InstaFlowModel<T>

Models & Types · AiDotNet.Diffusion.FastGeneration

InstaFlow model for one-step text-to-image via rectified flow straightening.

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

For Beginners

Diffusion models follow curved paths from noise to images. InstaFlow first “straightens” these paths (making them nearly straight lines), then trains a model to jump from start to end in one step. It’s like straightening a winding road so you can drive directly to your destination instead of following curves.

How It Works

InstaFlow straightens the probability flow ODE paths of a pretrained diffusion model using the Reflow procedure, then distills the straightened model into a single-step generator. The straighter paths make single-step generation possible with minimal quality loss compared to multi-step generation.

Reference: Liu et al., “InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation”, ICLR 2024

Example

var options = new DiffusionModelOptions<float> { LatentChannels = 4, DefaultInferenceSteps = 1 };
var noise = 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 InstaFlowModel<float>(options: options))
    .Build(trainX, trainY);
var generated = result.Predict(noise);

Constructors

Constructor Summary
InstaFlowModel(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, UNetNoisePredictor<T> predictor, StandardVAE<T> vae, IConditioningModule<T> conditioner, int? seed) Initializes a new InstaFlow model.

Properties

Property Summary
IConditioningModule<T> Conditioner { get; }
INoisePredictor<T> NoisePredictor { get; }
IVAEModel<T> VAE { get; }
int LatentChannels { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()