SUPIR is mind blowing - As you use, you get better of it, Comparison video generated via Gradio automatically - Juggernaut-XL_v9 works better with lower Text Guidance Scale
SUPIR is mind blowing - As you use, you get better of it, Comparison video generated via Gradio automatically - Juggernaut-XL_v9 works better with lower Text Guidance Scale
Download link : https://www.patreon.com/posts/supir-1-click-99176057
Comments (30)
Dr. Gözükara, I understand you train without captions typically, aside from the recent data analysis you put together in depth. Do you have an opinion, or thoughts otherwise, on using the Llava captions that were created when upscaling/improving image quality using SUPIR - to in turn train with Koyha_ss for example? The Llava captions are nice and detailed, would they cross over to being of actual benefit to training in a Dreambooth method? Thank you for all your time invested!
well i dont suggest to use AI processed images during training. you should try to get originally captured high detail images. but llava captioning is one of the best one. i suggest captioning while training a style but not a person
Is there a way to install the windows version for SUPIR from your zip file and not effect my current python installation or A111 install?
yep it wont affect. it generates its own venv so nothing will be affected. just use the latest zip file
I don't know exactly why, but the installer here does not work with Debian 12 out of the box. I did a fresh OS install to try and make sure I wasn't going insane, but no matter what I get the error below. The only thing I can think of is that Debian must be pulling a different/out of date package somewhere. Everything works fine on the PyTorch 2.1 Runpod template. qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/workspace/SUPIR_v45/SUPIR/venv/lib/python3.11/site-packages/cv2/qt/plugins" even though it was found. This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem. Available platform plugins are: xcb, eglfs, linuxfb, minimal, minimalegl, offscreen, vnc, wayland-egl, wayland, wayland-xcomposite-egl, wayland-xcomposite-glx. Aborted (core dumped) /usr/bin/xdg-open: 491: test: Illegal number: /usr/bin/xdg-open: 492: test: Illegal number: /usr/bin/xdg-open: 493: test: Illegal number:
yes your default debian system must be not setup properly with python, git and also cuda libraries and nvidia drivers
I would love to see a Creativity Strength like krea ai
maybe we can add good idea
Can I give one suggestion for future update please? 1. Can you add additional option for Turbo models? 2. Also for Lightning and Turbo models it usually recommended to use DPMSDE. Is it possible to add one? Anyway good job as always!
thank you so much good suggestions. can you remind me from discord as well in few days. currently eid so i may forget and i am super busy with family visiting
Hey, I own an RTX 3080, but since I usually do a lot of things for my university (I am in my final semester doing 3D and VFX for films and games), I am usually doing something on my computer. I decided to give Runpod a shot and I must say, I love the concept of still getting things done there while my main computer is free to do other things. I did try to run the Supir on RunPod and I just want to ask if the options that we see in the command prompt window as it launches should have specific choices made in them for RunPod, especially considering that the A6000 has much more ram than the 4080. What is the best way to make the most out of the speed of Supir on Runpod?
enable half precision but not tiling. it will give you most speed. also keep models in gpu all the time.
can someone explain to me how the upscale works? i have original image: 462x506 at upscale 1: it makes the image 1024x1120 at upscale 2: it makes the image 1024x1120 (same) at upscale 3: it makes the image 13864x1518
min upscale is 1024px at lower dimension. so whatever you put becomes lower dimension becomes 1024px
I have version 23.2.1. I should upgrade. Is this the problem? I emailed you.
hello. can you try fresh install of latest version and pay attention to every message? should be fixed now
Enter your choice (1-2): 1 Traceback (most recent call last): File "C:\Users\brand\Documents\SUPIR_v41\SUPIR\gradio_demo.py", line 13, in import einops ModuleNotFoundError: No module named 'einops' Press any key to continue . . . Then its shuts down. How do I satisfy this? Do I need to make a module or some unknown requirements?
hello. this happens when during initial installation you had error. can you tell me your python version? and also wait me to update to version 44 hopefully today. then you can try a fresh install and send me entire install cmd logs : monstermmorpg@gmail.com
Great song choice to go with the example :D
thanks :D we keep adding more stuff. Now version 38
Great tool, I really love it! A feature to save/load the config in the UI would be really great to quickly compare different settings and to save the best ones. Thanks for all your work!
hopefully we should add
Have a little suggestion to create a simple launcher - to have an ability tik checks for features user wanted to use and save, choose checkpoint and etc - without any need to 1,1,1 each launch in cmd)
we already have. rename the file direct_start_no_options .bat and edit it as you wish with starting options. then just use it to start immediately :)
Can you add a feature to compare the similarity of details between the upscaled images and the original images? I want to upscale 10 different image variations with random seeds and then choose the image that is most similar to the original before continuing with the next upscale. This will keep the quality as close to the original as possible. Thank you for your contribution to the community.
thank you but i don't know if there is such good way to check similarity. i will look
I see that you are very interested in the product and love Supir very much. Your updates constantly make me super excited every time I open my email.
awesome ty so much
Thank you, you're mega cool! I still need to learn how to process pics like you do! Good luck!
thank you so much for keep supporting me