Character LoRA (Z-Image Turbo)

This LoRA is trained on AI-generated character images and fine-tuned to produce a single, consistent adult character across many scenes, poses, and camera setups. It’s built for professional photo realism with strong character persistence, not for random-face generation.

what it’s trained to do

• Keep the same face and overall identity stable between generations.

• Keep consistent signature traits: baddie makeup, full lips, heavy lashes, smooth straight hair, tan glossy skin, tattoos, jewelry, long nails.

• Push a high-appeal curvy athletic hourglass silhouette while still reading as a pro fashion/photo shoot.

character sheet (what you’ll usually get)

• Adult woman (25+)

• Face: glam doll-like makeup, defined brows, long lashes, full lips, clean symmetrical look, confident/sultry expression

• Hair: long straight, most often jet black or honey blonde depending on prompt

• Skin: deep sun-kissed tan, glossy highlights, visible realistic pores when you ask for it

• Details: visible arm tattoos, belly piercing, gold necklace and rings, long manicure (often neon colors if prompted)

• Body type: athletic-curvy hourglass, snatched waist, full bust, wide hips, pronounced glutes and legs (high “Instagram model” vibe)

• Clothing: works best with bikini / lingerie-style fashion and other tight outfits.

training note

• Dataset: AI-generated character images (not a real person).

• Goal: identity consistency + “pro photo” finish across varied setups.

recommended use

• Model: Z-Image Turbo

• LoRA weight: start 0.7–0.95 for consistent face/body. Go lower (0.5–0.7) if you see overbaked features.

• Prompting: put everything in the positive prompt. Z-Image Turbo commonly ignores classic negative prompts, so use explicit constraints in the main text.

trigger and prompt style

Use your trigger token first, then describe the character in batches (identity → hair/skin → details → outfit → pose → camera → lighting → cleanup).

known behaviors

• If you under-specify camera/composition, it may drift into iPhone type of . Add “face fully visible” and “near full-body framing.”

Latina