One2345Model<T>

Models & Types · AiDotNet.Diffusion.ThreeD

One-2-3-45 model for single-image to 3D mesh generation in 45 seconds.

Inheritance: DiffusionModelBase<T> → LatentDiffusionModelBase<T> → ThreeDDiffusionModelBase<T> → One2345Model<T>

For Beginners

One-2-3-45 creates a 3D mesh from a single photo in ~45 seconds.

How One-2-3-45 works:

  1. Input image is encoded by CLIP into 768-dim features
  2. Zero123-based U-Net generates views from multiple angles
  3. Each view is conditioned on relative camera pose
  4. SparseNeuS reconstructs 3D mesh from the sparse multi-view images
  5. Texture is mapped from generated views onto the mesh

Key characteristics:

  • Two-stage: multi-view generation + 3D reconstruction
  • ~45 seconds total pipeline (fast for image-to-3D)
  • No per-shape optimization required
  • Produces textured meshes directly
  • Works from a single input image

When to use One-2-3-45:

  • Quick image-to-3D reconstruction
  • Single-image 3D mesh generation
  • When moderate quality at fast speed is acceptable
  • Prototyping 3D assets from reference images

Limitations:

  • Quality limited by sparse view generation
  • May struggle with complex occlusions
  • Texture quality depends on view consistency
  • Limited to object-centric scenes

How It Works

One-2-3-45 uses a two-stage pipeline: Zero123-based viewpoint diffusion generates multi-view images, then a SparseNeuS module reconstructs a textured 3D mesh from the sparse views without per-shape optimization.

Architecture components:

  • Zero123-based U-Net for multi-view generation (8 input channels, 768-dim CLIP)
  • SparseNeuS module for 3D reconstruction from sparse views
  • Standard SD VAE for view encoding/decoding
  • 8 input channels (4 latent + 4 view-conditioned)
  • DDIM scheduler for efficient multi-view generation

Technical specifications:

  • Stage 1: Zero123-based U-Net (8 input channels, 768-dim CLIP)
  • Stage 2: SparseNeuS reconstruction
  • Base channels: 320, multipliers [1, 2, 4, 4]
  • Input: 8 channels (4 latent noise + 4 view conditioning)
  • Pipeline time: ~45 seconds
  • Default point count: 4,096
  • Scheduler: DDIM

Reference: Liu et al., “One-2-3-45: Any Single Image to 3D Mesh in 45 Seconds without Per-Shape Optimization”, NeurIPS 2023

Example

var one2345 = new One2345Model<float>();
var mesh = one2345.GenerateMesh(
    prompt: "A ceramic vase with floral patterns",
    resolution: 256,
    numInferenceSteps: 50,
    guidanceScale: 3.0);

Constructors

Constructor Summary
One2345Model(NeuralNetworkArchitecture<T> architecture, DiffusionModelOptions<T> options, INoiseScheduler<T> scheduler, UNetNoisePredictor<T> unet, StandardVAE<T> vae, IConditioningModule<T> conditioner, int defaultPointCount, int? seed)

Properties

Property Summary
IConditioningModule<T> Conditioner { get; }
INoisePredictor<T> NoisePredictor { get; }
IVAEModel<T> VAE { get; }
bool SupportsMesh { get; }
bool SupportsNovelView { get; }
bool SupportsPointCloud { get; }
bool SupportsScoreDistillation { get; }
bool SupportsTexture { get; }
int LatentChannels { get; }

Methods

Method Summary
Mesh3D<T> GenerateMesh(string prompt, string negativePrompt, int resolution, int numInferenceSteps, double guidanceScale, int? seed)
ModelMetadata<T> GetModelMetadata()
Tensor<T> GeneratePointCloud(string prompt, string negativePrompt, int? numPoints, int numInferenceSteps, double guidanceScale, int? seed)