Type | |
Stats | 1,120 1,406 |
Reviews | (132) |
Published | Aug 19, 2024 |
Base Model | |
Training | Steps: 7,776 Epochs: 6 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Hash | AutoV2 8DBBAC11EA |
The LoRA is trained on MidJourney images that use my 'Personalized Style.'
This style is the base used in all of my LoRAs.
Apart from the general "MidJourney" aesthetic, it has strong influences from Art Nouveau, Watercolor, and Oil Pastel techniques.
The trigger word is ArsMJStyle.
Start ~1 weight and adjust according to your taste.
For quantized versions you can go a little bit higher from the start (1.2+).
This is my first attempt at a Flux LoRA, so I chose my most consistent dataset as a starting point. Trained on-site for ~7777 steps.
If you use quantized model versions of Flux, use GGUF loader nodes - they seem to work fine with the standard LoRA loaders in ComfyUI and the creator mentioned that he will add even better support in the future.
Simplistic Workflow for quantized models:
https://civitai.com/models/658639?modelVersionId=736954
The last 2 images from the showcase were created with it and the Q4 version of Flux Dev