The Lora is trained on MidJourney Images that use my "Personalised Style" + Mosaic.
Works without the trigger words ArsMJStyle, Mosaic
Have a an effect in the 0.4 - 0.8 range
Depending on the Lora stack optimal Mosaic look in the 0.8 - 1.5 range
Quality tags reduce the Mosaic effect,
Versions:
Flux - based on the latest updated dataset with new caption.
v1 - the lightest Mosaic Effect, but without strong biases towards anything. Trained on Clip skip 1. SD 1.5 and SDXL are trained with this "final" dataset as well.
Beta 2 - tries to generate Woman Portraits. Same dataset size as v1, but only 50% overlap. The Capture here was Free text + WD Tags.
Beta - 50% of the dataset of Beta 2. Not that flexible when generating new images.
The 3 Pony versions are practically different Loras, so try the other 2, if you liked one of them :)
Description
In the 0.2 - 0.6 Range the LoRA adds distinctive vibrant colors, but the Mosaic Effect might be missing or minimal.
Optimal Range is 0.8 - 1.5
For recommended usage on Quantized version check this really simple ComfyUI Workflow:
https://civitai.com/models/658639/super-simple-gguf-quantized-flux-lora-workflow
All the images with missing Metadata from the Showcase Gallery were created with it and Flux Q4 Dev model.