StableDiffusionSharp 1.0.9

StableDiffusionSharp

Use Stable diffusion with C# only.

StableDiffusionSharp is an image generating software. With the help of torchsharp, stable diffusion can run without python.

Features

• Written in C# only.
• Can load .safetensors or .ckpt model directly.
• Cuda support.
• Use SDPA for speed-up and save vram in fp16.
• Text2Image support.
• Image2Image support.
• SD1.5 support.
• SDXL support.
• VAEApprox support.
• Esrgan 4x support.
• Nuget package support.

For SD1.5 Text to Image, it cost about 3G VRAM and 2.4 seconds for Generating a 512*512 image in 20 step.

Work to do

• Lora support.
• ControlNet support.
• Inpaint support.
• Tiled VAE.

How to use

You can download the code or add it from nuget.

dotnet add package IntptrMax.YoloSharp

Or use the code directly.

[!NOTE] Please add one of libtorch-cpu, libtorch-cuda-12.1, libtorch-cuda-12.1-win-x64 or libtorch-cuda-12.1-linux-x64 version 2.5.1.0 to execute.

You have to download sd model first. If you need a seperate vae, and you have to download it too.

If you want to use esrgan for upscaling, you have to download model from RealESRGAN_x4plus.pth

Now you can use it like the code below.

static void Main(string[] args)
{
    string sdModelPath = @".\Chilloutmix.safetensors";
    string vaeModelPath = @".\vae.safetensors";

    string esrganModelPath = @".\RealESRGAN_x4plus.pth";
    string i2iPrompt = "High quality, best quality, moon, grass, tree, boat.";
    string prompt = "cat with blue eyes";
    string nprompt = "";

    SDDeviceType deviceType = SDDeviceType.CUDA;
    SDScalarType scalarType = SDScalarType.Float16;
    SDSamplerType samplerType = SDSamplerType.EulerAncestral;
    int step = 20;
    float cfg = 7.0f;
    long seed = 0;
    long img2imgSubSeed = 0;
    int width = 512;
    int height = 512;
    float strength = 0.75f;
    long clipSkip = 2;

    StableDiffusion sd = new StableDiffusion(deviceType, scalarType);
    sd.StepProgress += Sd_StepProgress;
    Console.WriteLine("Loading model......");
    sd.LoadModel(sdModelPath, vaeModelPath);
    Console.WriteLine("Model loaded.");

    ImageMagick.MagickImage t2iImage = sd.TextToImage(prompt, nprompt, clipSkip, width, height, step, seed, cfg, samplerType);
    t2iImage.Write("output_t2i.png");

    ImageMagick.MagickImage i2iImage = sd.ImageToImage(t2iImage, i2iPrompt, nprompt, clipSkip, step, strength, seed, img2imgSubSeed, cfg, samplerType);
    i2iImage.Write("output_i2i.png");

    sd.Dispose();
    GC.Collect();

    Console.WriteLine("Doing upscale......");
    StableDiffusionSharp.Modules.Esrgan esrgan = new StableDiffusionSharp.Modules.Esrgan(deviceType: deviceType, scalarType: scalarType);
    esrgan.LoadModel(esrganModelPath);
    ImageMagick.MagickImage upscaleImg = esrgan.UpScale(t2iImage);
    upscaleImg.Write("upscale.png");

    Console.WriteLine(@"Done. Images have been saved.");
}

private static void Sd_StepProgress(object? sender, StableDiffusion.StepEventArgs e)
{
    Console.WriteLine($"Progress: {e.CurrentStep}/{e.TotalSteps}");
}