Complete Guide to SUPIR Enhancing and Upscaling Images Like in Sci-Fi Movies on Your PC

Рет қаралды 6,141

Күн бұрын

In this video, I explain how to 1 click install and use the most advanced image upscaler / enhancer in the world that is both commercially and open source available. The upscaler that I am going to introduce you is open source #SUPIR and the model is free to use. SUPIR upscaler is many times better than both paid Topaz AI and Magnific AI and you can use this upscaler on your computer for free forever. The difference of SUPIR vs #Topaz and #Magnific is like ages. So in this tutorial you are going to learn everything about how to install, update and use SUPIR upscaler on your personal computer. The video shows Windows but it works perfectly fine on Linux as well.
Scripts / Installers Download Link ⤵️
/ 99176057
Samplers and Text CFG (Text Guidance Scale) Comparison Link ⤵️
imgsli.com/MjU2ODQz/2/1
How to install accurate Python, Git and FFmpeg on Windows Tutorial ⤵️
• Essential AI Tools and...
Full DreamBooth / Fine-tuning Tutorial ⤵️
• Full Stable Diffusion ...
Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild : arxiv.org/abs/2401.13627
Authors introduce SUPIR (Scaling-UP Image Restoration), a groundbreaking image restoration method that harnesses generative prior and the power of model scaling up. Leveraging multi-modal techniques and advanced generative prior, SUPIR marks a significant advance in intelligent and realistic image restoration. As a pivotal catalyst within SUPIR, model scaling dramatically enhances its capabilities and demonstrates new potential for image restoration. Authors collect a dataset comprising 20 million high-resolution, high-quality images for model training, each enriched with descriptive text annotations. SUPIR provides the capability to restore images guided by textual prompts, broadening its application scope and potential. Moreover, Authors introduce negative-quality prompts to further improve perceptual quality. Authors also develop a restoration-guided sampling method to suppress the fidelity issue encountered in generative-based restoration. Experiments demonstrate SUPIR's exceptional restoration effects and its novel capacity to manipulate restoration through textual prompts.
0:00 Introduction to the very best SOTA image enhancement and upscaler open source app SUPIR
0:45 The new features that we recently added to the SUPIR upscaler app
2:52 How to install SUPIR on Windows and what are the requirements
3:13 How to check and verify your Python version
3:55 How to check and verify you installed FFmpeg and Git accurately
4:19 How to download installer scripts and start install
5:03 How to manually download all of the necessary models
5:25 How to download and install uGet to download models file extremely fast
6:19 How to make uGet downloads much faster than default settings
6:47 How to update your existing SUPIR app installation
7:03 How to upgrade requirements of installed SUPIR app
7:16 How to make a fresh install but move existing models into new SUPIR installation
9:01 How to verify your installation were smooth or had errors
9:48 How to extract downloaded all models zip file into the accurate folder
11:00 Don't make parent folder name as SUPIR when installing SUPIR app
11:22 How to start SUPIR app with accurate configuration and options
13:26 Which command line arguments, starting options are available
13:45 How to start with a different checkpoints / models directory
14:33 How to save starting options to single click start
15:03 How to use SUPIR app interface to enhance and upscale images like in science fiction movies
16:09 How to auto download checkpoints
16:19 How to use your DreamBooth / Fine Tuned model to upscale your AI images
16:42 Where we put Stable Diffusion models to use
17:16 What is Text CFG (Text Guidance Scale) and what it does
17:32 What is Prompt Style and what it does
17:54 How to use batch processing and do batch upscale / image enhancement
18:22 How to generate a comparison video
19:00 How to set / increase your virtual RAM memory to prevent RAM related errors
20:06 If your upscale is too slow that means shared VRAM is used and how to fix it
22:06 Where the generated images are saved by default
22:34 How to look at the metadata of the generated images to see all of the configurations and
settings
23:15 Comparison of the samplers and super important effect of the Text CFG (Text Guidance Scale)
25:40 How to use SDXL lightning version for many times faster upscale

Пікірлер: 76

@SECourses 3 ай бұрын

Scripts Download Link ➡ www.patreon.com/posts/99176057 Samplers and Text CFG (Text Guidance Scale) Comparison Link ➡imgsli.com/MjU2ODQz/2/1 How to install accurate Python, Git and FFmpeg on Windows Tutorial ➡kzfaq.info/get/bejne/Y7SagdxoxsyzgWQ.html Full DreamBooth / Fine-tuning Tutorial ➡ kzfaq.info/get/bejne/Ztpln5l0tZ_Jcps.html Scaling Up to Excellence: Practicing Model Scaling for Photo-Realistic Image Restoration In the Wild ➡ arxiv.org/abs/2401.13627

