Description

Based on my extensive testing with SDXL anime-style models, I found that most base models stubbornly refuse to depict characters with "tanned skin" or "muscles" as the "Bottom (Receiver)" in BL/Yaoi contexts.

While individual character LoRAs for Cú Chulainn and Emiya already exist, this specific LoRA focuses entirely on their relationship dynamics. It is technically impossible to completely "lock" roles using only a LoRA, but this model drastically improves the success rate, which is otherwise extremely low without it. The dataset is based on my personal art style, regenerated and refined to capture a specific aesthetic.

• As of now, the Pony version is superior in terms of prompt flexibility and situational variety. However, for clearly defined scenes, Illustrious may produce better results when given precise, well-crafted prompts.

• 今処、lycoris(単独用よりはloraの性能サポート用)を制作中です。Please read the following guidelines before use:

1. Dataset Disclaimer

Since the training dataset includes AI-regenerated images based on my original sketches/drawings, I do not claim the training data as 100% my own hand-drawn art. However, it clearly reflects my specific taste and drawing style.

2. Permissions

You are free to merge this with other LoRAs or Checkpoints. However, re-uploading (reposting) this raw file is strictly prohibited.

3. Recommended Settings

This model is a calibrated merge of three distinct LoRAs (Logic, Texture, and Atmosphere).

Recommended Weight: 1.0

It depends on model and situation. 0.6~1.0 is sweet spot. pony is OK when stronger CFG, but illustrious needs more gentle SFG.

Note: Since the optimal ratio is already baked in, lowering the weight significantly may cause the role-locking logic to fail.

4. Role

Locking Limitations As mentioned, a LoRA alone cannot guarantee a 100% role fix. If you require absolute precision in positioning (e.g., specific poses), I highly recommend using Regional ControlNet alongside this model.

5. Prompting Tip (Crucial) For Archer's skin tone:

Avoid: dark skin (This tag often bleeds onto Lancer or fails to apply correctly).

Use: tanned skin (This yields the most accurate results).

6. Usage & Quality Note

Although explicit (XXX-rated) samples are not displayed here, this LoRA was fundamentally designed with NSFW generation in mind. It is capable of generating general slice-of-life scenarios, but please understand that the quality for broad SFW situations might be slightly lower compared to NSFW outputs. I will consider improving SFW stability in future updates.

(Optional) Recommended Tags: yrym, yariyumi, tanned skin, muscular