VideoPoetModel<T>

Models & Types · AiDotNet.Diffusion.Video

VideoPoet LLM-based zero-shot video generation.

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

For Beginners

VideoPoet from Google uses a large language model (LLM) architecture for zero-shot video generation. The original paper treats video generation as a token prediction task. This implementation approximates its capabilities using a diffusion-based backbone for compatibility with the framework.

How It Works

References:

  • Paper: “VideoPoet: A Large Language Model for Zero-Shot Video Generation” (Google, 2024)

VideoPoet uses a large language model for zero-shot video generation, treating video as a sequence of discrete tokens. The model synthesizes high-quality video with matching audio from diverse conditioning signals. Won the ICML 2024 award for its approach to video generation. Note: This implementation uses a DiT-based diffusion approximation of the original LLM architecture.

Technical specifications:

  • Architecture: LLM-Based Token Prediction + MAGVIT-v2 Tokenizer
  • Latent channels: 16
  • Default: 128 frames at 8 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, 128, 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 VideoPoetModel<float>(options: options))
    .Build(trainX, trainY);
var video = result.Predict(noise);

Constructors

Constructor Summary
VideoPoetModel(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 VideoPoetModel 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()