I've decided to release my latest cigarette smoking lora as a whole separate model from my existing smoking lora because this one is more stable but fundamentally more limited. Use the <code>cigsmoker</code> trigger to activate and be sure the rest of your prompt has a woman in it somewhere or you may get unpredictable results. Example prompt: <code>cigsmoker, a woman relaxing at the beach</code>I tend to use it with DreamShaper v6.31 and Realistic Vision v2 (as seen in the example images) The dataset includes only women, and most of those women are very scantily clad (though there are no explicitly nude photos in the dataset). This means it will likely produce a character that is normally fully dressed suddenly wearing lingerie, or in a pose that shows lots of cleavage. That wasn't my intent, but it's hard to find high quality smoking images with fully clothed women in them. You can try having it make a man smoking a cigarette, but you'll likely be fighting the model the whole way, and it will try to gender swap your character into a woman. This has many similar problems as my original smoking lora... multiple cigarettes in the mouth, or cigarettes don't line up quite right with the lips, etc, etc, etc. Unlike the original lora, it's bad at creating a cigarette in their hand and focuses only on images where the cigarette is in the woman's mouth. However, it produces a usable picture more frequently for me than my original lora. Hires fix will exaggerate and amplify the problems with this lora unless you turn down the denoising strength (0.5 to 0.6 usually). Oh, and don't smoke. Smoking is bad for your health.

Files with SHA256: 62ab450562b2a2ff0ff2aad70dcec34a1927a1906a5853b4f60eaee04324418a