IUnifiedVisionModel<T>

Interfaces · AiDotNet.VisionLanguage.Interfaces

Interface for unified vision models that can both understand and generate images.

Implements: IGenerativeVisionLanguageModel<T>

Known implementations: Chameleon<T>, JanusPro<T>, Janus<T>, OmniGen2<T>, SEEDX<T>, ShowO2<T>, ShowO<T>, Transfusion<T>

How It Works

Unified models combine visual understanding (captioning, VQA) and visual generation (image synthesis) in a single architecture. Approaches include:

  • Chameleon/Show-o: discrete visual tokens for unified autoregressive generation
  • Janus: decoupled visual encoding for understanding vs. generation
  • Transfusion: mixed autoregressive + diffusion loss in one transformer

Properties

Property Summary
bool SupportsGeneration { get; } Gets whether this model supports image generation (not just understanding).

Methods

Method Summary
Tensor<T> GenerateImage(string textDescription) Generates an image tensor from a text description.