Wan-gguf Open-source Text-to-Video Generation Model - Suitable for Old and Low-end Machines, Efficient Inference with GGUF Files

Wan Gguf

Developed by calcuis

The GGUF quantized version of Wan Video is a text-to-video generation model suitable for older or low-end machines, supporting efficient inference via GGUF files.

Text-to-Video EnglishOpen Source License:Apache-2.0 #Text-to-Video Generation #Low-Resource Optimization #GGUF Quantization

Downloads 26.46k

Release Time : 2/26/2025

Model Overview

This model is based on Comfy-Org/Wan_2.1_ComfyUI_repackaged and is a text-to-video generation model that supports quantized inference via GGUF files, making it ideal for resource-limited devices.

Model Features

GGUF Quantization Support

Supports the GGUF file format, making it suitable for older or low-end machines, improving inference efficiency.

Multi-Component Support

Supports various components (e.g., t5xxl-um encoder, VAE, etc.), offering flexibility for different needs.

CPU Offloading

Supports CPU offloading to optimize resource usage without affecting inference speed.

Model Capabilities

Text-to-Video Generation

Negative Prompt Support

Efficient Inference

Use Cases

Creative Content Generation

Winter Landscape Video Generation

Generates videos of animal movements in winter landscapes based on text descriptions.

samples/ComfyUI_00007_.webp

Anime Character Animation Generation

Generates turning animations of anime characters based on text descriptions.

samples/ComfyUI_00008_.webp

Artistic Creation

Glass Flower Blooming

Generates videos of glass flowers blooming based on text descriptions.

samples/ComfyUI_00010_.webp

🚀 gguf quantized version of wan video

This project provides a gguf quantized version of the wan video model, enabling users to easily integrate and run it in ComfyUI. It simplifies the deployment process and offers a smooth experience for generating videos.

🚀 Quick Start

Installation

Drag gguf to ./ComfyUI/models/diffusion_models.
Drag t5xxl - um to ./ComfyUI/models/text_encoders.
Drag vae to ./ComfyUI/models/vae.

screenshot

Workflow

For the i2v model, drag clip - vision - h to ./ComfyUI/models/clip_vision.
Run the .bat file in the main directory (assuming you are using the gguf pack below).
If you choose to use the fp8 scaled umt5xxl encoder (applicable to any fp8 scale t5), please use CPU offload (switch from default to cpu under device in gguf clip loader; it won't affect the speed). By the way, it works well for both gguf umt5xxl and gguf vae.
Drag any demo video (below) to your browser for the workflow.

screenshot

✨ Features

Model Deployment

The project simplifies the deployment of the gguf quantized model by providing clear instructions on where to place different model components.

Compatibility

It is compatible with various models, encoders, and VAEs, and offers solutions for different usage scenarios, such as using the fp8 scaled encoder.

Alternative Running Method

You can also run it with the gguf - connector using the command ggc v2.

screenshot

📚 Documentation

Review

pig is a lazy architecture for the gguf node. It applies to all model, encoder, and vae gguf files. If you try to run it in the comfyui - gguf node, you might need to manually add pig to its IMG_ARCH_LIST (under loader.py). It's easier than editing the gguf file itself. By the way, model architectures compatible with comfyui - gguf, including wan, should work in the gguf node.
The 1.3b t2v gguf works well, which is suitable for old or low - end machines.

Reference

Base model from [wan - ai](https://huggingface.co/Wan - AI/Wan2.1 - T2V - 14B).
ComfyUI from comfyanonymous.
Pig architecture from connector.
gguf - node ([pypi](https://pypi.org/project/gguf - node)|repo|pack)

📄 License

This project is licensed under the Apache 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご