If you do like this work, consider buying me a coffee :)
---
Who said SDXL couldn't do anime?? Following on from the recent Quiet Goodnight model for SD1.5, I thought it would be cool to release a SDXL model based off the lora from the same dataset, so here it is! Credit to @Lykon and @xDivine as it started with the AAM XL and Blank Canvas XL as the base with a wacky merge method (auto-MBW, then fold them back upon themselves) and then merged in the Horde HQR and SDXL_Divine_RC loras in to get the magic. For v2, this then got a negative alpha merge with a merge of ICBINP XL versions to move it further away from realism.
The samples were done on Forge using the Restart or DPM++ 3M SDE Exponential sampler, 20-30 steps, CFG 3-6, and 832x1216 or 1024x1024. Hi-res fix and adetailer not required but recommended.
Description
Made with a MBW mix of ICBINP XL v5a and v2, then combined with a flat -1.2 alpha merge with Quiet Goodnight v1
FAQ
Comments (2)
Was this trained with AI input?
Apologies - there's no training in v2, just in v1 - which was manually captioned