@reapicus557 Ай бұрын

I was able to integrate SUPIR into my recent dataset thanks to you. With the right touch, I was able to recover most of the cropped face images I had. This upscaler, in a few cases, felt like magic with how good the recovery was. I appreciate all your hard work! :D

@SECourses Ай бұрын

Thank you for comment. It is true this app is amazing

@VeganApps 3 ай бұрын

This is very, very useful! Thank you very much. Supir and your integration are amazing!

@SECourses 3 ай бұрын

thank you so much for the comment

@Cutieplus 3 ай бұрын

What is your recommendation for SDXL model ? Do any other SDXL model can give better result than Juggernaut-XL_v9 ? Thank you

@SECourses 3 ай бұрын

yes. For example RealVis XL 4 is great.

@MrJohnSnow 3 ай бұрын

Thank you very much for your work! Is it possible to download the models individually? It only failed on one part of the Lava model (the bigger ~10GB part), the rest downloaded successfully. Downloading all 45GB in one file takes forever, even with uget.

@SECourses 3 ай бұрын

You can download by looking those downloader files and download manually. You can also edit downloader files and delete downloaded ones

@user-qg4gx7be2j 3 ай бұрын

It was working fine before, I did a full 50 gigabyte upgrade, now after starting and loading the model I get an error : super().__init__(torch._C.PyTorchFileReader(name_or_buffer)) RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory After the error I clicked Windows_Update_Requirements After the update I run the same error again, I clicked Windows_Update_Version no updates, I run again and again the error. I uninstalled and reinstalled again (except models) and ran Windows_Update_Requirements and Windows_Update_Version, after that I ran suprin and again the same error. What could be the problem? Thanks!!!

@SECourses 3 ай бұрын

It looks like model file reading error. Perhaps your hard drive is corrupted? Or file was not downloaxed accurately? I would re-download

@user-qg4gx7be2j 3 ай бұрын

@@SECourses You're right, I reinstalled the model and everything worked for me. Thank you!

@SECourses 3 ай бұрын

@@user-qg4gx7be2j you are welcome

@WhySoBroke 3 ай бұрын

Looks like it is very slow on you 3090, even after loading the model. It seems Comfyui might be faster?

@SECourses 3 ай бұрын

It depends the settings but ye we have some speed problem right now. If you don't start with tiled vae still fast though.

@twilightfilms9436 3 ай бұрын

Great video! But I think I said it in your first video about the enhancer upscaler. The only way to know of it good or not or to test it with heavy motion blur scenes (or photos). Because it will be interesting to see what the app does with the halo around a moving object. And also how the model interpret, for example a leg running or a hand moving fast. I believe the Perturbed Attention Guidance does an excellent job interpreting objects in motion completely out of focus, that’s why debluring is extremely important in modern enhancers. Krea has managed this quite well and also provides a masking tool to retouch the parts that are not what you want. I’d like to see how this model reacts to extreme situations, but my test in Fal and Replicate were not good. But I can see you have more options in you UI, so it will be interesting to see results on moving objects. Thanks as always! Typos by iOS.

@SECourses 3 ай бұрын

i think it would be same as replicate. because i think the original supir is not trained for movies. the authors said they are trying to gather resources to train a movie upscaler. if you message me such image from discord i can test. by the way we added save and load presets just a moment ago

@JPnessYT 3 ай бұрын

Any chance this would work on an M1 macbook pro?

@SECourses 3 ай бұрын

not likely. but you can use on kaggle for free with our notebook or on massed compute very cheap we have installer for there

@gokhandanismend 3 ай бұрын

Hocam merhaba. Stable diffusion 3 için 512x512 çözünürlükte sonuçlar da görebilecek miyiz? En az 1024x1024 olacaksa da auto 1111 ve comfyui gibi local uygulamalarda bu sd3 checkpointlerini 1024x1024 için 8 gb vramli bir pcde kullanabilecek miyiz? Checkpointleri yayınlanmaya başladığı gibi orta veya düşük segmentli pcler için böyle bir kolaylıkla gelebilir mi sd3 acaba?

