Sign In
home
models
images
videos
posts
articles
bounties
tools
challenges
events
shop

SDXL / IL / PDXL | (BASE) Multiple Lora Merging in Google Colab!

8
HE
Herrscher_AGGA
New Years 2025 Badge
Apr 1, 2025
training guide
mergeloratestmerging lora
SDXL / IL / PDXL | (BASE) Multiple Lora Merging in Google Colab!

Sometimes you have a lot of images but very little space in online trainers like Tensor or Pixai? This is made for that. You can also use it to mix concepts or refine Loras.

The idea behind this was to make SDXL file mashups easier by trying to reuse my resources, which are considered outdated. I'm just too lazy to train loras for every checkpoint. Yeah, they probably won't work for every checkpoint you see, but it's good to try~ 7w7)/

PS: I did this as my code training in Google Colab. It'll probably include things like upscaling SD loras, extracting loras from checkpoints, or mash them... if they're useful to you, feel free to comment.

This was also done for personal reasons, cheers! n.n)/

PS: This is made to play with your own loras, if you want to merge someone else's lora I recommend you ask for permission.

New Version: Now refined to 4 steps!

Open In ColabWell, let's start with the Step by Step:

  • 1. Clone the sd-scripts repository

!git clone https://github.com/kohya-ss/sd-scripts.git

  • 2. Install PyTorch

!pip install torch torchvision

  • 3. Verification

import torch
print(torch.__version__)
print(torch.cuda.is_available())  # Verifica si CUDA está disponible

  • 4. Install requirements

!cd /content/sd-scripts && pip install -r requirements.txt

  • 5. Install Numpy

pip install --upgrade "numpy<2"

  • 6. Update to fix some issues

!pip install --upgrade tensorflow
!pip install --upgrade jax
!pip install --upgrade albumentations

  • 7. Download sample LoRA models

!wget Your LORA URL -O Lora1.safetensors
!wget Your LORA URL -O Lora2.safetensors
!wget Your LORA URL -O Lora3.safetensors

  • 8. Merge two LoRA models together

!python sd-scripts/networks/sdxl_merge_lora.py \
--save_precision fp16 \
--save_to /content/Lora1-Lora2-fp16.safetensors \
--models \
    /content/lora1.safetensors \
    /content/Lora2.safetensors \
--ratios 0.5 0.5

  • 9. Set up Google Colab (Optional)

from google.colab import drive
drive.mount('/content/drive')

  • 10. Merge three LoRA models together (Optional)

!python sd-scripts/networks/sdxl_merge_lora.py \
--save_precision fp16 \
--save_to /content/Lora1-Lora2-Lora3-fp16.safetensors \
--models \
    /content/Lora1.safetensors \
    /content/Lora2.safetensors \
    /content/Lora3.safetensors \
--ratios 0.3 0.3 0.3

Try upload your LoRA models and merge them

You can drag and drop your LoRA models from your computer to the hard drive on google colab

modify the scripts to use your uploaded LoRA models

Enjoy!

8

Comments