Lora Training using only ComfyUI!!

everyone welcome to another AI fuzz video my name is Marcus and we are here to cover a huge huge node for comfyi users yes you can now train fully trained luras exclusively income for UI you heard that right no more Coya no more kaggle no more Google collab you do all the trig in com UI just by one single node yo let's do this all right so uh first we have first we had I'm gonna put the link uh to the GitHub for this node it is allora trading by Larry Jane mine five something but it'll be I'll link it into the description he also has um a group of nodes and also would create caption text files for each of the images so this is how it starts you're going to start with creating a data set right you need a data set data set of images because that's all it is data set is just a pro images that you will train AI on so here I have a group of sketches that we are going to make a sketch style um and this is all just a group of sketches I usually get maybe 25 at the minimum to 50 for this demonstration we're doing 24 um but these are all different typ of sketches their own PNG format they don't have to be the same size um and you put them all into a databased folder it's very important for this note you have to name this folder folder five umore s and the name of your lore whatever you're doing that folder means to be in a data data base folder because that's the higher that's structure that this node Works off of so once you have your data set in there first we're going to make the text captions for that so what I did was I have a brand new fresh install of company because you need to have a certain Scorch CU 121 version to work this if you try not it it's going to break so I have a separate Comfort UI installed that strictly has that version on it and it's strictly for training luras for me all right so be once you download the nodes from the GitHub it's going to be right here ljr Laura and you're going to have big nodes here so order to do the Lura caption save we're going to do the Lura caption load okay and in the middle we're going to do W the 14 tagger which is you can download that from the manager to be go cons just them loads and just type in WG it's right here so just down with that we start comfy and you'll have it so be your simple this is simple all you have to do is plug the image list to the image of image here and C plug the string through the text and plug the path to the path Hoops to to L to the path leave the main list open and then we need to do the pick it in here hit okay default settings are fine for the w for tiger and all you do is you prompt who I forgot to view the name list and then you prompt and I guess you do have to do the nameless sorry so it's GNA go in and basically now analyze each image in or data set it's going to create a text file that describes what's in each of the pictures so when it trains it kind of has it kind of knows kind of what's in each of the pictures to give it some idea of what it's looking at all right so it's all done training and if you look here it shows some of the text the white se's and some of the photos you can use really use any of these for trigger words for your Laura I usually just use the name of the Laura as a Trader in that always works so if we go into that file the database file we will see now there is text files for each of these pictures do that all right so the text files are there along with the corresponding images so now we have to bring up the magic node um so let's clear this put it go at node ljr e then backed in and this is the magic note it's the war training in comy UI huge I remember epic was telling me this is the biggest thing that she's ever seen you love VII and he uses it all the time so comeu I this these are these are the the the options here you have your checkpoint name what you want to be the base checkpoint you want to train it on you have it path to your images to your gay set back size MAX train EPO save every at EPO the output mean what you want to save the output uh luras two with the the name could Skip and the output directory so we are going to do this as we will do igri was a natural let's do the path to the folder now this is very important for this you don't put the name of the folder in the path you just put the name of the folder that it's in be just G the database and it would go into database and look at that folder and automatically know that's where the data set is keep the back size of one train OTS you can do any anything below anything between 40 and above is fine I do 50 and save every an eox essentially what it's going to do it's going to go through 10 images then save an EPO save a uh Laura from those indes then it's going to go to another 10 so they have 20 Imes to trained on and it'll save a Laura at that step and keep going 10 10 10 as to get to the 50 and you'll have a full Laura with no you'll see it with no numbers in the name it's just going to say whatever the name of lore is say teniser okay the dire desired name I'm just going to Nam it SK because that's what the folder is named after you can name that whatever you want clip skip will keep a two and the output directory will be your models Laura so it's actually going to make put it into your models Lura fly a folder for you this is this is ridiculous all right so while that's done so all you do chew prompt what Che prompt that's all you do so if we go into here and look at what it's going on it's going to go through it's going to find values uh and it's going to find the buckets with the images and it's going to make the buckets and it's going to list the steps it takes to make this Laura now I usually keep this up while it does it so I will keep it up for you by that's all but training it took whoops 23 20 one seconds a trained so then a little B basic workflow here upload theora and uh set of these clips and there it is you will do ep the grill was um the what do um s ah all right well C that prompt so this is the be the prompt with the brand new sketched sketch Mo that we just made trying to directly into comei try and get direct way ofi then this is at R not that for that sketched well uh let's do another one let's do four this time all right there goes the Treeline Street sketch I'm using RF Brandy sketch Laura and again you can really use whatever you want so long as you have the right images in her data set and the training goes well and again I did not use any Kya no no kaggle no Google CL nothing outside of compi it was all done within compi and there is the or right there so again like I said I will put the link in the description and I hope you guys like videos like this I will r youw with a few images at the end of some train lers that Lura that Ed did prior to him leaving so we'll see you Days Later another AI fuzz video Markus is out and the crew will be [Music] back [Music]

Channel: AIFuzz

Views: 5,949

Keywords:

Id: gt_E-ye2irQ

Length: 11min 14sec (674 seconds)

Published: Wed Feb 28 2024