ComfyUI 2-Pass | Hires Fix: Your Secret Weapon for Stunning ComfyUI Images! 🖼️✨
Вставка
- Опубліковано 6 жов 2024
- The Hires Fix is all about creating an image at a lower resolution, making it bigger, and then using img2img (which is the same as txt2img in ComfyUI) to enhance it. It's a handy trick that can make your art look even better!
So, let's dive into it together in this video. Don't forget to subscribe and ring that notification bell, so you won't miss any of our future ComfyUI tutorials! 🚀🖌️
Hey thanks for these videos, I know there are a ton out there but I am just getting started and yours seem to have more refined and current setups. Will be learning from your videos going forward.
Glad to help
Your tutorial is perfection. Thank you!
A question, how do you change your ComfyUI background color? It look very neat in black.
This was EXACTLY what I needed, thank you!
Glad it was helpful!
Fantastic. BTW, I've tried a few upscale workflows and never got the quality of a lot of those over on reddit, until I tried your 2 pass method. Really cool. Thanks!
This is very similar to what I do. I use the lcm lora around 6-10 steps and .6 and .4 denoise for the 2nd and 3rd ksamplers, then run it through ultimate sd upscale. I've played around with just 2 ksamplers without the lcm lora and a .5 denoise with the ultimate sd upscale after that as well
good explained and is especially suitable for beginners, very nice, keep it up💪
Thank you 🙌
Enjoying the hires fix so much I decided to make a third pass of ksampler to see if it fixes the duplicated characters. Since I'm only on an RTX 3080 it's taking a while.
civitai.com/models/238891/hd-horizon-the-resolution-frontier-multi-resolution-high-resolution-native-inferencing try this perhaps
Thanks a lot for the tutorial, I did not know the latent way of upscaling.
I just tested the methods and just can not decide which method to use. How can I decide which method is more close to the original image and is more quality?
Maybe using pixel upscaling and running through sampler again is the way to go?
you can share the workflow
Can you upload this workflow or add a link? I don't know what I'm doing wrong but my upscaled results are wildly deformed
What is extension you have to manage workspace?
Here you go : github.com/11cafe/comfyui-workspace-manager
Some things aren't wired? Are the wires invisible?
Yeah they are. I'm using github.com/chrisgoringe/cg-use-everywhere
@@tech_bytebrain oh shoot - I forgot to come back and say I implemented your lesson. BEST results. I tried a few different options for upscaling and they all either added too much detail or no detail. Your version did what I needed.
@@wetterschneider Glad it helped!
Def not the same face/tattoo....
Lower the denoising strength or add ip adapter to keep likeliness (as seen in the other video for faceid v2)
You threw on the vastness of the Internet, another video about a method that everyone already knows, and who does not know, will be able to find in youtube hundreds of the same.
The questions on my videos suggest otherwise ¯\_(ツ)_/¯. Next videos will be more extended and complicated workflows.
yet you came all the way here in that vastness to comment 🤔