Type | |
Stats | 122 |
Reviews | (9) |
Published | Mar 14, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 |
Hash | AutoV2 5F7CFAC510 |
Exchange group 820725763
Recommended parameters:
Between CFG 5-15, steps between 20-40.
Use Euler or DPM++ SDE Karras for sampling.
Never use DPM fast, DPM++2M SDE, DPM++ 2M SDE Heun
HD restoration is recommended with R-ESRGAN 4x+ Anime6B
0.5-0.7 repaints, 0.3-0.5 repaints for Tupelo.
The resolution is recommended not to exceed 1024, for bigger resolution please use HD restoration.
------------------------- Testimonials ------------------------------
I was afraid of being called a mold melting brat, so I was like a bolt from the blue after the Dominant World 4.2 review results came out.
The models I had squeezed in on colab were so bad.
I was about to give up on the big models, but when I saw that GhostInShell and Vientiane Furnace had updated their SDXLs, I decided to keep working on Dominant Worlds.
I then decided to go ahead and make the SDXL version of the dominant world.
For the SDXL version, I've tried to make it as close to SD1.5 as possible.
This time, I'm still working on generalizability and have experimented a lot with lora compatibility and tag compatibility, so I trust its quality.
The test runs were all done with a single 2060, so there's no way to do in-depth testing.
So there may be a lot of unknown issues with the model, so feedback is welcome (does anyone really have any feedback on issues 233)
The previews were all drawn with the local AUTOMATIC1111 webUI, and can be reproduced by importing the images in the png info section.