This LORA was created for the Style Capture &amp; Style Fusion Contest <a target="_blank" rel="ugc" href="https://civarchive.com/articles/2755">https://civarchive.com/articles/2755</a>Check out the collection on this link <a target="_blank" rel="ugc" href="https://civarchive.com/collections/85024">https://civarchive.com/collections/85024</a>Forgive me for double posting, Im not familar with the UI of CivitAI and didnt realize you could post two different versions of the same model on the same page so I accidentally made two posts an SD1.5 and SDXL versions. Too late now, I guess I'll just keep it up.This is the SDXL version of my SD1.5 LoRa of the same name here&gt; <a target="_blank" rel="ugc" href="https://civarchive.com/models/181359">https://civarchive.com/models/181359</a>. For further details check out the link to the original/first version.I give up on SD1.5, I am absolutely unable to replicate the style of this artist in SD1.5, seems like the SD1.5 model just refuses to cooperate with me. But with SDXL the style transfers over flawlessly and effortlessly, it was a breeze to work with, while trying to style transfer in SD1.5 is just a huge a headache. I highly recommend using this version instead of SD1.5 if you are capable of using SDXL. The style transfer is perfect on SDXL, while on SD1.5 its barely noticeable. This goes for all LoRas, not just this one, SD1.5 is just a pain to create accurate LoRas for.USAGE: 1. Please use &lt;lora:Austin_Osman_Spare_1.1:1&gt;, a setting of 1 is the default tuning and you will get the most accurate style results with a setting of 1.2. A CFG of between 5-8 and a denoise of 0.5-0.8 seems to be the most accurate true to style settings for this LoRa style if you want to stay true to the style, if not, go crazy. 3. "CLIP SKIP 2" work very good with this LoRa, I recommend using it. Use 1 for ultra realism, 2 for cartoon/paintings.

Austin Osman Spare SDXL

1.1 SDXL 64 images trained of artist of title. All 64 training pics where manually captioned by me.

Files with SHA256: c48ba27ca70f031fa40d2cdc938c5c6481ab486a6c9a2249c1af5285dbffec5a