Velyrra — Anima LoRA

A character LoRA trained on Anima Base v1.0, built from a 50-image hand-posed dataset covering portraits, half-body, full-body, and dynamic action shots to keep the identity consistent across a wide range of poses and angles.

Trigger word: velyrra

About the character

Velyrra is a dragon-girl hybrid with dark purple-black shaggy hair, purple eyes, cream-colored curved horns, tiger-style facial markings, a long ringed tail, white paw-like hands and feet, small angel-style wings, and a signature black sweater with an orange/red swirl pattern and white fur collar.

Usage tips

• velyrra alone will generate the character with all core traits (horns, hair, eyes, markings, tail, paws) — no need to spell them out.

• Works well combined with pose/shot tags such as: portrait, half_body, full_body, standing, sitting_on_chair, walking, jumping, arms_crossed, hands_framing_face.

• Add sticker_outline if you want the thick white/lavender sticker-style border seen in some of the reference art — omit it for a cleaner painterly edge.

• Responds well to strong style overrides (fantasy, cyberpunk, gothic, kimono, etc.) — core traits like horns, hair color, and tail pattern generally survive stylistic reinterpretation.

Known limitations (v1)

• Wings are inconsistent — they may not always appear, especially under heavy stylistic prompts. If wings are important to your generation, consider explicitly reinforcing them in your prompt.

• Facial tiger markings can be faint in profile or heavily stylized outputs.

• Under very strong non-default art styles (painterly realism, moody cinematic lighting), horns may occasionally not render — closer to the training style tends to be more reliable.

Recommended settings

• CFG scale: ~6-8

• Sampling steps: ~25-30

• LoRA weight: 0.8-1.0 for strongest identity match; lower slightly (0.6-0.7) if blending heavily with another style LoRA

Version notes

• v1: trained on Anima Base v1.0

• Planned: additional training data emphasizing wings and facial markings for improved consistency in future versions