Generating Images from Text with C# and Open AI DALL-E

With the use of .NET C#, we can develop a POC (Proof of Concept) that uses the Open AI DALL-E model to generate images from text input.

In this blog post, we will explore the steps to develop a C# Console Application using Open AI's DALL-E model to generate images from text.

Also, check out my blog with other articles --> https://rmauro.dev

💬 What is DALL-E from Open AI

DALL-E can generate images of almost anything, from a yellow submarine to a pig with wings. It has been trained on a massive dataset of images and textual descriptions, allowing it to learn how to generate images from natural language input. In this article, we will explore how to integrate C# with DALL-E to generate images using code.

DALL-E Generated

🧵 Development with C

In this project we're going to use Visual Studio and C# with .NET 6 to create a Console Application.

💡 You may need to do some tweaks if you intend to use another.NET version.

Step 1: Console Application and dependencies

Let's create our Console Application with C# and .NET 6 let's install the dependencies.

Project Name : ConsoleAppOpenAI.DALL_E

dotnet add Microsoft.Extensions.Http

👇 these are for loading configuration from JSON files
dotnet add Microsoft.Extensions.Configuration
dotnet add Microsoft.Extensions.Configuration.Json

👇 this is optional
dotnet add Microsoft.Extensions.Configuration.UserSecrets

Commands to install the dependencies

Installed Dependencies

Step 2: IOpenAIProxy Interface

Within this interface, we'll expose only the methods to Generate and Download the images from Open AI.

namespace ConsoleAppOpenAI.DALL_E.HttpServices;

public interface IOpenAIProxy
{
    //👇 Send the Prompt Text with and return a list of image URLs
    Task<GenerateImageResponse> GenerateImages(
        GenerateImageRequest prompt, 
        CancellationToken cancellation = default);

    //👇 Download the Image as byte array
    Task<byte[]> DownloadImage(string url);
}

IOpenAIProxy.cs File

Step 3: Generate Image Models

Let's define our models using records. Records simplify the reading since they are only POCO classes.

namespace ConsoleAppOpenAI.DALL_E.HttpServices
{
    public record class GenerateImageRequest(
        string Prompt, 
        int N, 
        string Size);

    public record class GenerateImageResponse(
        long Created, 
        GeneratedImageData[] Data);

    public record class GeneratedImageData(string Url);
}

Generate Image Models DTO

Step 4: Create an Open AI Account

To use the OpenAI API, we need to create an account on the OpenAI platform. The registration process is straightforward and can be completed in a few minutes.

• We just need to visit the OpenAI website at https://platform.openai.com/overview.
• Then click on the "Sign Up" button in the top right corner.
• Click on the button to start the registration process.

Step 5: Set up the Configuration File / appsettings.json

To access the DALL-E model, we'll need to set up the Subscription Id and API key for our application.

Collect them from these menus:

Update the appsettings.json or secrets.json file with the values.

{
  "OpenAi": {

    "OrganizationId": "{Subscription Id goes here}",
    "ApiKey": "{API Key goes here}",

    "Url": "https://api.openai.com",
    "DALL-E": {
      "Size": "1024x1024",
      "N": 1
    }
  }
}

appsettings.json file

💡 Don't forget to set Copy to Output Directory as Copy if newer for appsettings.json.

Step 6: Open AI HTTP Service Implementation

Create a class named OpenAIHttpService with a single constructor receiving IConfiguration and read the configuration we just set in place.

using ConsoleAppOpenAI.DALL_E.HttpServices;
using Microsoft.Extensions.Configuration;
using System.Net.Http.Headers;
using System.Net.Http.Json;
using System.Text.Json;

namespace ConsoleAppOpenAI.DALL_E.Services;

public class OpenAIHttpService : IOpenAIProxy
{
    readonly HttpClient _httpClient;

    readonly string _subscriptionId;

    readonly string _apiKey;

    public OpenAIHttpService(IConfiguration configuration)
    {
        //👇 reading settings from the configuration file
        var openApiUrl = configuration["OpenAi:Url"] ?? throw new ArgumentException(nameof(configuration));
        _httpClient = new HttpClient { BaseAddress = new Uri(openApiUrl) };

