Edit

Share via

Generate images using Azure AI with .NET

  • Article

Get started with Semantic Kernel by creating a simple .NET 8 console chat application. The application will run locally and use the OpenAI dell-e-3 model to generate postal card and invite your friends for a hike! Follow these steps to provision Azure OpenAI and learn how to use Semantic Kernel.

Get started with the .NET Azure OpenAI SDK by creating a simple .NET 8 console chat application. The application will run locally and use the OpenAI dell-e-3 model to generate postal card and invite your friends for a hike! Follow these steps to provision Azure OpenAI and learn how to use the .NET Azure OpenAI SDK.

Prerequisites

Deploy the Azure resources

Ensure that you follow the Prerequisites to have access to Azure OpenAI Service as well as the Azure Developer CLI, and then follow the following guide to set started with the sample application.

  1. Clone the repository: dotnet/ai-samples

  2. From a terminal or command prompt, navigate to the quickstarts directory.

  3. This provisions the Azure OpenAI resources. It may take several minutes to create the Azure OpenAI service and deploy the model.

    azd up

Note

If you already have an Azure OpenAI service available, you can skip the deployment and use that value in the Program.cs, preferably from an IConfiguration.

Troubleshoot

On Windows, you might get the following error messages after running azd up:

postprovision.ps1 is not digitally signed. The script will not execute on the system

The script postprovision.ps1 is executed to set the .NET user secrets used in the application. To avoid this error, run the following PowerShell command:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass

Then re-run the azd up command.

Another possible error:

'pwsh' is not recognized as an internal or external command, operable program or batch file. WARNING: 'postprovision' hook failed with exit code: '1', Path: '.\infra\post-script\postprovision.ps1'. : exit code: 1 Execution will continue since ContinueOnError has been set to true.

The script postprovision.ps1 is executed to set the .NET user secrets used in the application. To avoid this error, manually run the script using the following PowerShell command:

.\infra\post-script\postprovision.ps1

The .NET AI apps now have the user-secrets configured and they can be tested.

Trying Generate Hiking Images sample

  1. From a terminal or command prompt, navigate to the semantic-kernel\05-HikeImages directory.
  1. From a terminal or command prompt, navigate to the azure-openai-sdk\05-HikeImages directory.

  1. It's now time to try the console application. Type in the following to run the app:

    dotnet run

    If you get an error message the Azure OpenAI resources may not have finished deploying. Wait a couple of minutes and try again.

Understanding the code

Our application uses the Microsoft.SemanticKernel package, which is available on NuGet, to send and receive requests to an Azure OpenAI service deployed in Azure.

The entire application is contained within the Program.cs file. The first several lines of code load secrets and configuration values that were set in the dotnet user-secrets for you during the application provisioning.

// == Retrieve the local secrets saved during the Azure deployment ==========
var config = new ConfigurationBuilder()
    .AddUserSecrets<Program>()
    .Build();

var config = new ConfigurationBuilder().AddUserSecrets<Program>().Build();
string endpoint = config["AZURE_OPENAI_ENDPOINT"];
string deployment = config["AZURE_OPENAI_GPT_NAME"];
string key = config["AZURE_OPENAI_KEY"];

The AzureOpenAITextToImageService service facilitates the requests and responses.

AzureOpenAITextToImageService textToImageService = new(deployment, endpoint, key, null);

Once the textToImageService service is created, we we provide more context to the model by adding a system prompt. A good prompt to generate images requires a clear description: what is in the images, specific color to use, style (drawing, painting, realistic or cartoony). The model will use this prompt to generate the image. To have the model generate a response based off the user request, use the GenerateImageAsync function, and specify the size and quality.

// Generate the image
string imageUrl = await textToImageService.GenerateImageAsync("""
    A postal card with an happy hiker waving and a beautiful mountain in the background.
    There is a trail visible in the foreground.
    The postal card has text in red saying: 'You are invited for a hike!'
    """, 1024, 1024);
Console.WriteLine($"The generated image is ready at:\n{imageUrl}");

Customize the prompt to personalize the images generated by the model.

Understanding the code

Our application uses the Azure.AI.OpenAI client SDK, which is available on NuGet, to send and receive requests to an Azure OpenAI service deployed in Azure.

The entire application is contained within the Program.cs file. The first several lines of code load secrets and configuration values that were set in the dotnet user-secrets for you during the application provisioning.

// == Retrieve the local secrets saved during the Azure deployment ==========
var config = new ConfigurationBuilder()
    .AddUserSecrets<Program>()
    .Build();

string openAIEndpoint = config["AZURE_OPENAI_ENDPOINT"];
string openAIDeploymentName = config["AZURE_OPENAI_GPT_NAME"];
string openAiKey = config["AZURE_OPENAI_KEY"];

// == Creating the AIClient ==========
var endpoint = new Uri(openAIEndpoint);
var credentials = new AzureKeyCredential(openAiKey);

The OpenAIClient class facilitates the requests and responses. ChatCompletionOptions specifies parameters of how the model will respond.

var openAIClient = new OpenAIClient(endpoint, credentials);

var completionOptions = new ChatCompletionsOptions
{
    MaxTokens = 400,
    Temperature = 1f,
    FrequencyPenalty = 0.0f,
    PresencePenalty = 0.0f,
    NucleusSamplingFactor = 0.95f, // Top P
    DeploymentName = openAIDeploymentName
};

Once the OpenAIClient client is created, we we provide more context to the model by adding a system prompt. A good prompt to generate images requires a clear description: what is in the images, specific color to use, style (drawing, painting, realistic or cartoony). The model will use this prompt to generate the image.

string imagePrompt = """
A postal card with an happy hiker waving, there a beautiful mountain in the background.
There is a trail visible in the foreground. 
The postal card has text in red saying: 'You are invited for a hike!'
""";

To have the model generate a response based off the user request, use the GetImageGenerationsAsync function, and specify the size and quality.

Response<ImageGenerations> response = await openAIClient.GetImageGenerationsAsync(
    new ImageGenerationOptions()
    {
        DeploymentName = openAIDalleName,
        Prompt = imagePrompt,
        Size = ImageSize.Size1024x1024,
        Quality = ImageGenerationQuality.Standard
    });

ImageGenerationData generatedImage = response.Value.Data[0];
if (!string.IsNullOrEmpty(generatedImage.RevisedPrompt))
{
    Console.WriteLine($"\n\nInput prompt automatically revised to:\n {generatedImage.RevisedPrompt}");
}
Console.WriteLine($"\n\nThe generated image is ready at:\n {generatedImage.Url.AbsoluteUri}");

Customize the prompt to personalize the images generated by the model.

Clean up resources

When you no longer need the sample application or resources, remove the corresponding deployment and all resources.

azd down

Next steps