L6: Model Switch and Masking in ComfyUI - Comfy Academy
Тәжірибелік нұсқаулар және стиль
Model Switching is one of my favorite tricks with AI. We render an AI image first in one model and then render it again with Image-to-Image in a different model. This allows us to use the colors, composition, and expressiveness of the first model but apply the style of the second model to our image.
Then we also explore Image Masking for inpainting in Comfyui, a hidden gem that is very effective.
Find my Workflow here: openart.ai/workflows/oliviosa...
Other Lesson Videos:
• L1: Using ComfyUI, EAS...
#### Join and Support me ####
Buy me a Coffee: www.buymeacoffee.com/oliviotu...
Join my Facebook Group: / theairevolution
Joint my Discord Group: / discord
AI Newsletter: oliviotutorials.podia.com/new...
Support me on Patreon: / sarikas
Пікірлер: 35
Amazing stuff, Olivio! Former A1111 dabbler going to ComfU! So thankful for your tutorials.
Thanks Olivio!You are appreciated by the Ai Art Community!
Thank you for your videos. They're very helpful and everything is explained quite well. Easy to follow along.
hey Olivio, today I decided to do all your workflows from your ComfyUI academy. I know comfyUI well, but there were some voids in my knowledge.. Soo, now the further courses are waiting for me.. :D Thank you a lot!
thank you! really stunning
Thanks for your lesson, it helps a lot, thanks !
Awesome tutorial especially for newbie! I can't thank you enough. 🤩But I don't understand why you need to convert the positive text to input and make a CR Prompt Text? I created a separate workflow without the CR Prompt Text, just use the same Positive Clip Text Code and link to 2 different Ksampler, I am able to achieve the same result. Appreciate anyone here can enlighten me 🙏
Very cool explanation~!
Thank you it was indeed beginner friendly! I did some comparison recently of tools accesible via hugging face.
Thanks!
Thank you! ♥♥♥♥
I don't quite understand why you would use CR prompt text instead of normal text incode for the positive nodes
@ethanhorizon
Ай бұрын
Same here, I am also wondering why.
I fell in love with the cover art, I'm wondering if there's a possibilty that you can maybe share the image. The elf girl looks so cute 🧝🏻♀💚
why did we have to use the skip clip layer? -2 is one from the end, just wondering why we don't include the end layer for the initial revAnimated image?
adamsınn olivio
This workflow is useful for a base model+refiner workflow?
This was really helpful! I have two questions regarding this: 1) Why does the first model need a clip skip of -2? What is that doing? 2) Would it be possible to start with a SD1.5 model (e.g. rev animated) and then model switch to an SDXL model? If I missed the explanation to either question in a previous question my apologies!
@kick851
4 ай бұрын
I actually don't know what is it doing but you can ask the civitai ai that, it did help me a lot
@maciejbala477
Ай бұрын
i've said it in another reply as well but actually the clip skip is just a model thing/model family thing. ReV Animated I assume is part of the NAI model family, trained from the NAI v2 anime model for SD1.5, which is a bit overtrained so it's recommended to set clip skip to -2 to skip the last layer and use 2nd-to-last. Another one that recommends it is Pony Diffusion which is a very popular model family for SDXL these days, so same reasoning there. Otherwise if not related to these, just keep clip skip as is unless the model page says otherwise
Great video as always! but I have something that I'm not quite sure about - I keep getting the same effect of the two model combination weather I use the Clip set last layer node or not, So I am not completely sure what is its use, Can anyone explain? Thanks!
@maciejbala477
Ай бұрын
some particular model types (particularly the NAI SD 1.5 model family, so basically anything anime-related, and the Pony Diffusion based SDXL stuff, might be others) recommend the -2 value. An opinion I've ran into was because they were a bit overtrained so actually skipping a generation layer/step would give the better result, hence the value.
👋
What is the purpose of the clip skip? Is that only for certain models?
@OlivioSarikas
26 күн бұрын
yes, clip skip is only for some models. mainly models that do digital painting style and anime. but you need to check the model info on civitai to see which model needs what clip setting
How do you get rid of red windows?
I realized that models can be combined, or rather line up steps one after another with different models, ok, but I didn't understand why you would make a drawn image first and then turn it into a more photo realistic image? Hmmm. Maybe there's some more logical useful usecase for combining models? How is that better than doing photorealism on one model at a time?
@wVAjaxVw
Ай бұрын
3:58 I think it’s because the first model is better at doing hands/hand poses and colors, that can then be translated into realism. My guess is starting with realism might give duller colors, or weird hand placements.
Jesus can we stop generating "beautiful women"
Honestly man I don't mean to hate on you or anything, but this is all very basic and I feel like I've seen this exact workflow idea on your channel multiple times already. I understand this is beginner oriented, but how about still throwing in some stuff for more advanced users? I also feel like even for beginners, you could improve your visual presentation by coloring the nodes and organizing the worflow better.
@JonRowlison
4 ай бұрын
I can't disagree. He *DID* however mark this series as a beginning series for ComfyUI... so it's not a surprise that he's covering things he's covered before, but in a progressive series of lessons for newbs.
@carstenli
4 ай бұрын
Why would you even watch it, if you are an experienced user?
@iulianpartenie6260
4 ай бұрын
Move on!
@poipoi300
4 ай бұрын
@@carstenli This very idea he's shown in this video is something I've learnt from him a while ago. I watch every video with the hopes of learning something new. This being oriented at beginners, I didn't have that as an expectation but still a hope. New information and ideas have the potential to be anywhere for anyone. You should never assume you can't learn anything from something/someone. BTW "why watch if x" is old and rarely relevant, just letting you know. You're not addressing the criticism, you're just running away.
Graciaaaaaaas Oliviiiioooooooooooooo