Skip to main content
Concept

LoRA

A plug-and-play add-on that teaches your model a specific character, style, or pose.

A LoRA is a lightweight add-on for an AI Model — think of it as a plug-and-play skill pack. The base model knows how to draw "anime characters in general"; add a LoRA and it suddenly knows your specific character, a watercolor style, or a tricky pose, without needing a whole new model.

Types of LoRA on PixAI

  • Character LoRA — generate a specific character from a game, anime, or your own imagination

  • Style LoRA — apply an artistic style: watercolor, oil painting, manga line art…

  • Pose LoRA — nail specific poses like squatting, jumping, or dynamic action shots

  • Detail Enhancement LoRA — polish specific elements: eye reflections, fabric textures, lighting

Using LoRAs

Add LoRAs from the generation panel or the Model-Market. Two things to get right:

  1. Match the base model. A LoRA only works with models sharing its base (Illustrious with Illustrious, DiT with DiT…). This is the #1 cause of "my LoRA does nothing" — see Base-Model.

  2. Tune the weight (0.0–1.0). Start at 0.8. Too strong and it overpowers everything; too weak and it barely shows. When combining two character LoRAs, lower the dominant one to 0.6–0.7 until both characters appear clearly.

When you load a LoRA, its Trigger-Words appear in your prompt automatically — they're the keys that activate it.

Can't find the LoRA you want?

Train your own in minutes — see LoRA-Training.

See also