LoRA of Elizabeth Banks
If you wonder why this LoRA was uploaded please check my new article and be amazed: https://civarchive.com/articles/1721
I'm creating Dreambooths, LyCORIS, and LORAs. Here is my base model Serenity: https://civarchive.com/models/110426
I upload 2-3 times a week, however, I write down suggestions all the time. If you want to see someone made into a model just leave a comment :)
Suggestions left on my donation page ( https://www.buymeacoffee.com/malcolmrey ) are taken with the highest priority and you will be able to enjoy them way before they get released here. (I also accept custom requests)
The reason is that I can't keep up with uploading and the models are being made pretty much all the time :-)
As mentioned, via the donation page I accept custom requests, so feel free to engage with me there, I will answer all your questions :)
Description
FAQ
Comments (11)
This one looks nice! I've been wondering, though: Have you tried using the same dataset with different base models - as in have you seen what this dataset looks like, if trained on a basic RealisticVision3.0 model?
great question, at the start I've been using the base 1.5 model, then I trained on some internal custom merge and then I moved to realistic vision 2 (and later 3)
now I am on serenity (v1), I might migrate to another model eventually but at the moment it is Serenity
some people say that they get better results when training on the base 1.5 model, however, I tried it in the past and I do not agree
also, I'm not training direct LyCORIS, I train dreambooth, and those who use those checkpoints would prefer to use better base model than 1.5 :)
@malcolmrey I mention it because I've noticed some of the same facial structures in your different models - eyes slanting and stretching a little, jaws getting a bit compressed - almost like the Serenity model has a little bit of a lensing effect to it. I think maybe it would be worth a try using different base models for different people, depending on what their appearance demands. I know it's a pain in the ass to run a model twice, but I think it's worth investigating.
I've been getting into using the Colab Notebook a little, so I might take a shot at running the same dataset and settings on some different models, to try and figure out exactly what's lost and gained with the different ones :)
@malcolmrey Is the Serenity model on HuggingFace so we can get a model card for it?
@joakimkunzdesign128 here it is, just uploaded :-)
@malcolmrey Aces! :D Gonna give that a try :) *Edit: I get a HTTP error when I try to use it with Colab Notebook. The same goes for a bunch of other models - can't tell what the error is, though.
@joakimkunzdesign128 oh, that's a shame :(
1. you could perhaps notify the creator of the colab so they could fix it
2. by bunch of, do you mean all that you've tested or it does work on some? could you link one (two would be best) that works? perhaps the model is in some subfolder or something (perhaps file name)
@malcolmrey Dunno - seems kinda random which ones in can download and which ones it can't. Gonna have to do a little digging, I think.
@malcolmrey Seems like you were right - the notebook has trouble with some newer models. It either can't download them, or it crashes during Unet training. Guess I'll have to look for a newer one :P
@joakimkunzdesign128 my friend is using RunPod and she says this is currently the best option (she tried rundiffusion but it was problematic at times)
I'm working on a service where people could train using my method (that is first step), if that becomes successful then I'm thinking of making it available to do generations there as well, so we will see :)
@malcolmrey Hey - sorry I didn't respond. Had a pretty grim situation develop out of nowhere, and haven't really been paying attention to what anybody's been saying.
Meanwhile, I've been gathering some different datasets - from some old movies mostly. Dolly Parton, Pam Grier - been trying to put together a Gillian Anderson/X-files thing, both from the classic and the new series. I've had some pretty varying results (although I did get a pretty good model of 2016 Scully). Anyway, I'd be happy to package it all up and send it over, if you want some of it for a rainy day (turns out ripping images from high def movies is a pain in the ass! XD )