---
license: apache-2.0
base_model: Wan-AI/Wan2.1-T2V-14B
base_model_relation: adapter
library_name: peft
pipeline_tag: text-to-video
model_name: LongLive-Plug-Wan2.1-T2V-14B-cfg
tags:
  - lora
  - distillation
  - cfg-distillation
---

# LongLive-Plug-Wan2.1-T2V-14B-cfg

A LoRA adapter for [Wan2.1-T2V-14B](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B) that distills classifier-free guidance into conditional-only video generation, reducing the need for a separate unconditional inference branch.

This adapter does not provide few-step acceleration on its own.

This is an adapter and requires the corresponding base model.

## Usage

Use with the matching [few-step LoRA](https://huggingface.co/Efficient-Large-Model/LongLive-Plug-Wan2.1-T2V-14B-few-step).

Recommended LoRA weights: **few-step : CFG = 1 : 0.5**.

These are adapter weights, not the inference CFG scale.