AuraFlowModel<T>

Models & Types · AiDotNet.Diffusion.FastGeneration

AuraFlow model — open-source flow-matching text-to-image model.

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

For Beginners

AuraFlow is a community-driven alternative to commercial models:

Key characteristics:

  • Flow matching: continuous-time formulation (not discrete DDPM steps)
  • DiT backbone: transformer-based denoiser
  • Open-source: fully open weights and code
  • T5 text encoder for strong prompt understanding
  • Competitive quality with commercial models

Use AuraFlow when you need:

  • Open-source flow-matching model
  • Alternative to SD3/Flux architecture
  • Research-friendly model

How It Works

AuraFlow is an open-source flow-matching model from Fal.ai that uses a modified DiT architecture with flow matching for high-quality text-to-image generation.

Technical specifications:

  • Architecture: DiT with flow matching
  • Text encoder: T5-XXL (4096-dim)
  • Resolution: 1024x1024
  • Latent channels: 4, 8x downsampling
  • Scheduler: Flow matching (Euler/midpoint)

Reference: Fal.ai, “AuraFlow v0.3”, 2024

Example

// Create an AuraFlow model for open-source flow-matching generation
var options = new DiffusionModelOptions<float> { LatentChannels = 4, DefaultInferenceSteps = 28 };
var noise = Tensor<float>.CreateRandom(new[] { 1, 4, 128, 128 });
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 AuraFlowModel<float>(options: options))
    .Build(trainX, trainY);
var generated = result.Predict(noise);

Constructors

Constructor Summary
AuraFlowModel(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, DiTNoisePredictor<T> dit, StandardVAE<T> vae, IConditioningModule<T> conditioner, int? seed)

Properties

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