Do you have a problem with shy subjects in your images? Try this love potion to really break the ice.
My Embeddings don't impart a style, don't use loads of extra VRAM, so it works great with other LoRAs and adapters.
Follow me to make sure you see new tools like this, or new styles, poses and Nobodys when I post them. More Clutter series coming too. Things move fast on this site, it's easy to miss.
Different checkpoint models will require different strengths to work, typically 0.2 to 1.8:
TOOLS:
Style-Interact Make Out: In small amounts you can get casual interactions, hugs, leaning in, etc. Dial it all the way up for full make out session and inappropriate groping. Put "kiss, kissing" in the negative if you want to get their faces unstuck.
Style-Cuddles Cuddles: A more affectionate hug or embrace without all the face time like Make Out
Style-Grabby Grabby: Very straight forward grab or grope. Low strength will grab arms or waist like a hug, high strength will go for the crotch or other sexy areas. Works with couples or solo.
more variations to come, make sure to like this page to get notifications.
Do you have requests? I've been putting in many more hours lately with this. That's my problem, not yours. But if you'd like to tip me, buy me a beer. Beer encourages me to ignore work and make AI models instead. Tip and make a request. I'll give it a shot if I can. Here at Ko-Fi
Description
FAQ
Comments (14)
I'm having an issue, the embedding is too strong. I even put it at -1.0 and the characters are still embracing/kissing etc. Any ideas?
This is an embedding and not a Lora. So a bracketed "Style-Interact" is used. A lower strength would be (Style-Interact:0.5). Whatever variable you decide on (0.8, 0.4) for less effect or (Style-Interact:1.2) for more effect. Using a negative would just mess up the whole prompt.
Another person mentioned a file format called EMBEDDING. Classification with "easy negative" etc.
You don't want easy negative:-1 or something like that, do you? That's what I mean.
@Aight_Geezer yes, I understand that. I am using it like <style-interact>:0.2 etc
It is quite interesting. Good sense of the author. Quite interesting. The author has good taste.
positive:rivalry, face-to-face,looking at another, fighting stance
Negative:yuri,.
Despite adding the prompts of hostility, looking at each other, fighting stance, and yuri rejection to the comparison image this time, they are still kissing fiercely.
I think it's so LOVE that I'm in love with it.
I think it is impossible because it is EMBEDDING, but I think it will be several times more interesting if you put it in Negative and make it antagonistic.
I think it would be good to have a separate drug EMBEDDING that is the exact opposite of this one and that they dislike each other to add to the Negative.
Interesting ideas. Thanks.
Empirical.
Somehow it is strong in describing a large number of people, but it seems to stabilize the description by keeping the number of people described small, even if you specify a level that creates a [crowd].
It can be used as a stabilizer when you want to see WAIFU and many mobs intertwined, maybe?
Are these concat embeddings? I converted it to work for SDXL (which it did marvelously) and noticed all of the vectors were identical to standard CLIP embeddings.
I can't remember. I'll pay attention next time. What do you use to convert to SDXL?
@Zovya I made a script that filled the clip_g vectors with the nearest token to clip_l
(Clip_l seems to have a much stronger effect on SDXL anyway)
I think there are more gains to be made before I release such a script.
@AICasanova whoa that would be really great to have.
Status: Styles should work rather well, but stuff like corneo embeddings would need to have a LoRA trained via PTI, simply because SDXL UNet doesn't know what some things are.
@AICasanova it is concat by the way. Most of my embeddings are. I'd be interested in that script, or if there's a way to do it with nodes in ComfyUI.
Added auto conversion of SD 1.5 embeddings to SDXL in SDNext (vladmandic/automatic). It'll be in dev for a while, but one can take a look at the code.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.