallow depth of field, ' +
          'natural window lighting, ' +
          'bokeh background',
});

Error Handling

Handle image generation errors:

import { CoreAIError, ProviderError } from '@core-ai/core-ai';

try {
    const result = await generateImage({
        model,
        prompt: 'A beautiful landscape',
    });

    if (result.images.length === 0) {
        console.error('No images generated');
    }

    result.images.forEach((image, i) => {
        if (!image.url && !image.base64) {
            console.error(`Image ${i} has no data`);
        }
    });
} catch (error) {
    if (error instanceof ProviderError) {
        console.error('Provider error:', error.message);
        console.error('Status code:', error.statusCode);
    } else if (error instanceof CoreAIError) {
        console.error('CoreAIError:', error.message);
    } else {
        console.error('Unknown error:', error);
    }
}

Provider-Specific Options

Use provider-specific features:

const result = await generateImage({
    model,
    prompt: 'A futuristic vehicle',
    n: 2,
    size: '1024x1024',
    providerOptions: {
        openai: {
            quality: 'hd',
            style: 'vivid',
        },
    },
});

Provider options are namespaced by provider name and validated by each provider adapter. Check the provider docs for available options.

Practical Examples

Logo Generator
Product Mockup
Illustration Generator

async function generateLogo(companyName: string, industry: string) {
  const result = await generateImage({
    model,
    prompt: `A modern, minimalist logo for "${companyName}", ` +
            `a ${industry} company, ` +
            `clean design, professional, vector style, ` +
            `white background, high contrast`,
    n: 3,
    size: '512x512',
  });

  return result.images;
}

const logos = await generateLogo('TechFlow', 'software development');
console.log(`Generated ${logos.length} logo options`);

async function generateProductMockup(
  product: string,
  style: string = 'photorealistic'
) {
  const result = await generateImage({
    model,
    prompt: `${product}, product photography, ` +
            `${style}, studio lighting, ` +
            `clean white background, 4k, commercial quality`,
    size: '1024x1024',
  });

  return result.images[0];
}

const mockup = await generateProductMockup(
  'a sleek wireless headphone',
  'premium and elegant'
);

async function generateIllustration(
  scene: string,
  artStyle: string = 'digital art'
) {
  const result = await generateImage({
    model,
    prompt: `${scene}, ${artStyle}, ` +
            `highly detailed, vibrant colors, ` +
            `professional illustration, ` +
            `trending on artstation`,
    size: '1792x1024',
  });

  return result.images[0];
}

const illustration = await generateIllustration(
  'a fantasy castle on a floating island',
  'concept art'
);

Best Practices

Handle both URL and base64 responses

Different providers return different formats:

const image = result.images[0];

let imageData: string;
if (image.url) {
  imageData = image.url;
} else if (image.base64) {
  imageData = `data:image/png;base64,${image.base64}`;
} else {
  throw new Error('No image data available');
}

Store generated images permanently

URLs may expire - save images you need:

if (image.url) {
  // URL might expire, download and store
  const response = await fetch(image.url);
  const buffer = await response.arrayBuffer();
  await saveToStorage(buffer);
}

Use appropriate sizes for your use case

Smaller sizes are faster and cheaper:

// Thumbnail: use smaller size
const thumbnail = await generateImage({
  model,
  prompt: 'icon design',
  size: '256x256',
});

// High quality print: use larger size
const print = await generateImage({
  model,
  prompt: 'artwork',
  size: '1024x1024',
});

Next Steps

Multi-Modal

Use images in chat conversations

Chat Completion

Generate text responses

Tool Calling

Combine image generation with tools

Embeddings

Generate vector embeddings

Get Started

Core Concepts

Guides

Observability

Examples

Image Generation

Basic Usage

Google image models

Configuration Options

Image Response Format

Multiple Images

Handling Image Data

Revised Prompts

Image Sizes

Prompt Engineering Tips

Error Handling

Provider-Specific Options

Practical Examples

Best Practices

Next Steps

Multi-Modal

Chat Completion

Tool Calling

Embeddings

​Basic Usage

​Google image models

​Configuration Options

​Image Response Format

​Multiple Images

​Handling Image Data

​Revised Prompts

​Image Sizes

​Prompt Engineering Tips

​Error Handling

​Provider-Specific Options

​Practical Examples

​Best Practices

​Next Steps

Multi-Modal

Chat Completion

Tool Calling

Embeddings

Basic Usage

Google image models

Configuration Options

Image Response Format

Multiple Images

Handling Image Data

Revised Prompts

Image Sizes

Prompt Engineering Tips

Error Handling

Provider-Specific Options

Practical Examples

Best Practices

Next Steps