        _subscriptionId = configuration["OpenAi:SubscriptionId"];
        _apiKey = configuration["OpenAi:ApiKey"];
    }

    public async Task<GenerateImageResponse> GenerateImages(GenerateImageRequest prompt, CancellationToken cancellation = default)
    {
        throw new NotImplementedException();
    }

    public async Task<byte[]> DownloadImage(string url)
    {
        throw new NotImplementedException();
    }
}

Next should be the implementation of the GenerateImages() method:

public async Task<GenerateImageResponse> GenerateImages(GenerateImageRequest prompt, CancellationToken cancellation = default)
{
    using var rq = new HttpRequestMessage(HttpMethod.Post, "/v1/images/generations");

    var jsonRequest = JsonSerializer.Serialize(prompt, new JsonSerializerOptions
    {
        PropertyNamingPolicy = JsonNamingPolicy.CamelCase
    });

    //serialize the content to JSON and set the correct content type
    rq.Content = new StringContent(jsonRequest);
    rq.Content.Headers.ContentType = new MediaTypeHeaderValue("application/json");

    //👇 Including the Authorization Header with API Key
    var apiKey = _apiKey;
    rq.Headers.Authorization = new AuthenticationHeaderValue("Bearer", apiKey);

    //👇 Including the Subscription Id Header
    var subscriptionId = _subscriptionId;
    rq.Headers.TryAddWithoutValidation("OpenAI-Organization", subscriptionId);

    var response = await _httpClient.SendAsync(rq, HttpCompletionOption.ResponseHeadersRead, cancellation);

    response.EnsureSuccessStatusCode();

    var content = response.Content;

    var jsonResponse = await content.ReadFromJsonAsync<GenerateImageResponse>(cancellationToken: cancellation);

    return jsonResponse;
}

Last the DownloadImage() method implementation:

public async Task<byte[]> DownloadImage(string url)
{
    var buffer = await _httpClient.GetByteArrayAsync(url);

    return buffer;
}

Step 7: Consuming the APIs

Back to Program.cs file let's wire everything together and start calling the APIs to generate images.

using ConsoleAppOpenAI.DALL_E.HttpServices;
using ConsoleAppOpenAI.DALL_E.Services;
using Microsoft.Extensions.Configuration;
using System.Reflection;

Console.WriteLine("Starting commandline for DALL-E [Open AI]");

var config = BuildConfig();

IOpenAIProxy aiClient = new OpenAIHttpService(config);

Console.WriteLine("Type your first Prompt");
var msg = Console.ReadLine();

var nImages = int.Parse(config["OpenAi:DALL-E:N"]);
var imageSize = config["OpenAi:DALL-E:Size"];
var prompt = new GenerateImageRequest(msg, nImages, imageSize);

var result = await aiClient.GenerateImages(prompt);

foreach (var item in result.Data)
{
    Console.WriteLine(item.Url);

    var fullPath = Path.Combine(Directory.GetCurrentDirectory(), $"{Guid.NewGuid()}.png");
    var img = await aiClient.DownloadImage(item.Url);

    await File.WriteAllBytesAsync(fullPath, img);

    Console.WriteLine("New image saved at {0}", fullPath);
}

Console.WriteLine("Press any key to exit");
Console.ReadKey();

static IConfiguration BuildConfig()
{
    var dir = Directory.GetCurrentDirectory();
    var configBuilder = new ConfigurationBuilder()
        .AddJsonFile(Path.Combine(dir, "appsettings.json"), optional: false)
        .AddUserSecrets(Assembly.GetExecutingAssembly());

    return configBuilder.Build();
}

With all of this, we should have a running POC integrating with DALL-E model.

Generate our very first Image

Here is the output of my first try.

Prompt: Wide and green garden with a lot of flowers, with sunflowers, and a small dog running around

Take a look at this beautiful image generated by our application and DALL-E.

DALL-E Generated

Conclusion

Integrating C# with DALL-E is a straightforward process that allows us to generate images programmatically.

By using Open AI's API, we can easily send textual descriptions and receive high-quality images in response.

This integration opens up many possibilities, such as generating images for data visualization, creating custom artwork, or automating image creation tasks. As DALL-E continues to improve, we can expect even more exciting applications in the future.

Source code at: https://github.com/ricardodemauro/OpenAILabs.Console

Posted at: https://rmauro.dev/generating-images-from-text-with-csharp-and-open-ai-dall-e/