HunyuanDiTModel<T>

Models & Types · AiDotNet.Diffusion.TextToImage

Hunyuan-DiT model — bilingual (Chinese-English) DiT text-to-image model by Tencent.

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

For Beginners

Hunyuan-DiT generates images from Chinese or English prompts:

Key characteristics:

  • Bilingual: understands both Chinese and English prompts natively
  • DiT backbone: transformer-based denoiser (1.5B parameters)
  • Dual text encoders: CLIP-L/14 + mT5-XL for multilingual support
  • Multi-resolution training with aspect ratio bucketing
  • Human preference alignment via RLHF-like training

How Hunyuan-DiT works:

  1. Text goes through CLIP (visual) + mT5 (multilingual) encoders
  2. DiT transformer denoises with cross-attention to both embeddings
  3. Multi-resolution support through positional embedding interpolation
  4. VAE decoder produces final image

Use Hunyuan-DiT when you need:

  • Chinese text-to-image generation
  • Bilingual applications
  • Open-source multilingual alternative

How It Works

Hunyuan-DiT is Tencent’s bilingual text-to-image model that uses a DiT transformer backbone with dual text encoders (CLIP + multilingual T5) for both Chinese and English prompt understanding.

Technical specifications:

  • Architecture: DiT-XL with dual text conditioning
  • Parameters: ~1.5B (DiT backbone)
  • Text encoders: CLIP ViT-L/14 (768-dim) + mT5-XL (2048-dim)
  • Native resolution: 1024x1024
  • Latent space: 4 channels, 8x downsampling
  • Guidance scale: 6.0 recommended

Reference: Li et al., “Hunyuan-DiT: A Powerful Multi-Resolution Diffusion Transformer with Fine-Grained Chinese Understanding”, 2024

Example

var options = new DiffusionModelOptions<float> { LatentChannels = 4, DefaultInferenceSteps = 30 };
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 HunyuanDiTModel<float>(options: options))
    .Build(trainX, trainY);
var generated = result.Predict(noise);

Constructors

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

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