This is the full version of the View Between Legs!
15imagies dataset.
With this LoRA and just 1girl, you can create illustrations peeking from between the legs of other people.
strentgh: 0.9~1.1
This LoRA can be applied to a variety of situations.
I am satisfied with my learning result.
(1) view between legs ⚠it means anothers legs⚠ ←It tends to result in (3), so it's most practical to use inpainting in i2i on the 'legs' generated with this LoRA. or NP: (looking through own legs:1.5)
(2) view between legs fellatio
The LoRA for specific situations⇩
https://civarchive.com/models/588511/view-between-legs-fellatio-or-concept-lora
(3) looking through own legs
(4) view between legs (reflection mirror) ← I will fully support it in my next LoRA release.⇩
https://civarchive.com/models/1614348
example⬇
Description
15 imagies
FAQ
Comments (3)
It seems that "looking between legs" interacts with "looking through own legs" and "looking through legs," resulting in a high probability of generating a bent-over, backward-looking posture (i.e., (3) looking through own legs).
To achieve (1) "view between legs (another's legs)," "looking between legs" must be removed. The success rate is approximately 80% (based on 20 randomly seeded images). Most cases feature a single person or two people (one bending over and looking backward, while another is positioned between their legs). Adding "multiple girls" and placing "looking through own legs" and "looking through legs" in the nagative prompt (with a weight of 1.2) increases the success rate to around 95%.
Thank you for your comment. This is a phenomenon I was concerned about. As a result, I added "It tends to result in (3), so it's most practical to use inpainting in i2i on the 'legs' generated with this LoRA." to the explanation. In my opinion.
I think that LoRA designed to generate legs in the top half of the screen is difficult for IL. It's a cut and paste job.That's because I gave up distinguishing the person's legs from the foreground legs. this LoRA is premised on t2i and i2i use.
As a supplementary explanation,
The tag 'looking through own legs' exists 1.3K times on Danbooru and is effective just by using the tag, so those images weren't included in the training dataset. However, it's probably triggered by 'view between legs' due to their high cosine similarity.
Personally, I wasn't concerned and posted this because it demonstrates the high degree of freedom in IL.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.