LoRA LyCORIS support - wkpark/sd-webui-model-mixer GitHub Wiki

LoRA/LyCORIS export/merge

Extract Method

Extract merged LoRAs

If you generate an image with LoRAs included in the prompt, the current model you've selected(or a merged) model will include the weights or LoRAs. This feature extracts the weights of the currently included LoRAs and stores them as a LoRA/LyCORIS(locon) file.

Difference between base and current

This feature will save the difference between two selected models as a LoRA or Locon. the second selected model could be the currently used model(or merged model)

LoRA export

image

  1. select LoRA
  2. select extract method (perform extract any LoRAs used in the prompt. so, all LoRAs will be merged into extracted LoRA)
  3. LoRA dimension: SVD calculation will be performed on weights, and the size of the weights can be reduced based on the selected dimension.
  4. advancd LoRA extraction options
    • TextEncoder minimum diff: if you choose 0, all differences in weights of textEncoder(BASE) will be extracted.
    • Quantile: if you choose 1.0, all differences in weights will be extracted.
  5. Calculation device: select GPU for fast calculation or select CPU for high precision calc.
  6. select LoRA name.
  7. Press the button to perform extraction

screen shot

image