This model is trained on 1,000 images generated by MidJourney V3 whose aesthetics I absolutely love and I think it was their biggest mistake to abandon. Dataset contained 50% of 1024x1024 (with varying aspect ratios) and 50% 256x256 (also with varying aspect ratios.)

This model is great at: Abstract prompts, adding creativity, introducing stuff that is great, colors, some specific styles, wallpaper material, art and artistic stuff, aesthetics.

EXPAND THE DESCRIPTION :)

Check out the comments for all the different things it can do! All my images have their prompts and gen config

I also added some specific stuff Stable Diffusion struggles with like "honey river" never draws an actual river filled with honey. This LORA kinda works like 80% of the river is honey but SD is quite stubborn, base model lacking in fantasy images much?

You can also prompt for names of songs, Genshin Impact OST tracks have really cool names, I have a bunch of showcases in the comments. YouTube Music link for album

If you keep getting mountains, landscapes or some concept that you didn't ask for and you don't want, use negative prompts! They work really well.

I heavily suggest you to not use any LORAs or TIs for your first 100-200 images or perhaps ever because most stuff require specific ways to look good. This LORA makes drastic changes to the text encoder and UNET.

I would recommend the detail lora after you play around for a bit.

I also strongly suggest you to use dreamshapers 8, all of the preview images were non-nitpicked and generated used Dreamshapers 8 using DPM2 40 steps at CFG 4 512x512 with no post processing or inpainting. CFG 4 is recommended.

Legal: Rest assured, all images were all generated by me and at the time of generation, MJ allowed commercial rights to the images. Even if they didn't, AI outputs are not copyrightable. You agree to indemnify me from any sort of harm or problem that arises with the usage of this LORA. You should obey the law always.