Character Description ✨

Savita is an elegant South Asian woman with warm, golden-brown skin, long dark hair (straight or softly waved), and expressive almond-shaped eyes. She conveys a blend of poise and quiet confidence—often styled in luxurious silk sarees, embroidered lehengas, delicate sheer drapery, or tasteful boudoir pieces and artistic drapery—posed in rich, painterly backdrops (velvet, marble, or sunlit studio settings) for a refined, cinematic look.

AI Image Generation Model: Z-Image-Turbo 🔧

Z-Image-Turbo is a high-performance image generation model optimized for fast, high-quality outputs across styles and resolutions—ideal for detailed portrait and fashion synthesis.

Training Details 🧪

LoRA generated with approximately 2500 steps

Network: LoRA (linear: 32, linear_alpha: 32, conv: 16, conv_alpha: 16)

Training configuration: 3000 steps (configured)

Batch size: 4

Optimizer: adamw8bit

Learning rate: 0.0001

Optimum KSampler Settings for Best Results ✅

Base model: Z-Image-Turbo

Trigger word: savita (LoRA applies only when this token is present in the prompt)

Sampler: Steps 9, er_sde; Scheduler: beta

Upscaler: SeedVR2 Upscaler recommended for final high-res outputs

Usage Tips 💡

Use the trigger word exactly as savita in the prompt to enable the LoRA.

Combine with stylistic tokens (e.g., “golden hour lighting, 85mm portrait, cinematic, film grain”) for consistent aesthetic.

For cultural authenticity, add clothing/accessory keywords (e.g., “silk saree, zari embroidery, kundan jewelry, bindi”) and pose/style notes (“soft smile, three-quarter pose, studio backdrop”).

Disclaimer ⚠️

The training images were generated using Juggernaut XL Ragnarok and edited with Qwen Image Edit 202511.