OpenSora2Model<T>

Models & Types · AiDotNet.Diffusion.Video

Open-Sora 2.0 commercial-level video generation model.

Inheritance: DiffusionModelBase<T> → LatentDiffusionModelBase<T> → VideoDiffusionModelBase<T> → OpenSora2Model<T>

For Beginners

Open-Sora 2.0 is a commercial-grade open-source video generator trained for under $200K. It produces high-quality videos competitive with proprietary models, using an improved architecture with better temporal modeling and data curation.

How It Works

References:

  • Paper: “Open-Sora 2.0: Training a Commercial-Level Video Generation Model in $200k” (HPC-AI Tech, 2025)

Open-Sora 2.0 achieves commercial-level video quality comparable to HunyuanVideo and Runway Gen-3 alpha while being trained for only $200k. Uses an improved STDiT backbone with enhanced temporal attention, better VAE, and data curation pipeline. Human evaluation shows it matches or exceeds leading commercial models.

Technical specifications:

  • Architecture: Enhanced STDiT + Improved 3D-VAE + Rectified Flow
  • Latent channels: 16
  • Default: 93 frames at 24 FPS
  • Supports I2V: Yes | T2V: Yes | V2V: No

Example

var options = new DiffusionModelOptions<float> { LatentChannels = 16, DefaultInferenceSteps = 30 };
var noise = Tensor<float>.CreateRandom(new[] { 1, 16, 93, 90, 160 });
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 OpenSora2Model<float>(options: options))
    .Build(trainX, trainY);
var video = result.Predict(noise);

Constructors

Constructor Summary
OpenSora2Model(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, DiTNoisePredictor<T> predictor, TemporalVAE<T> temporalVAE, IConditioningModule<T> conditioner, int defaultNumFrames, int defaultFPS, int? seed) Initializes a new instance of OpenSora2Model with full customization support.

Properties

Property Summary
IConditioningModule<T> Conditioner { get; }
INoisePredictor<T> NoisePredictor { get; }
IVAEModel<T> TemporalVAE { get; }
IVAEModel<T> VAE { get; }
bool SupportsImageToVideo { get; }
bool SupportsTextToVideo { get; }
bool SupportsVideoToVideo { get; }
int LatentChannels { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()