CosmosModel<T>

Models & Types · AiDotNet.Diffusion.Video.WorldModels

NVIDIA Cosmos physics-aware world generation model.

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

For Beginners

Cosmos from NVIDIA is a physics-aware world generation model that understands real-world physics. It comes in Nano/Super/Ultra sizes and generates videos where objects obey gravity, collisions, and other physical laws.

How It Works

References:

  • Reference: NVIDIA Cosmos (2025)

Cosmos is NVIDIA’s world foundation model for physical AI development. Split into Nano (4B), Super (8B), and Ultra (14B) tiers, it generates physics-aware video and synthetic sensor data. Cosmos Predict 2.5 unifies Text2World, Image2World, and Video2World into a single architecture for consistent, controllable multi-camera video worlds.

Technical specifications:

  • Architecture: Physics-Aware DiT + World Generation + Multi-Camera
  • Latent channels: 16
  • Default: 120 frames at 24 FPS
  • Supports I2V: Yes | T2V: Yes | V2V: No

Example

var options = new DiffusionModelOptions<float> { LatentChannels = 16, DefaultInferenceSteps = 50 };
var noise = Tensor<float>.CreateRandom(new[] { 1, 16, 120, 135, 240 });
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 CosmosModel<float>(options: options))
    .Build(trainX, trainY);
var world = result.Predict(noise);

Constructors

Constructor Summary
CosmosModel(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, DiTNoisePredictor<T> predictor, TemporalVAE<T> temporalVAE, IConditioningModule<T> conditioner, int defaultNumFrames, int defaultFPS) Initializes a new instance of CosmosModel 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()