Skip to main content
A Low-Rank Adaptation, or LoRA, is a small file that adjusts a base model toward a specific style, character, or concept. Mage carries a catalog of them and runs them in the browser, so you stack LoRAs without installing a model or owning a graphics card.

​
Which base models take LoRAs

LoRAs are architecture-specific, so an SDXL LoRA will not load against Flux. Check what the LoRA was trained on before you look for it.

​
How many you can stack

On Stable Diffusion 1.5, SDXL, and the SDXL checkpoint families, one LoRA or Textual Inversion runs unlimited on Basic, two on Pro, and more on Pro Plus and Max. Video LoRAs are gated per model: on Wan 2.1 one LoRA needs Pro Plus and two need Max, on Framepack any LoRA needs Pro Plus, and on MiniMax-H3 two LoRAs need Max. Every combination also runs on Gems.

​
How to use a LoRA

  1. Pick your base model first. This determines which LoRAs are compatible.
  2. Open the LoRA panel and search the catalog.
  3. Add the LoRA and set its weight.
  4. Add the LoRA’s trigger words to your prompt if it has them.
  5. Generate, then adjust the weight rather than the prompt if the effect is too strong or too weak.

​
Importing your own

Pro Plus and Max can import LoRAs along with checkpoints, Textual Inversions, and Lycoris. See importing custom models.

​
Limitations

  • Stacking many LoRAs degrades output. Each one pulls the model in a different direction, and past two or three they average into mud.
Last modified on September 14, 2026