Lisa is an experimental character LoRA trained directly on Tongyi-MAI Z-Image Base and designed for use with both Z-Image Base and Z-Image Turbo.

Lisa is a fully synthetic adult character and is not based on any real person. All training images were generated using AI models.

The LoRA is designed to preserve Lisa’s facial identity and body proportions across close-up portraits, medium shots, full-body compositions, different camera angles, expressions, outfits, environments, and lighting conditions.

The training dataset contains 24 AI-generated images, including detailed facial portraits, medium-body references, fashion photographs, and full-body studio images. Clothing, pose, framing, expression, lighting, and background were described separately in the captions to keep the character flexible and responsive to prompts.

Trigger word: lsc
Recommended strength: 0.8–1.0

This is an experimental release, so results may vary depending on the selected checkpoint, prompt, sampler, resolution, and LoRA strength. Feel free to experiment with different settings.

Feedback and shared generations are always welcome.