Type | |
Stats | 227 |
Reviews | (23) |
Published | Sep 4, 2023 |
Base Model | |
Hash | AutoV2 053D136AA1 |
A general-purpose merge of some of the best SDXL models currently available. In my experience, it compares favorably to its constituent models.
Version 1 was created by performing the following sequence of merges:
50/50 mix of WyvernMix XL v1.1 and NightVision XL Alpha v0.6.5.0
70/30 mix of RunDiffusion XL beta and Dreamshaper XL1.0
75/25 mix of v1.0 Mix #1 and v1.0 Mix #2
Then, Version 2 was created by performing the proceeding further merges:
50/50 mix of FormulaXL v1.0 and Realities Edge XL v1.0
65/35 mix of SDVN7-RealArtXL beta2 and BriXL v1.0
65/35 mix of ProtoVision XL Beta v0.5.2.0 and ChimeraMix(XL) v1.2
60/40 mix of v2.0 Mix #2 and v2.0 Mix #3
75/25 mix of v2.0 Mix #1 and v2.0 Mix #4
50/50 mix of v2.0 Mix #5 and Edvard v1.0
Version 3 was created as follows:
60/40 mix of NightVision XL Beta v0.7.0.2 and WyvernMix XL v1.1
75/25 mix of v3.0 Mix #1 and v1.0 Mix #2
50/50 mix of v3.0 Mix #2 and v2.0 Mix #5
50/50 mix of Juggernaut v2 and Crystal Clear XL
60/40 mix of v3.0 Mix #3 and v3.0 Mix #4
I personally prefer to use the DPM++ SDE Karras sampler, the Real-ESRGAN X4+ upscaler, a CFG in the 4-6 range, and 30-45 steps, but your preferences may differ. HighRes Fix seems helpful. A refiner does not.
I've made no effort to improve SDXL's NSFW capabilities here, and I don't plan to.