mayaXL — Original Character LoRA for Illustrious XL

Maya is a cool, mysterious original character LoRA trained on prefectIllustrious XL v8. She has a distinctive look that holds up consistently across a wide range of prompts and styles — from cosy bedroom shots to cinematic night city scenes. Whether she's in a black hoodie, a slip dress, or lingerie, her face, hair and eyes stay locked in.

Character Features

• Long straight platinum blonde hair

• Saturated violet eyes with strong limbal ring and vibrant iris detail

• Fair, soft skin with natural subsurface scattering

• Subtle dark eyeliner, soft glossy lips, light natural blush

• Tall, slender build with slim waist

• Default look: black oversized hoodie, hood down

Trigger Word

mayaXL

Recommended Base Model

prefectIllustrious XL v8 (trained on) — works best on Illustrious XL and its derivatives

Recommended Settings

• Resolution: 1152x1536 (portrait) or 1024x1024

• CFG: 4.5–6

• Sampler: DPM++ 2M Karras

• Steps: 20–25

• Clip Skip: 2

• LoRA strength: 0.8–1.0 (sweet spot is 0.9)

Example Prompt

mayaXL, masterpiece, best quality, perfect quality, ultra-detailed, absurdres, 1girl, solo, long straight platinum blonde hair, saturated violet eyes, vibrant iris, detailed iris, strong limbal ring, fair skin, smooth complexion, subsurface scattering, detailed face, detailed eyes, subtle dark eyeliner, soft glossy lips, light blush, natural makeup, detailed eyelashes, detailed hair strands, realistic shading, cinematic lighting, soft rim light, [outfit here], looking at viewer, [composition here]

Signature Look

mayaXL, masterpiece, best quality, perfect quality, ultra-detailed, absurdres, 1girl, solo, long straight platinum blonde hair, saturated violet eyes, vibrant iris, detailed iris, strong limbal ring, fair skin, smooth complexion, subsurface scattering, detailed face, detailed eyes, subtle dark eyeliner, soft glossy lips, light blush, natural makeup, detailed eyelashes, detailed hair strands, realistic shading, cinematic lighting, soft rim light, (black oversized hoodie:1.6), (dark black hoodie:1.4), hood down, looking at viewer, slight smile, upper body, (dark background:1.4), indoors

Negative Prompt

lowres, (bad), bad anatomy, bad hands, extra digits, multiple views, fewer digits, extra limbs, missing limbs, text, error, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, artistic error, username, scan, armour, accessories, jewelry, earrings, weapons, multiple girls, child, loli, aged down, deformed, disfigured, blurry, out of focus, glossy skin, plastic skin, shiny skin, heavy makeup, overdone makeup, clown makeup, thick eyeliner, dark lipstick, two heads, two faces, multiple faces, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, ugly, bad proportions, extra limbs, fused fingers, too many fingers, long neck, floating limbs, flowers, floral, petals, purple flowers, iris flowers, crocus, garden, botanical

Tips

• Black hoodie needs a push — the dataset covers many outfit categories so the hoodie isn't strongly baked in. Use (black oversized hoodie:1.6), (dark black hoodie:1.4) and add grey hoodie, grey clothing to your negative to force it black

• Flowers in background — purple flowers are baked in from the dataset. Add flowers, floral, petals, purple flowers, iris flowers, crocus, garden, botanical to your negative and use (indoors:1.4) in your positive to avoid them

• Violet eyes drifting — if eyes go blue or heterochromia appears, add (saturated violet eyes:1.3), (purple eyes:1.2) to your prompt

• Skin tone — Maya looks best at strength 0.9. Going above 1.1 can darken and flatten the skin

• Clip Skip 2 is important — using 1 may reduce identity consistency

• FaceDetailer recommended — use face_yolov8m.pt at denoise 0.35 for best face detail on full body shots

• Works best on Illustrious XL derivatives — other base models may give inconsistent results

Known Quirks

• The LoRA primarily locks in face identity, hair and eye colour — it's been trained across 62 outfit categories so it's intentionally flexible with clothing

• Purple/warm atmospheric backgrounds appear naturally due to dataset aesthetic — prompt against them if you want a clean dark background

• Best results at strength 0.85–0.95 — below 0.7 the identity weakens, above 1.1 quality starts to degrade

Training Details

• Base model: prefectIllustrious XL v8

• Training steps: 6800 (with epoch 1 checkpoint at ~5432 steps)

• Network dim: 128 / Alpha: 40

• Resolution: 1152x1536

• Dataset: 5431 images across 62 categories including portraits, expressions, outfits, poses, turntables, mouth shapes, and more

• Regularisation: 1705 images

🎨 Show me what you make! Post your generations in the gallery and share your prompts — the best ones get featured. She's versatile so I'd love to see what you come up with!