Type | |
Stats | 714 |
Reviews | (112) |
Published | Aug 30, 2023 |
Base Model | |
Hash | AutoV2 33A9065DE4 |
A very simple merge of publicly available models, I'm putting it here mostly for my own reference. Mainly for characters, but it can produce some nice detailed backgrounds. I use inpainting and i2i upscaling for uploading on Pixiv, but for this repo I will just give example images of t2i + Hires.
Great for original characters, and it works with a good range of character LoRAs.
Both SFW and lewds are good.
Can also accommodate some stylistic or concept LoRAs, but it's a bit hit n' miss.
Let me know what you think! I'm always open to feedback, and happy to see what images you produce :)
Settings to get you started:
Clip skip: 2 (1 can be a bit chaotic but also valid)
Sampler: DPM++ 2M Karras (good starting point) / Euler a (can sometimes give better but more random images)
VAE: I like orangemix.vae but you can swap out for whatever you want. There is no baked vae.
Steps: 20-30 steps. You can go higher but it doesn't make too much difference (except with Euler a)
CFG scale: 5-7
Hires upscaler: 20 steps with around 0.5 denoising. Play around with upscalers to get your desired aesthetic
Negative prompts: EasyNegative, (bad-hands-5:0.8), (realistic, lips, nose, tooth:1.0)
Prompts: (masterpiece, best quality:1.1)
Enjoy :>
XV