Does what you want. Does what you love. Does what you're good at! (except hands)
Your Ikigai!
I made this mix because I really love fine details provided by some models, but didn't want be bound to a certain look for faces, lighting, etc.
This model focuses mainly on 2.5D and semi-realism, and handles loras, TIs, long/short prompts, and everything in between very well!
V2:
Now with even better lighting, more detail, and no loras needed to achieve the look!! This iteration plays a lot nicer with Textual Inversions without the need of realism, detail, or artist loras, but they're still welcome!
Added the following models to the mix
V1:
My usual settings:
Sampler: DPM++ 2M Karras
Sample Steps: 20-35
Upscaler: ESRGAN_4x
Deniose: 0.5
CFG: 5-9
Model has no VAE baked in, I personally use Graprfruit-VAE
By no means is this law, feel free to experiment and I'd love to see your results!!
For inpainting, I generally limit to Only Masked to 800 x 800 or lower and 0.5 denoise, as it will start to get a little too detailed otherwise. You shouldn't need to change your prompt at all when inpainting faces to get a good result. This was a goal of mine, as I prefer not to have to do any extra steps in my workflow if possible.
This mix comprises of:
MajicMix Lux
MIX-Pro-V4
Babes 2.0
For the semi-realism look in a couple of the example images, I HIGHLY recommend the lora below at a weight of 0.1-0.2:
Danni-2D (looks to be removed, but any lora that would add realism to models like AOM3)
Model also plays reaaaally well with 2D artist style loras like DevilHS, Gopher, SamDoesArt, etc. and will often help with the hands issue.
All example images have no post-work beyond inpainting and SD upscale within the webUI.
Description
Fp16
Pruned