NVIDIA SANA 1-Click Installers & Gradio App: 4K Images on 8 GB VRAM (Patreon)
Downloads
Content
NVIDIA Labs SANA text-to-image model in an improved SECourses Gradio app; with all 4 VRAM optimizations, 1K images need 4 GB, 2K 6 GB and 4K 8 GB GPUs.
From NVIDIA Labs, with 1-click installers for Windows, RunPod, Massed Compute and a free Kaggle account notebook, and RTX 5000 series support since July 2025. Includes aspect ratio presets, the Diffusers pipeline and a resumable model download fix.
1-Click installers for NVIDIA Labs SANA model with an amazing Gradio APP developed by SECourses. Windows, RunPod, Massed Compute and Kaggle
Patreon exclusive posts index to find our scripts easily, Patreon scripts updates history to see which updates arrived to which scripts and amazing Patreon special generative scripts list that you can use in any of your task.
Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord
Please also Star, Watch and Fork our Stable Diffusion & Generative AI GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)
=======
Latest zip file : Sana_v18.zip
Full tutorial video : https://youtu.be/KW-MHmoNcqo
14 July 2025 Update
The app now supports RTX 5000 series on Windows, RunPod and Massed Compute
Newer models and features will be added later hopefully
13 January 2024 Update
Example images : https://www.reddit.com/r/SECourses/comments/1hzx1d6/it_is_now_possible_to_generate_16_megapixel/
New 4K Tutorial Video : https://youtu.be/GjENQfHF4W8
Now the APP will use Diffusers Pipeline and it has huge VRAM optimizations
You need to reinstall
The models will be downloaded into your Hugging Face cache folder when you first time generate something
Please make a fresh install
When you enable all 4 optimizations the VRAM usages are like below
Make sure shared VRAM is enabled because initial loading of the model need more VRAM
Enable VAE Tiling + Enable VAE Slicing + Enable Model CPU Offload +
Enable Sequential CPU Offload
1K (1024x1024) : 4 GB GPUs
2K (2048x2048) : 6 GB GPUs
4K (4096x4096) : 8 GB GPUs
Still in any case may work on your GPU test it
Just Enable VAE Tiling + Enable Model CPU Offload works great in many cases
29 December 2024 Update
Following aspect ratios added
1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, 9:21, 4:5, 5:4
When you select the aspect ratio, it will auto set the resolution according to either 1K model or 2K model
Windows_Fix_Model_Download.bat added which will resume download if your model download fails
27 December 2024 Update
With V6 update I have significantly optimized the pipeline
Now it will keep only model or VAE at the same time on the GPU, thus it will work way faster when decoding with VAE
Now works way faster on low VRAM GPUs
CUDA not found error fixed
Removed set CUDA_VISIBLE_DEVICES=0 from bat file
When changing model, wait until model is fully loaded
If you had V5, just using Windows update file should be sufficient but also replace your windows start bat file
Kaggle notebook updated
2K model is not working on Kaggle since GPUs are lacking BF16 capability
Pipelines and other code changes from main repository added to our APP
I have improved the Gradio APP significantly and now the Gradio APP supports the newest following model as well
Sana 1K (1024x1024)
Sana 2K (2048x2048)
The model will be loaded when you first time generate an image
The very first generation will be significantly slower than the consequent ones - totally related to their pipeline
I recommend make a fresh install - it will auto download newest models from the official repo automatically
======
Official page : https://nvlabs.github.io/Sana/
I have developed a better version Gradio APP than they have and made 1-click installers for Windows, RunPod, Massed Compute and a free Kaggle Notebook
It took me literally more than 12 hours of work :(
Speeds for generating per image as below for 1024x1024
Free Kaggle : 15 second
RTX 3060 : 9.5 second
RTX 3090 : 4 second
RTX 4090 : 2 second
Uses 9.5 GB VRAM but someone reported works amazing even on 8 GB GPUs
Windows Requirements
Python 3.10, FFmpeg, CUDA 12.8, cuDNN 9.7, C++ Tools, MSVC and Git
If it doesn't work make sure to below NEW tutorial and install everything exactly as shown in this below tutorial
How To Install And Use
Windows Requirements
Python 3.10, FFmpeg, CUDA 12.8, cuDNN 9.7, C++ Tools and Git
If it doesn't work make sure to below NEW tutorial and install everything exactly as shown in this below tutorial
RunPod Instructions
Please register for RunPod at : https://get.runpod.io/955rkuppqv4h
Select a GPU with a minimum of 24 GB VRAM, such as the RTX 3090.
Get a machine with minimum 48 GB RAM, getting multiple GPU machine better you will get more RAM
Choose a RunPod template:
RunPod Pytorch 2.2.0
runpod/pytorch:2.2.0-py3.10-cuda12.1.1-devel-ubuntu22.04
Follow Runpod_Instructions_READ.txt
Massed Compute Instructions
If you get 2x A6000 GPU it would work best with 96 GB RAM but not mandatory
Watch this instructional video to learn how to use Massed Compute:
Register for Massed Compute using the following link:
Important Notes:
Ensure each copy-paste action is accurate, as Massed Compute may not always copy-paste correctly.
Select an A6000 GPU or A6000 GPU ALT (if other one is not available) and use our code SECourses for huge discount.
Follow Massed_Compute_Instructions_READ.txt
Kaggle :
Upload Kaggle notebook and follow the instructions
Make sure verify phone so you can enable internet and select GPU
Example Kaggle install video : https://youtu.be/LeHfgq_lAXU
Related posts
Automatic1111 SD Web UI 1-Click Installer: SDXL, ControlNet, xFormers
Stable Diffusion WebUI Forge NEO: 1-Click Installer for Windows & Cloud
InstantID Gradio App: 1-Click Face Transfer With Xinsir ControlNet
All posts in this topic: AI Image Generation and Editing Apps · New to SECourses? Start Here: Best and Most Popular SECourses Posts