InstantIDModel<T>

Models & Types · AiDotNet.Diffusion.Control

InstantID model — zero-shot identity-preserving image generation.

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

For Beginners

InstantID generates images of a specific person from one photo:

Key characteristics:

  • Single reference image: no training or fine-tuning needed
  • Face identity preservation: maintains facial features
  • Text prompt control: change style, pose, expression
  • Works with SDXL for high-quality output

How InstantID works:

  1. Face encoder extracts identity embedding from reference image
  2. IP-Adapter injects identity into cross-attention layers
  3. IdentityNet provides spatial face guidance (landmarks)
  4. Text prompt controls style, background, and expression

Use InstantID when you need:

  • Personalized image generation from one photo
  • Face-consistent character generation
  • Style transfer while preserving identity

How It Works

InstantID enables identity-preserving image generation from a single reference face image without any fine-tuning. It combines a face encoder, IP-Adapter, and IdentityNet to preserve facial identity while following text prompts.

Technical specifications:

  • Architecture: SDXL + IP-Adapter + IdentityNet
  • Face encoder: InsightFace (512-dim embedding)
  • Base model: SDXL U-Net
  • Control: face landmarks via IdentityNet
  • Resolution: 1024x1024

Reference: Wang et al., “InstantID: Zero-shot Identity-Preserving Generation in Seconds”, 2024

Example

// Create InstantID for zero-shot identity-preserving generation
var options = new DiffusionModelOptions<float>
{
    LatentChannels = 4,
    DefaultInferenceSteps = 30
};
            
// Generate images of a specific person from one reference photo
var faceInput = Tensor<float>.CreateRandom(new[] { 1, 512 });
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 InstantIDModel<float>(options: options))
    .Build(trainX, trainY);
var personalized = result.Predict(faceInput);

Constructors

Constructor Summary
InstantIDModel(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, UNetNoisePredictor<T> unet, StandardVAE<T> vae, IConditioningModule<T> conditioner, int? seed)

Properties

Property Summary
IConditioningModule<T> Conditioner { get; }
INoisePredictor<T> NoisePredictor { get; }
IVAEModel<T> VAE { get; }
int FaceEmbeddingDim { get; } Gets the face embedding dimension.
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)

Fields

Field Summary
int DefaultHeight
int DefaultWidth