Type | |
Stats | 813 377 |
Reviews | (137) |
Published | Aug 4, 2023 |
Base Model | |
Hash | AutoV2 DCE7E29013 |
This is a STYLE lora and it's experimental
Now many sample images people uploaded have to be done in the comfyUI. That's... just not good. Miss the good old days when we got a nice pic, we dragged it to the ui and reproduced it easily.
The base model is 国风4 GuoFeng4 XL V1.0 which I listed in Suggested Resources. (Note that it's an early version). It is a great model. You can get a sharp image in 512x768 while not using the refiner or Hires. Light as a feather. Like what we did in SD1.5.
I'll test other models later. Everything's getting bigger and heavier in SDXL.
UI: a1111
Sampling steps: >40
Lora Weight: around 0.9
Resolution: 512x768, 512x1024, 1024x1024 or higher
Sampling Method: most samplers are compatible; you can find it in the samples.
(I didn't figure out why we need a refiner)
Hires.fix:
Upscaler: ESRGAN_4x
Denoising strength: 0.3
Hires steps: ≥10