MAGI1Model<T>

Models & Types · AiDotNet.Diffusion.Video

MAGI-1 video model with strong temporal coherence and multi-task support.

Inheritance: DiffusionModelBase<T> → LatentDiffusionModelBase<T> → VideoDiffusionModelBase<T> → MAGI1Model<T>

For Beginners

MAGI-1 is a ~10B parameter video model with strong temporal coherence, meaning objects and people stay consistent across frames. It supports multiple tasks (text-to-video, image-to-video) and produces smooth, natural-looking motion.

How It Works

References:

  • Paper: “MAGI-1: Autoregressive Video Generation at Scale” (Sand AI, 2025)

MAGI-1 is a ~10B parameter model from Sand AI achieving strong temporal coherence through autoregressive generation with latent space consistency mechanisms. It supports multiple tasks including text-to-video, image-to-video, and video continuation. The model uses a DiT backbone with enhanced temporal modeling.

Technical specifications:

  • Architecture: Autoregressive DiT + Temporal Coherence
  • Latent channels: 16
  • Default: 65 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, 65, 60, 106 });
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 MAGI1Model<float>(options: options))
    .Build(trainX, trainY);
var video = result.Predict(noise);

Constructors

Constructor Summary
MAGI1Model(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 MAGI1Model 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()