L8: Loras in ComfyUI - Comfy Academy
HTML-код
- Опубликовано: 18 фев 2024
- Let's learn how to use Loras in ComfyUi. Loras is a very effective way to create more styles and better images with the models you have. You can use Loras to create various styles, including specific clothing styles, actions, faces, and colors. You can even add more details to the image and get better skin details in your AI image.
Find my Workflow here: openart.ai/workflows/oliviosa...
#### 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 
Хобби
![Eminem - Houdini [Official Music Video]](http://i.ytimg.com/vi/22tVWwmTie8/mqdefault.jpg)
Thank you so much Olivio, you've taught me more than all the teachers put together. Your pedagogy, your clarity and your commitment to the subject make it hyper accessible and I'm very happy to learn so much, and it's all thanks to you.
Been so overwhelm lately and the least I can do is to follow this guide and I will (no more excuses)
And I already marked the date for the live in my calendar so I can catch the next one or watch the live streams.
Awesome teacher!
Great video Olivio....many thanks
Whats the reason for using different VAE than the checkpoints gives?
excellent tutorial, and the detail of the explanation
when part 9? hehe maybe animatediff? or IPa?
thanks a lot
@Olivio I would love to see an academy series on model/lora/ti training please :)
Hey Olivio, thank you for the awesome tutorials!! I have used your lessons almost exclusively so far for ComfyUI - and I've learned so much from you.
I am curious though, why your 'Learn ComfyUI' playlist only goes up to video 8, when there are many more videos.
The funny thing is, I can't seem to find where the other videos reside on your channel, and when I search for Lesson 9, 10, etc. it brings up nothing - though if I watch to the end of one, I will get a link for the next one.
Just wanted to let you know, because if I hadn't watched this video all the way to the very last second, I wouldn't have discovered there were more in the series - so if your metrics show that lesson videos after lesson 8 are getting significantly less hits, that might be why.
Keep up the great work as always, cheers! :)
Thank you! 🙂
You’re welcome 😊
Gracias Oliviooooooooooo
Hey Olivio, I created a ComfyUI node called "Lora Info" that helps with trigger words and example prompts. might be useful in your workflows
@OlivioSarikas Will you continue the series on Comfy Academy?
Was there a lesson on embeddings and I missed it or will you create one? :D
Great video, thanks:) I have one question. Can I call the LoRa just through the prompt so it's not activated when the prompt is not asking for it?
thx friend :) goof tutorial
Hi , thank you for your shares, I have a question : how to to creat a image have same ration with controlnet image ?
Hello Olivio,
I followed your excellent presentation step by step, but I cannot obtain models under the "safetensors" extension! In the "Models" folder there is only one .json file...
Do you have a solution to offer me?
Thanking you.
I wish usage information for every Lora would be saved within that file since it is quite tedious to re-visit each Lora's download page to look up the correct paramters used.
Yeah, especially trigger words and weights
you can actually place notes into each workflow that are saved with each image. that way they will always stay with you. including links
@OlivioSarikas yo!! Great tip!
@@OlivioSarikas Also if you right click the lora node and select "view info..." it will show you a load of that important information.
It's kind of unrelated. But I am hyped for X adapter to be implemented in comfy ui/auto111 . Using sd1.5 lora in sdxl is really great.
But using sd1.5 control net tile on sdxl is going to be heavenly 😊.
And of course, thank you for comfy ui tutorials.
I love the concept of workflow in CompyUI, but......The quality of images are so bad that I don't know why it is not working. I can't make the same quality as in A1111. I use the same models and parameters, samples. In ComfyUI all images are blurry and pixelated. I hope that it will get better. It is so time-saving to build all in one workflow and don't have to switch models and samples.
Would you recommend Comfy over ForgeUI?
if you want to build your own AI workflows that automate complex processes then yes :)
Great Video! Sorry I've not been more active recently. Had a potential business thing going on which then fell through and led to some health issues xDD But I should be back alive and well soon enough.
You're doing great work here. And it will pay off over time. You already have some of the most highly viewed tutorials on ComfyUI on YT, so the only way is up at your current pace. Even if it is a rocky start!
Forget comfi UI y XL model is not working with any lora properly
👋
9:25
nier anime style, anime screencap, megumin, 1girl, bare shoulders,
phoenixdress with gold ornaments, intricate dress, black cape, black gloves, black hair, blush, cape, choker, collarbone, dress, hair between eyes, hat, long sleeves, looking at viewer,
it was never explained what clipskip is??
I think i explain that in a different part. Sadly I can't explain every setting in every video. Clip skip means that CLIP as 12 layers of complexity in what the model understands from your prompt. For example if you write "a dog" there are many breeds of dog. So with clip skip it skips layers of complexity and does, in this example, not go into all the dog breeds it knows, but stays with a higher layer of dog breeds in understands better. that can help with Anime and Digital painting images as they are more abstract than photos and have less detail. That's a rather crude explenation, but i think it gives a basic understanding of what it does
@@OlivioSarikas very helpfull, thankyou for replying!
@@goodie2shoes my pleasure
I mean this as constructive criticism but this really should have been part of the first or second video. It's an issue with video instruction on comfyui. The same people who walked me through learning webui jump from "here's how you add the prompt window" to string theory then jump back to basics like adding loras sometime in the down the line. Again I mean this as constructive. Your regular videos are great it just seems to be a problem that once advanced users figure comfy out they're so enamored with what it can do that they forget the confused noobs just trying to do a basic render on a ui that uses recourses more efficiently.
Whilst these videos are still useful to see some example workflows it feels like they went off track a bit. From the first video I thought these were going to be a series on how to use Comfy and somehow we reached video 8 and only now get to using loras and still no mention of embeddings.
Most people thinking about using Comfy are likely to be people that started off with Auto1111. I doubt many people are starting off straight into using ComfyUI. I think a much better series woud be on how you can replicate the functionality of Auto1111.
Many people myself included see that you can do some really interesting stuff with comfy but all the things that are just a few simple clicks with 1111 requires a whole different workflow to be set up or loaded up in comfy and if you don't know how to set it up it just adds to the chore. If you have a whole bunch of loras even finding a lora takes longer because you have no thumbnails like in 11111. I can write a prompt, stick in a lora and an embedding and generate an image before I've even worked out how to add a lora in comfy. I think that's the main issue that puts people off using it.
It feels a bit like if your turned Photoshop node based, yes you could do some cool stuff you couldn't do before but every other basic function now becomes more complicated and tedius.
If you get stuck for content at some point I think a transition from auto1111 to comfy series would be really great. Generally just all the basic stuff, using loras, embeddings, inpainting, adding and using extentions like adetailer etc.
the series started out fine but then you quickly jumped over some important features without any explaination of what they do or why, such as embeddings in your negative prompts, clip skips etc.
Tutorials for beginners should explain what each thing does with examples, to help understand what's being done and why instead of just jumping to the end results or moving on to other topics