Tencent-Hunyuan/HunyuanImage-3.0

Employs an autoregressive Transformer architecture departing from conventional diffusion-based approaches, unifying text and image understanding within a single framework. Features a Mixture-of-Experts design with 80 billion total parameters (13 billion active per token) for enhanced capacity, supporting both text-to-image and image-to-image generation with integrated reasoning capabilities. Integrates with HuggingFace Transformers, vLLM for accelerated inference, and provides distilled checkpoints optimized for efficient deployment with minimal sampling steps.

2,921 stars.