Morgana xi

morganaXi_v10.safetensors

v1.0

Christina — Realistic Mature Female Character LoRA V1.0 · V2.0 · V2.0aChristina is a realistic mature female character LoRA trained across two generations of CyberRealistic Pony. She is designed to generate a consistent, believable woman in her mid-40s with natural, realistic proportions rather than idealized beauty standards.Christina woke up one morning to a quiet house and golden light spilling through the curtains — the kind of morning that feels like a beginning. At 45, she'd built a life she was proud of. Not without its complications — the marriage hadn't lasted, and there had been harder years than she'd anticipated. But the kids were hers, the house was hers, and somewhere in the process of putting herself back together she'd found a steadiness she didn't know she was capable of. Now the kids were away at school. The house was gloriously, completely hers. And something she hadn't felt since she was a teenager was stirring back to life — a curiosity, an aliveness, a slow-burning excitement about who she still was and who she might yet become. For years she'd been someone's wife, someone's mother, someone's ex. She'd been responsible and capable and fine. But fine wasn't the same as alive. Christina isn't having a crisis. She's having an awakening. She's not trying to be younger. She's not trying to be different. She's just finally giving herself permission to be fully, unapologetically herself — and discovering that the woman she is at 45 is far more interesting, far more alive, and far more ready for whatever comes next than she'd ever given herself credit for.<hr />What's new in V2.0 / V2.0a: Complete dataset rebuild — 59 curated images across a wide range of expressions, lighting conditions, angles, hair styles, environments, seasons, and clothing. Significantly improved identity consistency, expression flexibility, profile angle robustness, and lighting range over V1. Glasses coverage dramatically improved. New trigger token — see below.V2.0 and V2.0a were trained on the identical dataset and parameters — the only variable is the base model. V2.0 uses CyberRealistic Pony v17. V2.0a uses CyberRealistic Pony v18.V2.0a (v18) is the recommended primary. It produces better skin texture, more nuanced lighting, and more natural detail overall. Use V2.0a unless you have a specific reason to prefer V2.0.V2.0 (v17) may be preferable when:<ul><li>Your existing workflow is built around CyberRealistic Pony v17 and you want to keep your LoRA stack consistent</li><li>You're stacking with other LoRAs trained on v17 — matching base models can improve compatibility</li><li>You prefer a slightly softer, less detailed skin rendering — v17 produces a subtly smoother look that some may prefer for dreamier or lower-contrast scenes</li><li>You're generating in softer diffuse lighting (overcast, cool window) where v18's added texture detail is less of an advantage</li></ul>For most users and most scenes: use V2.0a.<hr />What this LoRA does well: Consistent facial identity across different scenes, lighting conditions, and outfits. Strong expression range from tired and introspective to laughing and candid. Handles profile angles, 3/4 angles, and looking-away compositions well. Pairs naturally with age and body realism tags to generate a genuine, relatable mature female character.<hr />Recommended settings:<ul><li>Trigger word: <code>cstna</code> (changed from V1's <code>christina_lora</code>)</li><li>LoRA strength: 0.85 standard / 0.90 outdoor or hair-up scenes / 0.95 robe or intimate scenes</li><li>Clip Skip: 2</li><li>Resolution: 832×1216</li><li>Steps: 30, Sampler: DPM++ SDE Karras</li></ul>V2.0a (v18) — recommended: CFG: 6.0V2.0 (v17) — fallback: CFG: 5.0 (Skimmed 4.0)<hr />Recommended positive additions for best results:<pre><code>45 years old, mature female, slight smile lines,
oval face, defined cheekbones,
wavy hair, hair down, loose waves,
skin pores, natural skin texture, skin imperfections</code></pre>Recommended negative additions:<pre><code>perfect body, athletic body, slim, skinny, model,
young, teenager, flawless skin, idealized beauty,
unrealistic proportions, smooth skin, perfect skin, airbrushed, plastic skin</code></pre><hr />Identity tip — hair down matters: Christina's wavy hair worn loose is her strongest identity anchor after her face. Prompting hair-up styles (bun, ponytail, updo) causes face shape and age to drift noticeably. Always add <code>wavy hair, hair down, loose waves</code> unless the scene specifically requires hair up. For hair-up scenes, also add <code>oval face, defined cheekbones</code> and bump LoRA strength to 0.90.Skin plasticity fix: The <code>score_9</code> tags push toward smooth skin by default. The negative additions above address this. For best results also add <code>skin pores, natural skin texture, skin imperfections</code> to positives.Smile bias note: Christina has a smile bias from dataset composition. For neutral or serious expressions:<ul><li>Drop ExpressiveH strength from 0.5 → 0.2–0.3</li><li>Add to positives: <code>closed mouth, lips together, no smile, resting face</code></li><li>Add to negatives: <code>smiling, smile, open mouth, teeth, happy, cheerful</code></li><li>Use explicit expression tags: <code>neutral_expression</code>, <code>serious_expression</code>, <code>tired_expression</code></li></ul>Smile bias persists in warm/cozy contexts (firelight, candlelight, robe) even at low ExpressiveH — this is a known dataset-level characteristic.Outdoor lighting tip: Use <code>golden_hour</code> rather than <code>overcast_lighting</code> for outdoor scenes. Flat overcast light can contribute to plastic-looking skin. Golden hour directional light produces more natural skin texture.Blue eyes fix (outdoor scenes): Bright outdoor light occasionally causes eye color drift. Add to negatives: <code>blue eyes, light eyes, hazel eyes</code>Warm interior / robe scenes: In warm-lit home scenes, add <code>restaurant, bar, public place</code> to negatives to prevent the model reading the scene as a date or glamour context. Also add <code>curly hair, tight curls, ringlets</code> to negatives — warm lighting can cause hair to render tighter than canonical.<hr />Stack recommendations:V2.0a (primary):<pre><code>&lt;lora:cstna_face_v2a:0.85&gt;
&lt;lora:ExpressiveH_jspp3733410:0.3&gt;</code></pre>V2.0 (fallback):<pre><code>&lt;lora:cstna_face_v2:0.85&gt;
&lt;lora:ExpressiveH_jspp3733410:0.3&gt;</code></pre><hr />What this LoRA is NOT good at: Generating male subjects, anime style content, or fantasy/non-realistic aesthetics. Does not work well with non-Pony based models. Not intended for generating minors or non-consensual content.Note: This is a fictional character LoRA. Christina is not based on any real identifiable person.

Christina - A Realistic, Mature, Female Character LoRA

V2.0a — Parallel trained variant. Identical dataset and parameters to V2.0 — 59 curated images (generated by Nano Banana 2 (Gemini 3.1 Flash Image)), same expressions, angles, lighting, environments, hair styles, and clothing. Same Gemini anchor method. Same trigger token: <code>cstna</code>. Only variable is the base model: V2.0a targets CyberRealistic Pony v18 vs V2.0's v17. v18 produces improved skin texture, more nuanced lighting, and stronger overall photographic realism. Recommended as the primary version for most users and scenes. V2.0 remains available as a valid alternative — see main description for guidance on when v17 may be preferable.

Files with SHA256: 60e98bafa7c005647ac12522409dae83b1ef040e8fb8667ac74548c120d28e55