Ojambo

Optimize Wan 2.1 on AMD Mi60 with Fedora

Fedora 43 AI Video Setup

Written by

Edward

in

Revised 1 min, 57 sec read

Introduction

Speed up Wan 2.1 video generation on Fedora 43 today. This guide uses stable-diffusion.cpp with full ROCm acceleration.

The AMD Instinct Mi60 powers this high performance setup. It features 32GB of VRAM and 24GB of RAM.

Software and Model Licenses

Visit https://github.com/leejet/stable-diffusion.cpp for the source code. This application uses the flexible MIT License for developers.

The Wan 2.1 1.3B model uses the Apache 2.0 license. Both licenses allow for personal and commercial video projects.

High Performance GPU Tweaks

The --diffusion-conv-direct flag increases your GPU efficiency. It bypasses slow layers to process video frames faster.

Using --cache-preset ultra keeps weights ready in VRAM. This prevents reloading data during the sampling process today.

The --vae-tiling option is vital for 832×480 resolution. It breaks frames into small blocks to save memory.

Your 24GB of system RAM supports the UMT5-XXL encoder. Quantizing to Q4_K_M makes this large model fit easily.

Optimized Command Line Usage

Use aria2c to download the large model files quickly. The multi-connection tool handles the GGUF files with ease.

Run the sd-cli with the --type q4_K parameter. This quantization maintains quality while boosting the generation speed.

The --cache-mode easycache setting reduces redundant math calculations. It reuses previous results to finish the video sooner.

Set --flow-shift 3.0 to improve the motion consistency. This ensures your lovely cat video looks smooth and natural.

Screenshot

Wan 2.1 1.3B Weights
File Manager Displaying Wan 2.1 1.3B Model Weights

stable-diffusion.cpp Options
Command Line Displaying stable-diffusion.cpp Options For Wan 2.1 1.3B T2V

stable-diffusion.cpp loading tensors
Command Line stable-diffusion.cpp Running Wan 2.1 1.3B T2V Loading Tensors

stable-diffusion.cpp Completed
Command Line stable-diffusion.cpp Running Wan 2.1 1.3B Completed

AI Generated Video
Video Player Playing Wan 2.1 1.3B AI Generated Video

Live Screencast

Screencast Of AI Generated Video Using Wan 2.1 1.3B T2V And stable-diffusion.cpp

Addendum Wan 2.1 1.3B T2V And stable-diffusion.cpp

Take Your Skills Further

🚀 Recommended Resources

Disclosure: Some of the links above are referral links. I may earn a commission if you make a purchase at no extra cost to you.

About Edward

Edward is a software engineer, author, and designer dedicated to providing the actionable blueprints and real-world tools needed to navigate a shifting economic landscape.

With a provocative focus on the evolution of technology—boldly declaring that “programming is dead”—Edward’s latest work, The Recession Business Blueprint, serves as a strategic guide for modern entrepreneurship. His bibliography also includes Mastering Blender Python API and The Algorithmic Serpent.

Beyond the page, Edward produces open-source tool review videos and provides practical resources for the “build it yourself” movement.

📚 Explore His Books – Visit the Book Shop to grab your copies today.

💼 Need Support? – Learn more about Services and the ways to benefit from his expertise.

🔨 Build it Yourself – Download Free Plans for Backyard Structures, Small Living, and Woodworking.

View all posts | Website

More posts