PaintByExampleModel<T>

Models & Types · AiDotNet.Diffusion.ImageEditing

Paint-by-Example model for exemplar-based inpainting using reference images.

Inheritance: DiffusionModelBase<T> → LatentDiffusionModelBase<T> → PaintByExampleModel<T>

For Beginners

Paint-by-Example fills in parts of an image using another image as a guide.

How it works:

  1. You provide a source image, a mask indicating the region to fill, and a reference image
  2. The reference image is encoded by CLIP into a visual embedding
  3. The U-Net takes 9 input channels (latent + masked image + mask) and denoises
  4. The filled region matches the style and content of the reference image

Key characteristics:

  • Based on Stable Diffusion 1.5 (512x512)
  • U-Net has 9 input channels (4 latent + 4 masked image + 1 mask)
  • Uses exemplar images instead of text for conditioning
  • Uses DDIM scheduler for deterministic inpainting results

How It Works

Paint-by-Example fills masked image regions using exemplar images as visual references instead of text prompts. The U-Net receives 9 input channels: 4 latent channels from the noisy latent, 4 channels from the masked source image latent, and 1 channel for the binary mask. The exemplar image is encoded via CLIP and injected through cross-attention.

Technical specifications:

  • Architecture: SD 1.5 U-Net modified with 9 input channels for inpainting
  • Conditioning: CLIP image encoder (768-dim) via cross-attention
  • Input channels: 9 (4 noisy latent + 4 masked source latent + 1 binary mask)
  • Output channels: 4 (latent space prediction)
  • Cross-attention dimension: 768
  • VAE: 4 latent channels, scale factor 0.18215
  • Noise schedule: Scaled linear beta [0.00085, 0.012], 1000 timesteps
  • Scheduler: DDIM

Reference: Yang et al., “Paint by Example: Exemplar-based Image Editing with Diffusion Models”, CVPR 2023

Example

var options = new DiffusionModelOptions<float> { LatentChannels = 4, DefaultInferenceSteps = 30 };
var input = Tensor<float>.CreateRandom(new[] { 1, 4, 64, 64 });
var trainX = Tensor<float>.CreateRandom(4, 8);
var trainY = Tensor<float>.CreateRandom(4, 2);
var result = new AiModelBuilder<float, Tensor<float>, Tensor<float>>()
    .ConfigureModel(new PaintByExampleModel<float>(options: options))
    .Build(trainX, trainY);
var prediction = result.Predict(input);

Constructors

Constructor Summary
PaintByExampleModel(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, UNetNoisePredictor<T> unet, StandardVAE<T> vae, IConditioningModule<T> conditioner, int? seed) Initializes a new instance of PaintByExampleModel with full customization support.

Properties

Property Summary
IConditioningModule<T> Conditioner { get; }
INoisePredictor<T> NoisePredictor { get; }
IVAEModel<T> VAE { get; }
int CrossAttentionDim { get; } Gets the cross-attention dimension (768 for CLIP image encoder).
int InputChannels { get; } Gets the number of U-Net input channels (9: 4 latent + 4 masked + 1 mask).
int LatentChannels { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> GenerateFromText(string prompt, string negativePrompt, int width, int height, int numInferenceSteps, double? guidanceScale, int? seed)
Tensor<T> ImageToImage(Tensor<T> inputImage, string prompt, string negativePrompt, double strength, int numInferenceSteps, double? guidanceScale, int? seed)

Fields

Field Summary
int DefaultHeight Default image height for Paint-by-Example (SD 1.5 native resolution).
int DefaultWidth Default image width for Paint-by-Example (SD 1.5 native resolution).