LumiereModel<T>

Models & Types · AiDotNet.Diffusion.Video

Lumiere Space-Time UNet for single-pass 80-frame video generation.

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

For Beginners

Lumiere from Google uses a Space-Time UNet (STUNet) to generate entire video sequences in a single pass rather than frame-by-frame. This produces globally consistent motion across all 80 frames simultaneously.

How It Works

References:

  • Paper: “Lumiere: A Space-Time Diffusion Model for Video Generation” (Google, 2024)

Lumiere uses a novel Space-Time UNet (STUNet) architecture that downsamples in both space and time, generating 80 frames at 16 FPS in a single pass. This holistic approach produces more globally coherent motion compared to cascaded models that generate keyframes then interpolate.

Technical specifications:

  • Architecture: STUNet (Space-Time UNet) + Single-Pass Generation
  • Latent channels: 4
  • Default: 80 frames at 16 FPS
  • Supports I2V: Yes | T2V: Yes | V2V: No

Example

var options = new DiffusionModelOptions<float> { LatentChannels = 4, DefaultInferenceSteps = 30 };
var noise = Tensor<float>.CreateRandom(new[] { 1, 4, 80, 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 LumiereModel<float>(options: options))
    .Build(trainX, trainY);
var video = result.Predict(noise);

Constructors

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