This model is an experimental LoRA inspired by a unique D&D-themed concept.
V2:
This version builds on V1.
The dataset was refined by curating only the strongest outputs from V1 into a moodboard, combined with the most iconic images from each model release as style references.
The result is a sharper, more coherent aesthetic.
Like most of my LoRAs it works seamlessly both with Original Flux dev and Krea (see sample images).
V1:
The "DnD Darkest Fantasy" style was crafted with input from my primary D&D party, each contributing 1-3 fantasy images. These images were used as reference material in Midjourney, where I blended them with my own "Personalized Style" to create a distinctive aesthetic. The training dataset was exclusively generated from various "random D&D-themed requests" they provided.
Тhe trigger words are ArsMJStyle, dnddarkestfantasy .
Start around strentgh 1 and adjust accordingly.
Distortions may occur at strengths above 1.3
Update:
Added SD 1.5 that works well in the 0.2 - 1.2 range
And a big THANK YOU! to you all for making it so popular and making this the 3rd place of the "Flux Training Contest".
Description
FAQ
Comments (4)
Good day! Tell me, is it suitable for the flux1-dev-bnb-nf4-v2 model?
Not sure to be honest. I never used nf4 versions. I usually use Q4 k for Quantized version
Good afternoon! Can you tell me which one, for example?
@Maratek Only the base flux dev Q4_0 and q4_k from hugging face, but you can search for any of those here as well - alot of people reuploaded the original quantizations. You can open any image in Comfy UI and will show you my config. I intentionally upload everything RAW and use REALLY basic workflow.