@SECourses 3 ай бұрын

Güzel sorular. Bence SD3, 8gb makinede çalışacak ama çözünürlükleri henüz bilmiyorum.

@gokhandanismend 3 ай бұрын

@@SECourses Pixeldojo adlı bir sitede bir deneme hakkıyla sd3'ü kullanma şansım olmuştu ve oluşturduğum görsel 1024x1024 çözünürlükteydi hocam ama tabi yine de nolacağı belli olmaz. 512x512 seçeneği olur mu diye sormamın sebebiyse sdxl'de 1024x1024 için daha fazla vram istendi ve ondan ağzımız yandı, şimdi acaba sd3 için de yine aynısı olur mu diye şüphelendiğimden dolayı hocam. Ancak min 1024x1024 çözünürlükte sd3 8 gb vramde de çalışacak olursa ondan ötesi bal-kaymak olur zaten😅

@SECourses 3 ай бұрын

bence 1024x1024 de 8gb da çalışır. ama hemen olmayabilir optimizasyonlar sonrası

@gokhandanismend 3 ай бұрын

@@SECourses İnşAllah 8 gb vramde çalışabilmesi için optimize edilmiş bir şekilde en tezinden sd3 checkpointlerini yayınlayabilirler hocam. Herkes için çok daha kolaylık sağlanmış olur ve erişilebilirliği de artar ama büyük ihtimal dediğiniz gibi biraz sürecektir muhtemelen o seviyeye gelebilmesi

@SECourses 3 ай бұрын

muhtemelen ben Kaggle notebook yaparım ona çıkınca

@markschrader7869 28 күн бұрын

Would you consider adding TensorRT acceleration for NVIDIA RTX GPU's to make Supir even more Super? Additionally, consider doing a video on how to enable Resizable Bar aka Re-BAR with your UEFI Bios & Nvidia app by the maker of your Nvidia RTX GPU. This will boost your Supir & Stable Diffusion rendering speed & performance.

@SECourses 28 күн бұрын

sadly super busy right now :(

@markschrader7869 28 күн бұрын

@@SECourses Understood, keep up the good work...

@SECourses 28 күн бұрын

@@markschrader7869 ty so much

@WhySoBroke 3 ай бұрын

If you could integrate the use of loras... would make it much better. Also batch processing from a specific folder from a preconfigured presets config file please

@SECourses 3 ай бұрын

Yes I plan to add some config save feature hopefully. LoRA usage we should add I agree

@user-xu8zy7ge1x 3 ай бұрын

@SECourses can you please do a video about flashFace - no one know how to install it

@SECourses 3 ай бұрын

thanks i saw it. gonna make a 1 click installer for it hopefully

@mocmocmoc13 3 ай бұрын

how to change path to v0F.ckpt v0Q.ckpt? I am also using comyfui supir, so would want to share the ckpt.

@SECourses 3 ай бұрын

edit options/SUPIR_v0_tiled.yaml and options/SUPIR_v0.yaml and give full path of your files it should work

@MrGTAmodsgerman 3 ай бұрын

I wish you would have showed a comparsion of your face upscale with the standard model vs your trained one.

@SECourses 3 ай бұрын

here tested for you on this image interesting result : imgsli.com/MjU3NzMx/0/1

@MrGTAmodsgerman 3 ай бұрын

@@SECourses Interesting. Thx. Crazy, on your face i expected major differences between those models. But maybe it would show more on a lower res image where less information is given. But it shows how good the base model already is. While on the clothing i can tell it make it way more accurate.

@VasilisKoutlis 13 күн бұрын

I'm stuck here.... Any suggestion? An generally to be honest I get a lot of errors so far which I solved but whatever I tried for this one didn't work. Traceback (most recent call last): File "C:\Users\User\Downloads\SUPIR_v49\SUPIR\gradio_demo.py", line 18, in import torch File "C:\Users\User\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\__init__.py", line 148, in raise err OSError: [WinError 126] The specified module could not be found. Error loading "C:\Users\User\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\lib\fbgemm.dll" or one of its dependencies. Press any key to continue . . .

