Inspired by some collage images I saw on Midjourney which emulates newspaper clippings which are glued to form shapes or characters, i was inspired to create a SDXL LoRA (Low Rank Adaptation) style that would emulate this to some degree.
As I have documented by process for creating LoRAs using Runpod.io in this post, I started by creating an image dataset and it captions.
Then I ran my Runpod to create the various different LoRA file at different Network Rank settings. Network Rank is one of the parameters used to train your LoRA but higher the number the larger the file.
And my base model used is SDXL 1.0 so the learning is of highest quality.
In the end I selected a few keepers and have been using to create some unique art pieces myself. However I wanted to share one of these LoRAs with you so you can enjoy, experiment and create your own art using this LoRA.
Here is a comparison with keyword alone and then keyword + LoRA at various CFG scales (5, 7 and 9), you get the idea on how much the Newspaper Collage LoRA influences the generated image.
What you get with this LoRA?
The package comes in a zip file that you can download after purchase and extract. The contents include:
- newspaper-collage.safetensors – LoRA file (435 Mb)
- license.txt – License agreement
- Newspaper Collage LoRA.pdf – Installation Instructions and sample images with their prompts
- You must have Stable Diffusion setup on your PC or Mac or Cloud GPU (I use Runpod)
- You must have SDXL 1.0 base model downloaded
- Download the newspaper-collage.safetensors file and place it in your models/Lora folder
- For Automatic1111: place the file under “SD-Web-UI\models\Lora” folder
- For ComfyUI: place the file under “ComfyUI\models\loras” folder
- Use trigger word “newspaper collage style” in your prompt
Below is a collection of results created with SDXL 1.0 Base Model and Newspaper Collage Lora.