If you'd like to support me and do more:
https://ko-fi.com/edgonkofi - commissions open!
If you're looking for a >>LoRA Making Tutorial<<
Visible Thong Straps - by EDG
Formerly "Whale Tail" because some people looooove to make remarks on technicalities...
I did this to help out someone who wanted "a whale tail from the front". Apparently, that doesn't exist? Didn't know.
Someone explained to me what a whale tail is. I found that it was an interesting concept to explore.
Please let me know down below what you think.
Love,
EDG
Description
FAQ
Comments (19)
Not to be that guy but without a shot from the rear, they're just highleg panties!
I recommend you use Goofy's version for rear shots. I was told his version did rear shots well so this one is more for front shots.
That being said, nothing stops me from updating this model so rear shots can be done well in the future.
Thank you for your comment and stay tuned!
@EDG No worries! I didn't mean anything serious by it, and largely I was poking fun in jest!
The Whale Tail is the back (ass) and not the front. Your model shows a G-String from the back and not a thong with a "T". A thong is what makes a Whale Tail from the ass, You cannot make a Whale Tail with a G-String (too thin).
So technically it's not Whale Tail. But a nice model.
Thank you for your comment.
I hope to see your review
Hey EDG, glad to see you're back creating again! Does this have much training with behind shots as well? That's usually what people think of when they assosciate "whale tale" as the thong is what creates the illusion of the animal's actually tail.
Either way, this looks great, looking forward to trying it out.
Nah those are for front shots, and I'm going to change the model name since people keep making this remark...
Like it has been said twice and you say it a third time. Why?
Thank you for welcoming me back btw, it means a lot.
@EDG Well, I wanted to combine it with asking as well if there was training as well for the shots from behind. It looks great either way, was just wondering about the training and stuff.
@Balthazar99 No harm done, I prefer to focus on your warm welcome, which is what I will do.
1- Training has been done using almost exclusively front shots
2- At this point it is entirely feasible to make the dataset more diverse to include a variety of angles, according to data availability
3- Due to how LoRA training works (averaging data) it is yet unclear if I must separate images folders according to angles or not
In order to stay updated on the process, I encourage you to like this model, if I ever come around to making a more complete Whail Tail LoRA you will be notified.
@EDG Awesome! Thanks for the detailed response, I always love gaining more understanding in this regard, knowledge is power! I do TIs/embeddings, but so far just on people, and mostly just personal training and use, but I would love to learn how to do styles and other stuff, so I love hearing the information. Thanks for all your awesome work, as always!
@Balthazar99
If you're interested in training people:
1- Gather 30 something good images where the face is clearly visible and don't take only face shots, make sur to have plenty of full body shots in the mix
2- Flip your dataset, meaning mirror all images. Why? Because mirroring images with help the training, the AI will get more used to your character's face and you will have increased likeness.
3- Caption your dataset so that you have in your captions the most important features, both facial and other features. If you're lazy just go with captions such as 'MyCharacter_face,MyCharacter_body' those will become your trigger words. It's not perfect, but it's a good start.
Then for training it depends if you're training a realistic looking character or anime. For realistic characters you can go with base SD1.5, for anime you can go with Lykon's AnyLoRA or Anything3.0 if you're the conservative type. + MAKE SURE YOU HAVE BOORU TAGGING IF YOUR CHARACTER IS ANIME.
If you're interested in training styles:
1- Gather a great number of images drawn in the style you want, I can't tell you how much but they definitely need to be varied.
2- Caption your dataset however you see fit (BLIP or BOORU)
3- Send it to train, and I would recommend dabbling with AnyLoRA for styles since Lykon does it so well. Use 75 tokens max with your trigger word 'MyStyle'.
This advice is by no means exhaustive, it's the bare minimum you should know, and according to your own training experience, you will gain other knowledge. For this kind of thing you gain more by experimenting than by contemplating theory.
Get to work! :D
Welcome back
Thank you! It's good to be back
mmmmmmmmmm
So fast! I will be forever thanking you!
I'm glad you like it
You're welcome
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.