– Version 1.5
– Steps: 4690
– Strength: 0.5 – 0.8
– Recommended: 0.5 – 0.6
This update focuses on improving generation consistency and body-type control for male characters and common yaoi/NSFW visual scenarios.
The dataset contains around 469 images, prioritizing themes such as:
male body forms (toned male, mature male, muscular male, bara)
– glossy body lighting and skin highlights
– detailed clothing and fine outfit elements
– darker skin tones and varied body proportions
– common yaoi-style compositions and situations
Dataset Changes:
Some images have been added, and a small number of images have been removed. Each image was carefully reviewed again to improve the overall quality, consistency, and dataset balance.
Tagging Improvements
All images have been re-labeled with more precise tags.
The tagging process was refined to better describe:
body proportions
– lighting and motion effects
– interaction poses
– clothing and visual details
This should help the LoRA respond more accurately to prompts and provide better generation stability.
Body Type Balance:
A new tag mature male was introduced to balance body types between:
toned male – lean body with light muscle definition
muscular male – larger, strong muscle build
Additionally:
muscular male + bara tends to produce an even larger and bulkier physique.
This is currently based on my own theory, so feel free to experiment further and see how it behaves in your generations.
These adjustments aim to give clearer control over body proportions when prompting.
Tag Examples Included:
Some tags included in the dataset now cover more specific visual elements, for example:
motion lines, motion blur, foreskin, penis shake, muscular, autofellatio, large pectorals, groping, holding artificial vagina, artificial vagina, male masturbation, gigantic penis, instant loss, 2koma, and other scene or motion-related tags, but sometime not work too😅
Prompt Reference:
I also added base prompts used during dataset generation as references for some images. These reflect the typical prompt structure I personally use when creating images.
Example result from this LoRA:
https://x.com/lyht111222333/status/2031018065203540326
Link: https://tensorhub.art/models/965032398813164498?source_id=njuwoVDrnUe0rfcra3f88BUh
Leave a Reply