Character LoRA for Matthew Kantakouzenos — an imagined version of a historic Byzantine noble navigating the tension between monastic life, political power, and personal conviction in the 1300s.Trained on a curated dataset of painterly, cinematic images to capture:<ul><li>consistent facial identity</li><li>contemplative and serious expressions</li><li>multiple roles (monk, noble, soldier)</li><li>Byzantine-era atmosphere and lighting</li></ul>The model is designed to maintain identity across varied outfits and settings while preserving a cohesive medieval aesthetic.<hr />Trigger Wordmatthew_byz<hr />Recommended SettingsLoRA weight:0.6 – 0.9<ul><li>~0.6 → more flexibility</li><li>~0.8 → best balance</li><li>~0.9 → strongest identity</li></ul><hr />Example Promptsmatthew_byz, young Byzantine man, dark robe, contemplative expression, candlelight, medieval monasterymatthew_byz, young Byzantine noble, ornate robe, palace interior, composed expression, cinematic lightingmatthew_byz, wearing light armor, holding spear, courtyard, serious expressionmatthew_byz, simple tunic, no robe, natural light, neutral background<hr />Notes<ul><li>Works best with SDXL-based models</li><li>Strong identity retention across different roles and clothing</li><li>Trained with varied lighting (candlelight, window light, interior/exterior)</li><li>Designed for cinematic, painterly outputs</li></ul><hr />Limitations<ul><li>May favor medieval / Byzantine aesthetics due to training style</li><li>Strong lighting bias (warm, dramatic tones)</li><li>Extreme modern prompts may require lower LoRA weight</li></ul>

Matthew Kantakouzenos

Initial release of Matthew Kantakouzenos LoRA. Trained on a curated dataset of Byzantine-inspired, painterly images capturing multiple roles (monk, noble, soldier) and varied lighting conditions. This version balances identity strength and flexibility: - consistent facial features across angles - works across different outfits (robes, armor, tunics) - strong performance in cinematic and medieval scenes Best results achieved around mid-to-late training epochs.

Files with SHA256: b02fc5d86be0584b8d81cb500716304610ceaf407a249ffbe56c61e3701dd6ae