Files
2026-06-07 20:43:31 -07:00

773 B

HunyuanVideo Native Support in Hanzo Studio

Date: 2024-12-20 Source: https://blog.hanzo.ai/p/hunyuanvideo-native-support-in-comfyui Category: model-release

Key Facts

  • HunyuanVideo is a 13-billion-parameter open-source video foundation model from Tencent
  • "Dual-stream to Single-stream" Transformer architecture fuses text and visuals
  • MLLM text encoder outperforms CLIP and T5 for instruction following and complex reasoning
  • Custom 3D VAE compresses videos into compact latent space preserving resolution and frame rate
  • Prompt Rewrite model with Normal Mode (improves intent) and Master Mode (optimizes composition/lighting)
  • Can generate both videos and still images (setting video length to 1)
  • video_hunyuan_video_t2v