This embedding is for creating microskirts and miniskirts. It should work across a variety of models. I've noticed in NSFW-centric models it tends to produce bikini bottoms instead of skirts, but more SFW models behave a little better.
Enjoy!
Description
80 images, 40b/2g, 2000 step alternating training
FAQ
Comments (17)
Mind if I ask you a couple things about training embeddings? Do you use tags when you train them? Did you use one of the default style txts or did you make one?
If I may chime in, also which model did you create the embedding on and did you use the VAE?
And another question, what does batch size does for embeddings? I've only been able to train hypernetworks successfully. Embeddings elude me.
Yes I captioned them, which for embeddings means pointing out the non-critical info (background, etc) not the subject itself. Used the default subject_filewords.txt, I've tried others but it's fine as is.
Batch size is just a good way of speeding up training with a larger dataset as your VRAM allows.
As far as I know, all embeddings are best trained on SD 1.5 with no VAE loaded.
Thanks. I'm gonna give it a try later Maybe for once I will be successful with an embedding.
I guess that explains what I have been doing wrong, by describing the entire image in the training data, including the subject.
This looks like a nice embed by the way, haven't had a chance to try it yet though, thanks for putting your work out there.
Hi - what's the trigger word(s)? I've had pretty hit & miss results in getting this to activate. Tried: clothing, miniskirt, microskirt, micromini : but nothing is getting reliable activations.
Can Anyone help?
It's just the embedding name, micromini - try a prompt like 'photo of a micromini woman'
OK - Thanks - Will try some more experimentation.
It's "micromini". Just watch examples from the author. Most of them have prompt data.
@limpokoso I know this is 3 months old, but the trigger word is whatever the filename is. So you can change it and make it whatever you want.
Only saying this because I've seen plenty of authors change the filename they upload after making their example images, so their prompts dont always line up with the filename.
.pt does not appear in my SD 1.5, how would one convert a .pt file into a .safetensors or .ckpt?
Please let me know how I can resolve this issue, thanks.
It's an embedding, put it in your embeddings folder and then just reference the embedding name in your prompt.
This is not a checkpoint file... it's a textual inversion embedding.
trigger word in InvokeAI is "jan5test1"
Plans for XL?
TAG YOUR FUCKING NUDITY