Check out the SDXL version of this LoRa at &gt;&gt;&gt;<a target="_blank" rel="ugc" href="https://civarchive.com/models/182675/austin-osman-spare-sdxl">https://civarchive.com/models/182675/austin-osman-spare-sdxl</a>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>WARNING: This LORA was trained on standard SD 1.5 model. YES!, that very old default model that comes packaged standard with SD1.5. You will get the absolute most style transfer if you use that model. Example pics were not made on SD1.5 default model but the style transfer is still noticeable.NOTE: Keep in mind that all my example pics are created with RealisticVisionV5.1 or a realistic checkpoint, so it is not an accurated representation of Austin Osman Spares work, as his paintings/drawings are not realistic. The style does blend to realistic stuff though. RealisticVision and other 1.5 realistic checkpoints are HIGHLY biased towards realism. You're better off using a cartoony checkpoint for more painting like features.Trained on 32 Artworks of totally unknown artist Austin Osman Spare(United Kingdom English artist 1886-1956). I picked a dead artist to use cause most alive artists get their panties in a bunch when they find out their artwork is used to train AI; dead guy wont mind. Im using CivitAIs training feature that uses 500 "Buzz" tokens. My first ever Lora...cause Im too lazy to learn how to use KohyaSS or any other Lora creator lol.I didnt realize how much time I was gonna take to make such a small LORA, 12 hours to make this. Research, finding pics, learning how to make a lora, creating LORA example pics, and testing it took a lot of time.I used <a target="_blank" rel="ugc" href="http://stable-diffusion.site/image-to-prompt/">stable-diffusion.site/image-to-prompt/</a> to create prompt ideas for my training dataset images. I coulda used the built in clip from my stable diffusion a1111 but I was too lazy to download it, and using the website saved me some electricity as I didnt need to run my GPU.A lot of the example pics seem to have a dark undertone, because Austin Spares artwork does have a dark and occult like undertone. Im a huge fan of dark fantasy/occult/esoteric art, so I googled for a dead occult artist and his name popped up, so I picked him since he is realatively unknown and so his training data is least likely to exist in the standard stablediffusion models."Austin Osman Spare" without the quotation marks was used in the training dataset. So it may be a trigger word.Suggestion: Use (Austin Osman Spare:1.2) trigger word to increase its effect along with Lora strength. Play around with CFG, denoise, and clip skip. It drastically alters the art, and Austin Spares art is supposed to be extemely surrealistic if you want to be true to form. My example pics are extremely conservative.

Austin Osman Spare SD 1.5

64 trained images at higher epoch and steps than v1.0

Files with SHA256: 9c6d52c3c1be9cd90a35ebc9fbe782677aea45ccdbfb4d8f2725f53f8a070fe7