Liko - Pokémon [PixAI DiT]

A high-fidelity character LoRA trained directly on PixAI using the Anima DiT v1.0 architecture. Built on a hand-curated dataset of 50 multi-aspect images, this adapter features a fully decoupled wardrobe for maximum styling flexibility across diverse environments without identity bleed.

💡 Core Features

• Decoupled Identity: Facial geometry, signature dual-tone hair, and eye characteristics are separated from clothing captions, allowing seamless outfit swaps and custom costume prompting.
• Native Anima DiT Geometry: Trained with multi-aspect ratio bucketing (Square, Portrait, and Landscape) to prevent distortion, stretching, or anatomical drift across varied canvas shapes.
• Modular Adapter Stacking: Clean weights designed to stack smoothly with external aesthetic, lighting, and texture LoRAs (such as Cinematic Film Grain) without burning or over-saturation.

⚙️ Recommended Generation Settings

• LoRA Weight: 0.75 – 0.85 (Recommended baseline: 0.80)
• Base Model Checkpoint: Tsubiki.3 DiT
• Sampling Steps: 30 – 32
• CFG Scale: 4.0 – 4.5
• Sampler: Euler or Euler a
• Native Resolutions:
  • Square: 1024x1024 (1:1)
  • Portrait: 832x1216
  • Landscape: 1216x832

✨ Universal Positive Base Tag Anchor

masterpiece, solo, 1girl, liko (pokemon), black hair, blue inner hair, two-tone hair, aqua eyes

👗 Classic Outfit Prompt Tags

teal sleeveless hoodie, long tunic hoodie, front pocket, light blue hood lining, black long-sleeved undershirt, white shorts, black ankle socks with teal trim, white sneakers with teal laces, green hair clip, double hairpin, yellow crossbody bag, strap across chest, ancient crystal pendant

🚫 Universal Negative Prompt

(worst quality, low quality:1.4), score_1, score_2, score_3, photorealistic, realistic, bad anatomy, bad hands, missing fingers, extra digits, blurry, cropped, frame, watermark, signature

📊 Training Details

• Platform: PixAI LoRA Trainer
• Base Architecture: Anima DiT v1.0
• Dataset Size: 50 curated high-resolution images
• Bucketing: Aspect Ratio Bucketing (Portrait, Square, Landscape)
• Captioning Strategy: Comma-separated tag-based captions with strict character-feature decoupling