ComfyUI HF Hugging Face Repo Node

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
hey everybody welcome to another AI fuzz video and today we're going to be covering a special node uh called comfyi HF which is basically a gpt2 text expander node that will actually add on to your prompts to give you a better well hopefully a better result cool thing about this node it also um tra you can actually use hugging face repositories on so it actually has models that were tested and you can use those models that were trained on same mid Journey or Lexica or cart and you can actually use those models and have it add on extra prompts and give you a different result than just the regular base prompt so this is kind of cool to play with it's an experimental thing you can see you might be able you might like you might not but uh when you open comfi you want to go to the comfi manager install custom nodes and you want to do a search for for HF that's it HF and it's going to be on the bottom by uh brami it's the comfyi knowde to work with hugging face repositories you want to click install and then uh you wanted to um restart compi I'm going to go to their hugging face which is right here I'll link this in the description and uh it shows you the models that were tested it's a gpt2 node loads the given GP model to generate a prompt and there's many different ones here that were tested these are default one the the gustava magic prompt stable diffusion uh you also have the uh text image prompt generator trained on M Journey prompts some were actually trained on lexic Art modal art and there a few more so I'm going to show you how I can use these to add more prompts to your current positive prompt so let's go ahead and that was a tongue twister wasn't it so let's go ahead and build this real quick um like uh always I'm going to keep load this workflow into the uh patreon and you guys can download it uh so we're going to load a checkpoint first okay so we're going to do a we're going to do a um comparison between the regular prompts and the new gp2 2 so this is going to be the regular I'm just going to color this like that this is the that didn't need to change the same color negative and positive okay uh we are going to go into a k sampler this is going to be the case sampler for the regular prompt okay the model into the model and I will choose let's just choose ah let's do um [Music] um let's do the realistic Vision all right and then this one will go to a vae decode which will then go into a preview image and again this will be the uh normal prompt just going through the regular right okay so I got to connect the vae to vae over here all right so now I'm going to do another case sampler so we can run in the new prompt okay the negative is also going to go into the negative here and instead of this regular Pro positive we're going to search for GPT and it's going to give us this gpt2 text generator if you need to see it right here so this has where you can put deposit prompt but here is where you can load those trained models to give you more prompts so if I go back to the hugging face let's say I want to do this mid Journey one I just go to that I copy go to there and I paste right there and I hit okay okay so that is how you load those up okay so I can hit the clip to here and then for here I'm going to do the conditioning right here here model going come back from here and the Vie as well I'm sorry whoops not that there I need a empty latent for this one did I do one for this no I did not so I need another one for [Music] here what I can't spell today I need another one for here I'm GNA make a little room because it's getting a little crowded Okie doie um okay so I'm going to just make this I'll make it 512 since it's a SD okay and then I need to do a VA decode for here and a preview right here so I'm going to hook up the vae to the vae there and this preview I'm going to set up right here okay all right so I got to figure out the prompts now so I'm going to put this regular prompt I'm going to do um female I'm going to do 24 year old female closeup um uh blonde uh realistic 8 K textured skin okay and here I'm going to do blurry ugly deformed smudge and malformed okay so that is the regular prompt up and here is the gpt2 text I'm going to actually copy this and put it into here now in order to see what the extra uh prompts it adds I'm going to do a show text we can do it right [Music] here you connect it to the string text and it'll show you what it's going to add to this so I have that all set up I have both in here and then we are going to Q prompt and we are going to get two different images let me just set up the samplers to be the same I'll do four as the seed uh let's do 40 steps and let's do seven CFG let's do sde and caras I'll do the same thing for this [Music] one 40 7 s cross all right so everything is about the same everything is set up and we're going to hit the Q prompt here all right so here is the extra prompts that it added as you notice we've had we had this prompt and it added character concept art art station cinematic detailed octane HD photo realistic and here are the here are the results so the first one is actually the regular prompt with the regular positive which is it's pretty nice and the second is the gpt2 prompt which is also pretty nice they both look good um and actually going to T I'm going to name this2 just so I know they both look pretty good but it's a different image obviously based on those extra prompts that it added all right let's try a different model as well as a different repo from hugging face and let's try this one for inst so I don't know what this is going to do I'm going to copy that stick that here okay and then let's try a mail this time and not a closeup so let's do um 40y old male oops let's do medium shot uh I'm not doing hair color we will let that decide and then we will put uh at a construction site which is barely stereotypical but maybe he's just there to deliver coffee I don't know okay so it gave me these two opt these two images uh the first one is the regular the second one is the gpt2 I still have a female for gpt2 for some reason trying to see why not sure did I leave anything for female no okay well I mean in this case it screwed that up so I think I'll take the regular image over that so let's try another one to see what comes up now that's a pretty decent uh prompt that I put in there I'm going to see what it adds if it adds any anything at all I think it may uh because we're at the D Sharma stable diffusion uh stable diffusion model uh from here okay so it added by onean Luna by B not sure if that was just a typo or what but let's see what it does and dream shaper eight and let's see what the output it gives us interesting okay so it gives us both this kind of in art style and and this one face is kind of messed up here but it added that extra prompt which made the [Music] difference all right so these are the results fairly similar but this is the regular prompt and this is the um the ones from the dolly repo not bad they look they look fairly similar so why don't I actually compare different repos against each other so that might work out better so let's do I'm going to add two more to get this started so okay so we're going to do a um let's do this we're going to alt alt alt I'm holding alt down guys I'm not just saying alt to say alt but um yeah we're going to do another one real quick okay oh all right I'm going to go here the model could use a reroute there but I will not this goes into the positive and the negative goes into the negative [Music] oops and the L image it connected all right so this is one this is another I will it's over [Music] here all right so we're going to do this prompt all right [Music] oops okay going to take this copy going to put it to here oops paste all right that's one all here's another one paste and the fourth paste all right so we're going to keep it dream shaper now we're going to load the repos okay so let's check all right let's uh get some repos [Music] here all right we'll do this one okay that's one we'll put that one right here and then let's do another one let's do this well that's a safe one cuz there's an not safe for work one all right and we do one more and let's grab um this one is that right here okay so now since they're all different they all they they all going to add different uh prompts all right so I'm going to keep everything else the same and then we will see what happens with [Music] that okay so this is what it came up with this is the regular prompt with nothing added this is the gpt2 prompt from I think it's the it's uh yes from the the maor trained on the M Journey prompts [Music] this uh okay is that a dead octopus on a tree branch or something not sure what happened there and this one is the last one which added uh some it is the it's this one the r l HF Crum Bloom and it added a lot of um word a lot of prompts to that so you can mess around and play and compare the different model repos those are the these are just the ones that are tested you can try probably more you know you can add more to this list or you can just mess around with it see what you can get but um again I will put the link to where to get all this information from the repo the GitHub will be on the the description and if you guys have any other questions let me know if you like these this video please subscribe and uh like and comment uh takes me a while I will to to it takes me a little bit time to make these videos and to come up with something at least that is going to work for you guys or something that's kind of interesting that you guys can play around with I have a patreon page you want to show support there I will be releasing this workflow on the patreon it's just going to have the one and the two though but you can add as you see it's pretty easy to add you can add more on there if you want to so I'll see you guys later on another EP AI fuzz video bye
Info
Channel: AIFuzz
Views: 752
Rating: undefined out of 5
Keywords:
Id: 7mqm-jOD5hk
Channel Id: undefined
Length: 15min 37sec (937 seconds)
Published: Mon Feb 19 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.