<img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e6ffb064-605c-4000-107a-59af0d693200/width=525/e6ffb064-605c-4000-107a-59af0d693200.jpeg" />I've decided to release this to gauge interest in future releases of styles and artists I've been working on. Please don't rip this off as your own, or do anything crazy with it. I am not releasing it for professional or commercial use, this is just for personal use and non-commercial research purposes. Please respect Patrick and his legacy, RIP. And please share your results!<img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8214ae92-932b-4d3b-cf69-7d06edbe1e00/width=525/8214ae92-932b-4d3b-cf69-7d06edbe1e00.jpeg" />The most decade-defining artist of the 1980's, Patrick Nagel has always been a major personal inspiration and I've decided to share my model for anyone who also enjoys his infamous style. The training data in all of the SD models on his work is really awful, so this fixes that and then some. This is my first time training a LoCon LyCORIS and it works quite well. <a target="_blank" rel="ugc" href="https://github.com/KohakuBlueleaf/a1111-sd-webui-lycoris">You will need an additional extension to use this! </a>(I know there was a TI posted for him awhile ago on here, but it's quite outdated already)<img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f56cd7ed-36ee-4019-2675-ce324ec2da00/width=525/f56cd7ed-36ee-4019-2675-ce324ec2da00.jpeg" /><ul><li>Use "patrick nagel" as your keyword.</li><li>This stacks really well on some models, not so much on others. I recommend <a target="_blank" rel="ugc" href="https://civarchive.com/models/4384/dreamshaper">Dreamshaper v5</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/22922/lyriel">Lyriel</a>, and <a target="_blank" rel="ugc" href="https://civarchive.com/models/4823/deliberate">Deliberate</a> from my own experiences. Play around with it on your favorite models. It also stacks stylistically with some anime models.</li><li>Even though it was trained on the 1.5 base model, it looks pretty bad overall on top of it. I recommend using a trained checkpoint.</li><li>This works very well with flexible LoRAs, so feel free to stack them!</li><li>I trained this on clip-skip: 2. I've tried it with 1 and 2, and it seems like a complete toss-up.</li><li>I trained this at 1024, so feel free to push up your resolution if your card supports it. It should work at any resolution though.</li><li>I usually kept it at :1.0 but you can tone it down if it's too stylized.</li><li>For deeper skin tones, use "white skin" and "pale skin" in negative prompts.</li><li>I kept some framing in training data, because his work was almost always presented with borders. Crop yours manually or use "frame, border" etc. in negative prompts if you don't want them.</li><li>People come out better than objects or scenes, but it can still be done. Negative prompt "person" and "face" if you have people popping up.</li></ul>

Files with SHA256: cb1d84da6d5ec465027e3bb116701a50b7acf2638d7a5db359572cffc96e0123