SDXL Refiner Extension: simplify the image generation process!
Вставка
- Опубліковано 6 сер 2024
- Generating images with SDXL is now simpler and quicker, thanks to the SDXL refiner extension!
In this video, we are walking through the installation and use of the SDXL refiner extension.
We are going to compare images generated using different sampling steps and images generated using the img2img tab.
💬 Social Media:
[Discord] / discord
[Patreon] patreon.com/IntelligentArt?ut...
[Instagram] / lacarnevali
[TikTok] www.tiktok.com/@lacarnevali?i...
____________________________________________________________________
🤙🏻 Learn More:
/ membership
/ lauracarnevali
📰Article:
/ simplify-image-creatio...
📌 Links:
SDXL Refiner Extension: github.com/wcde/sd-webui-refiner
SDXL Paper: arxiv.org/abs/2307.01952
SDXL Base Model Download: huggingface.co/stabilityai/st...
SDXL Refiner Model Download: huggingface.co/stabilityai/st...
00:46 Download the SDXL Refiner Extension
01:40 SDXL REfiner Extension GitHub page
02:37 Where to find the Refiner in A1111
03:25 What is the tutorial about?
04:12 Generation Data from CivitAI
04:44 Generate an image with SDXL Base Model (No Refiner)
05:27 Generate an image using the SDXL Refiner Extension
05:56 What is the Refiner?
07:32 Image generated with Base Model vs. Refined Image
08:05 Results using different numbers of Refinement Steps
08:15 Increase the total number of Sampling Steps
09:21 Image generated using the img2img approach
10:20 Denoising strength
10:53 Img2img: more outputs
11:10 SDXL Refiner Extensions vs. Img2img
#aiart #stablediffusion #generativeart #stabilityai #stablediffusiontutorial
#sdxl #sdxlrefinerextension
Well done for the fast updates and news, thank you Laura :)
I can not see that using the refiner actually give you more detail in the image... But using Ersgan4x upscaler and writing "highly detailed" in the promt (sometimes it give to much detail leading to distortions) will give you a lot more details in the image specially combined with 60 - 80++ sampling steps.
Hi Laura, thanks for your video. 👍
Fun fact, you still can use the refiner while using any checkpoint instead of base SDXL and use 512X512...
Another lil fun fact, there is a "new" extension, the SDXL style extension. (77 styles) and third fun fact, you can use them with anycheckpoint and there is a command (a box to check) that allow you to generate an image in each style, so 77 images. (10 minits for my pc) with refiner on. I encourage you to use this super defined black belt prompt : "highly detailed", I also use add detail and epinoise offset loras... just look at the output and you'll probably end up just like me... like wtf who needs a prompt ? 😮
So many fun facts I'd think you're the royal jester :D
Or you can go batshit crazy with the ultradetailed prompt ;) It often create more of a acid trip image, even "highly detailed" is often to much and I just write "detailed"...
Hi, love the videos! Very helpful.
Muchas gracias!!
Cool Night here in Queensland Australia as we are in Winter, and I'm running Auto111 and don't need a heater as the 4090RTX is keeping me warm, no kidding PCs running AI now double as heaters go figure!
Same LOL it is winter here in Norway now -8c for a month... My RTX4090 runs at 60 - 80% power most of the time and if it is really cold I set it at 100% while mining with Kryptex 450w of heat spewing out of that thing lol :) Or I set it at making AI pictures in A1111 at (generate forever) setting :) .
Good job
Great
Hi, thanks for the video. May I know is 8gb vramenough for using SDXL? I keep getting out of memory errors. Thanks.
12GB VRAM - this is the recommended VRAM for working with SDXL.
@@LaCarnevali I see.. thank you.
Isn't it built in already? There is a Refiner tab inside the Generation tab in SD version 1.6 by default
yes, probably after I made the video ;)
Which gpu do you use and itel i5 is enough for sdxl
It is the GPU that matter most and it need +12GB video RAM (less will give all sorts of troubles and need workarounds...). It worked fine on my RTX3060 with 12GB Vram. I now have RTX4090 with 24GB Vram.
I usede too many stepse and percente for my imagese :(
My Automatic1111 doesn't have the clip skip, vae selection, etc at the top next to the model selection box. I should have the latest version since I added a line to the startup to update it every time it starts. So do you know where those extra options come from and why it may not be showing up on mine?
Settings, Ui, Scroll down to Quicksettings list. Type clip and it should show up then click on it. Same for rest, just type the name. UI settings is what you want to go to for any program to change the UI.
@@ATLJB86 Thanks.
refiner are not really good