Lady Death LoRA — v1 (Experimental)

A LoRA trained to reproduce the comic book character Lady Death on an Anima-based model.

Recommended base model: Anima (base)

⚠️ Known issues:
This is a first experiment and not yet perfect. Generated images frequently include unwanted signatures, watermarks, or logos — a side effect of the training dataset, which I'll be cleaning up and improving in future versions.

✅ What works well:

• Blends smoothly with other style LoRAs, often noticeably improving overall output quality when combined.

• Compatible with both Euler Ancestral and ER_SDE samplers.

• Recommended CFG Scale: 4.5–7, with the higher end of the range working better on Euler.

• Suggested steps: ~40 for Euler, ~30 for ER_SDE.

Recommended LoRA weight: 0.8–1.0
Higher weights often produced garbage/broken output, so it's best to stay within this range.

Notes:
This is my first attempt at training this character, so feedback and suggestions are very welcome! Let me know how it performs for you and what could be improved — future versions will address the signature/logo issue with a cleaner dataset.

Some examples with notes:

checkpoint: animabase/anima_baseV10.safetensors
sampler: ER_SDE
steps: 30
guidance: 5.8
prompt: "1girl, solo, white hair, lipstick, long hair, jewelry, red lips, earrings, white eyes, weapon, red background, colored skin, makeup, signature, navel, underwear, smile, traditional media, panties, sword, elbow gloves, gloves, arm up, pale skin, no pupils, pasties",

strength: 1.0

Same as above - different prompt:
"1girl, long hair, weapon, solo, sword, cape, thighhighs, blood, holding weapon, white hair, holding sword, holding, torn clothes, very long hair, elbow gloves, gloves, signature, navel, colored skin, cleavage, garter belt, blue skin, underwear, dated, traditional media, blood on face"

Now with some "tricks" and styles, sorry they are my personal loras can't expose.