Jubjub_v1 – Photorealistic Character LoRA

Jubjub_v1 is a character LoRA designed for photorealistic SDXL image generation, with a focus on consistent facial identity across different images.

The training dataset includes a variety of camera angles, facial expressions, hairstyles, outfits, lighting conditions, backgrounds, and portrait compositions to improve character consistency and flexibility.

Recommended Settings

• Trigger word: jubjub
• Recommended LoRA strength: 0.6–0.8
• Suggested starting strength: 0.7
• Base model: SDXL 1.0

Tested with

• RealVisXL

Works best with realistic and photorealistic SDXL checkpoints. LoRA strength can be adjusted depending on the checkpoint and desired level of character resemblance.

Example images were generated using different poses, expressions, lighting conditions, outfits, and environments to demonstrate the flexibility of the LoRA.