LORA of Sadie Sink
Welcome to my corner!
I'm creating Dreambooths, LyCORIS, and LORAs. If you want to know how I do those, here is the guide: https://civarchive.com/articles/7/dreambooth-lycoris-lora-guide
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 ko-fi page ( https://ko-fi.com/dreambooths ) are taken with the highest priority and you will be able to enjoy them way before they get released here. (I accept also 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 ko-fi page I accept custom requests, so feel free to engage with me there, I will answer all your questions :)
Description
FAQ
Comments (22)
Hi, I've been using the Lycoris model and see you just published the Lora ... is there instances when one should use the Lora and when one should use the Lycoris? I've been confused about that. I read that when using several Lora in a prompt that if the weights add up to more than 1 then they won't all be working correctly, but not sure that's still true?
Hey hey, great question!
Two main reasons, one is rather plain: not everyone can use LyCORIS (the computer is not good enough or they use an online service that has no support for LyCORIS but does support LORA).
The second one is that some people prefer flexibility (in poses, distance) and don't set top priorities on photorealism. So for them, LORA might be better.
And there are some that want to mix it.
So, to answer your question: "is there instances when one should use the Lora and when one should use the Lycoris? I've been confused about that.".
Depending on your preferences, you may want to go with LyCORIS or with LORA or perhaps try playing with both :)
"I read that when using several Lora in a prompt that if the weights add up to more than 1 then they won't all be working correctly, but not sure that's still true?"
It's not a specific science that it has to sum up to "1" but it is true that the more LoRAs you'll use - the worse effect you may get (something kind of like a "model collapse").
In general, some LoRAs straight don't want to cooperate with other models (and need very low weights to even work properly) and some are more forgiving.
If I may comment, yes LoRas break beyond 1 strength and are also very strong at 1 to begin with,
From my Observation I found that Lycoris models are really hard to break and you can go upto 1.4 without breaking,
its kinda hard for me to go back to LoRa now
@dogu_cat @malcolmrey Thank you both for the info, and of course for the models!
Hey, great work! Just curious, what checkpoint did you train your LORA on?
Hey hey! RealisticVision 2.0
I have a whole guide for my process (in the article section)
This Lora has really good quality!
Would you consider doing a high quality one of Emma Watson? At age 25 to now.
Thanks! I'm glad you like it :)
Would you be able to grab me some high-quality photos of Emma Watson from that age period? It would help a lot and definitively speed up the process :)
@malcolmrey I can definitely help you with that. Question is what kind of images do you need? I've tried making a LORA with the ones I had before but it didn't turn out too well. Maybe if you can give me one of your final data sets I can give you similar images.
@unstablestDiffuser This was the latest dataset: https://imgur.com/gallery/ROU2JRN
But it will be fine when you provide around 30 photos in high quality where the face is not obstructed, I prefer to crop myself as I know what I'm looking for :-)
thanks!
@malcolmrey Ok, perfct, so is it face only? or is it better to also include some full body shots?
@unstablestDiffuser some half body are okay for sure, but it is important to have the face in good quality :)
@malcolmrey Hey, where/how can I send you the images?
@malcolmrey Hey, I don't know if you missed my comment, but anyway, I put all the images on my GDrive, zipped, seperately and upscaled versions if you need better quality: https://drive.google.com/drive/folders/1bFSSzifc2-mZ_vvr0z0l9qWauEIATDSa?usp=sharing
@unstablestDiffuser I did, in fact, miss it, I'm sorry, but thanks for providing the link again, I've saved it and will download it tomorrow (so please don't remove it yet)
recent days were quite busy (sdxl being one of reasons, as you can imagine :P) but i will certainly handle/finalize EW before this weekend is over :)
@malcolmrey Awesome, I'm excited to see the result :)
And no worries!
@malcolmrey heya, how's it going? Are the photos I provided good enough?
@unstablestDiffuser hey hey! yes, they helped! thank you :)
Emma is already trained, I will try to upload her soon enough :)
Does this lora perform well with full body shots as well? Or is it only for closeups?
my models work better with close-ups but they can work with wider shots, but you need to use high.res.fix with higher denoise to get great results
otherwise just do the scene you want and then fix with inpainting :)
@malcolmrey Thanks for the great advice! Your Loras are some of the best :)
@DrPepapTheKnave Oh thank you! :-)
I'm currently doing something top secret but because of that I actually had to train some loras in kohya (as you know I usually do extractions) and I really liked the results, so it may be that in the future i will be doing more of the lora trainings :)
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.