Mixon_WL — Qwen 2512 VL Edition

A Style LoRA combining two iconic fantasy aesthetics: WLOP-inspired digital painting with luminous ethereal glow, and high-quality 3D CGI rendering with cinematic precision. Now fully trained on the advanced Qwen 2512 VL Image architecture.

What It Generates

Fantasy female characters — elves, nekomimi, mages, dark warriors — with exceptional detail in accessories, enchanted clothing, and magical environments. Every output sits at the boundary between fine art and game concept.

Two Style Modes

Painterly Mode — WLOP Aesthetic

Soft impressionistic brushwork, luminous moonlit lighting, dreamy ethereal glow. Pastel-to-deep gradients, flowing hair and fabric, bokeh backgrounds. Perfect for enchanted night scenes and poetic character portraits.

CGI / Cinematic Mode

Sharp cinematic renders with physically accurate fabric physics, intricate armor detailing, specular highlights, deep shadow contrast and rim lighting. Production-ready quality for game concept art and character sheets.

Best For

• Dark fantasy illustrations and moonlit atmospheric night scenes

• Game concept art and character design sheets

• Magical companions — nekomimi, elves, mages, sorceresses

• WLOP-style fan art and inspired digital portraits

• Beach and summer fantasy with soft natural light

Prompting Tips

Trigger

ohwx — this trigger word is required to activate the core style.

Painterly look

ohwx, ethereal glow, soft brushwork, moonlit, atmospheric, luminous, impressionistic background

CGI / Cinematic look

ohwx, cinematic render, 3d render, sharp detail, armor, fabric physics, game-ready, dramatic lighting

Recommended Strength

0.7–1.0 for full effect. Use 0.5–0.7 to blend subtly with your base model. Both modes respond well across the full range — experiment freely.

Technical Details

• Type: Style / Character LoRA

• Base Model: Qwen 2512 VL Image BF16

• Text Encoder: Qwen 2512 VL 7B BF16

• Resolution: 1328×1328

• Network Dim / Alpha: 128 / 128

• Trainer: Musubi Trainer v27.1

• Notes: Trained with blocks_to_swap=47 and gradient checkpointing. Requires a compatible Qwen Image inference pipeline.

(For the FLUX.1 [dev] version of this style, click here)