Model Description

Similar to its Illustrious counterpart, this LoRA is designed to add decorations (resembling stabbed objects) attached to characters at various body positions. While Anima base models already have some ability to depict stabbing scenes, this LoRA improves control and variety over such compositions.

This model is trained on both tags and captions — tags handle general composition, while captions provide precise control over stabbing details.

Note: This is a beta version for testing training on the Anima[preview] base model, trained on a pruned dataset of 183 images. A larger dataset will be used once the official Anima base model is released.

⚠️ Disclaimer: I am not responsible for any inappropriate content generated using this model.

Recommended Settings

• Model: Tested primarily with Anima[preview] and AnimaYume[v0.1]

• LoRA Weight: 0.8–1.0

• Steps: 28 | Sampler/Scheduler: Euler_a / beta | Resolution: 1MP
  (personal preference — feel free to experiment)

Trigger Words

1. Position Control — Anchors the stabbed location on the character: abdomen_stabbed chest_stabbed head_stabbed neck_stabbed back_stabbed

2. Wielder & Scenario Control

• independent_stabbed (Default) — No wielder. Use this when the object appears unsupported, even in multi-character scenes.

• assisted_stabbed — Someone else is holding the object.

  • 💡 Tip: This requires more canvas space. Add solo focus to prioritize the stabbed character; otherwise the composition may split focus evenly. Tags like "confrontation" or "fight" help establish atmosphere.

• self_stabbed — The stabbed subject is holding the object themselves.

• impaling_stabbed — The stabbed subject appears to be impaled by the object.

Prompting Structure Guide

Recommended prompt order for best results:

[Quality tags] → [Artist tags] → [Position trigger] → [Wielder/Impaling trigger] → [Characters involved] → [Other Danbooru tags] → [Caption]

For maximum accuracy, begin your caption with a concise sentence summarizing the stabbing scenario. For example:

"A girl is stabbed in the back and impaled by a baguette held by a man."

This single sentence captures all the key information the model needs:

• Subject: a girl

• Object: a baguette

• Location: her back

• Impaled: yes

• Wielder: a man

You can then add further descriptions — especially about how the object interacts with the subject — to refine the scene further. This level of caption-driven control is one of Anima's strongest advantages over SDXL-based models. Refer to the prompts on my example images for reference.

Negative Prompt

old, early, ugly, bad anatomy, artistic error, (worst quality, low quality), score_1, score_2, score_3, blurry, artist name, signature, watermark, jpeg artifacts, mosaic, censored, multiple weapons, (blood:1.2), (guro:1.2), (bleeding:1.2), wound, injury, blood spray

• multiple weapons — Reduces weapon duplication

• blood, guro, bleeding, wound, injury — Stabbing/impaling concepts in Anima are highly correlated with these terms; using them as negatives helps prevent unintended graphic content