HunyuanVideo
HunyuanVideo by Tencent: an open video model with a language-model text encoder; version 1.5 is a separate, lighter family.
What it is
HunyuanVideo (late 2024) is a 13-billion-parameter text to video model; an image to video version followed. HunyuanVideo 1.5 (late 2025) is a newer, smaller model with its own files and graphs. Tencent's Hunyuan image models are separate families again.
Files and nodes
Load Diffusion Model with the HunyuanVideo model; DualCLIPLoader with clip_l and llava_llama3 (type hunyuan_video); Load VAE with hunyuan_video_vae; EmptyHunyuanLatentVideo; image to video adds a CLIP Vision encoder and its own conditioning node. Version 1.5 has its own template.
Settings that work
About 848×480 to 1280×720; up to about 129 frames at 24 fps (4n + 1); about 20 to 30 steps; guidance through a FluxGuidance-style node in the templates; euler with simple. Prompt: detailed sentences.
What fits
LoRAs made for HunyuanVideo (and for 1.5 separately). Wan LoRAs do not work.
How the app helps
Shown as HunyuanVideo on model cards. The node EmptyHunyuanLatentVideo, named after this family, is also used by Wan graphs. The lesson Video from text: HunyuanVideo 1.5 builds a HunyuanVideo 1.5 graph.