ComfyUI-HunyuanVideoWrapper

Loading the LLM that reads your prompt

Sharper detail without cranking guidance

A cheaper way to get CFG's benefit on part of the network

How this pack fits a 13B video model on a normal GPU

Turning real negative prompting back on

Generating longer clips than the model was built for

Rewriting the system prompt behind your prompt

Turning latents into actual frames

A null prompt without loading the LLM

The leapfusion image-to-video route

Start-and-end-frame video with the Dashtoon LoRA

A nearly-free quality bump you should probably just leave on

Matching your image to what the model was trained on

The official image-to-video path in Kijai's Hunyuan wrapper

The first half of prompt-driven video-to-video

A cheap peek at your video before the real VAE decode

Baking a seamless loop into generation, not stitching one after

Applying a LoRA to only part of the model

How LoRAs actually load in kijai's wrapper

The front door to Kijai's Hunyuan wrapper

Blending two prompts into one generation

Turn an inverted video into a new one, same motion

Where the video actually gets generated

Spatio-temporal guidance without doubling your compute

Skip redundant steps, trade a little quality for speed

Skip the LLM, use native ComfyUI text encoding

Skip the text encoder, load a saved prompt

Stop re-running the text encoder every iteration

The CLIPTextEncode of kijai's Hunyuan wrapper

Text + reference images in one embed

Squeeze more speed out of the transformer

The piece that turns latents into pixels