@SECourses 13 күн бұрын

yes please do the followings. uninstall your existing python. install into C drive as shown in here and don't forget ffmpeg and c++ tools as well : kzfaq.info/get/bejne/Y7SagdxoxsyzgWQ.html also install supir into c:/supir_ai and also send me installation logs. if you dont know how to get installation logs check this : www.patreon.com/posts/108419878 if you message me from discord i can reply better there with images

@MrJohnSnow Ай бұрын

How can I change the path to my A1111 checkpoint folder?

@SECourses Ай бұрын

easy. edit the .bat file you are using and add --ckpt_dir "folder_path"

@MrJohnSnow Ай бұрын

@@SECourses I meant so that SUPIR uses my A1111 checkpoints, to save storage on my harddrive.

@SECourses Ай бұрын

@@MrJohnSnow yes it is SUPIR command. we designed that way in our app

@MrJohnSnow Ай бұрын

@@SECourses Do you mean the Windows_Start_SUPIR_app bat? Where in there do I put the command?

@SECourses Ай бұрын

@@MrJohnSnowedit that bat file and you will see how arguments are provided. Message me from discord I can help

@cemilhaci2 3 ай бұрын

Hocam linearcfg linearstage2guidance, s-churn, s-noise, stage1-stage2 noise değerlerinin etkisi, Prompts detaylandırmanın görsele etkisi ve advanced options kısmında görseli izin kalitesini artırıcı detaylara ilişkin bir video hazırlarsanız, render süreleri uzun olduğu için deneme yapmak çok uzun sürüyor. Böyle bir çalışmanın yararlı olacağını düşünüyorum. Tekrardan eğitim için teşekkür ederim

@SECourses 3 ай бұрын

O ayarlar ile ben biraz oynadım ama anlamlı sonuç elde edemedim. En çok Text cfg ve sampler etki ediyor gözlemim. Ama haklısınız onlara bir imgsli yapabilirim

@cemilhaci2 3 ай бұрын

@@SECourses bekliyoruz hocam

@SECourses 3 ай бұрын

@@cemilhaci2 teşekkürler

@AkArchitectsConsultancy 3 ай бұрын

Thank you from the bottom of my heart

@SECourses 3 ай бұрын

Thanks

@iseahosbourne9064 3 ай бұрын

Hey secourses, could you make a video on the latest and greatest video to ai methods and ai please 🙏

@SECourses 3 ай бұрын

hello. thanks for comment. AI has so many different areas. which areas are you interested in?

@iseahosbourne9064 3 ай бұрын

@@SECourses Just the ability to stylise a video in the most creative way possible, with the most control and flexibility. My goal is to make an animation with ai as realistic and quick as possible:

@SECourses 3 ай бұрын

@@iseahosbourne9064 i see. yes that is in my research field and i should do such video soon hopefully

@iseahosbourne9064 3 ай бұрын

@@SECourses Thats great to hear, cant wait!

@Thecroods923 3 ай бұрын

Download is paid ?

@SECourses 3 ай бұрын

Yes it is

@eaglecloud9223 3 ай бұрын

I think your thumbnail is kind of uncanny valley

@SECourses 3 ай бұрын

so you think it is interesting to click or not :D

@djivanoff13 3 ай бұрын

No module named 'gradio'

@SECourses 3 ай бұрын

That means you had error while installing. What is your python version?

@djivanoff13 3 ай бұрын

@@SECourses Python310.9, Python311.6

@SECourses 3 ай бұрын

ok you need to use 3.10.9 while installing. please make sure that it is default and do a reinstall

@bgtubber 3 ай бұрын

Awesome tutorial! Even though I already use it, I still learned a few new things. Yes, restoring parameters from the PNG metadata would be a nice improvement. Hopefully in a future update this will be possible. :)

@SECourses 3 ай бұрын

thanks hopefully coming

@lisaexplains 2 ай бұрын

Hey can you please tell me how many repeats to set for regularisation images in one trainer?

@SECourses 2 ай бұрын

they added a new option "samples" directly setting repeat. make it equal to the number of training images you have. or watch this tutorial : kzfaq.info/get/bejne/Ztpln5l0tZ_Jcps.html

@lisaexplains 2 ай бұрын

@@SECourses what about the training images? Do we set them as repeats or samples?

@SECourses 2 ай бұрын

@@lisaexplains repeat 1

@lisaexplains 2 ай бұрын

@@SECourses thank you. You are a Godsend to the AI community.

@SECourses 2 ай бұрын

@@lisaexplains thanks