Running SANA Text-to-Image locally on Windows

SANA is the new, blazing-fast Text-to-Image model from NVIDIA, where you can create images with a size up to 4096x4096 pixel in just a few seconds. While an Online-Demo has been out for a little while, it's now possible to run it on a local machine with an NVIDIA-GPU. By this time, it's only meant to be for Linux, but in this tutorial I will show you how to run it on Windows, using the built-in Windows Subsystem for Linux (WSL). The installation process will be explained step by step and you can find the commands for each step further down below. Hope it can be helpful to you! Chapters: ========= 00:00 Intro, Requirements 00:55 Installing the Windows Subsystem for Linux (WSL) 01:29 Cloning SANA from Github 02:41 Installing Conda on WSL 04:16 Create and activate the virtual environment for SANA 04:36 Installing Cuda, Torch, xFormers and SANA libraries 05:20 Create and install Huggingface-Token 06:30 Starting SANA for the first time 07:10 Creating images with SANA 08:03 How to start SANA after the installation Useful Links: ========== SANA Paper: https://nvlabs.github.io/Sana/ SANA Online-Demo: https://nv-sana.mit.edu Installation Process: ================ 1.) Installing the Windows Subsystem for Linux Link: https://learn.microsoft.com/en-us/windows/wsl/install in Windows Powershell type: wsl --install (choose Ubuntu Standard installation) 2.) Open WSL in the directory, where you want to install SANA in the WSL-shell update & upgrade WSL: sudo apt update sudo apt upgrade 3.) Clone SANA repository from Github: Link: https://github.com/NVlabs/Sana git clone https://github.com/NVlabs/Sana.git cd Sana 4.) Install Conda for Linux https://www.anaconda.com/download save file: https://repo.anaconda.com/archive/Anaconda3-2024.10-1-Linux-x86_64.sh into your SANA directory and run it ./Anaconda3-2024.10-1-Linux-x86_64.sh copy installation path (in my case: home/gerald/anaconda3) reload wsl-shell: source ~/.bashrc initialize conda (use your own installation path