Aims to improve generation of Kiseru / Smoking PIpe poses and items.

Should support a wide range of outputs, and uses a gentle touch. I see good results with Pony-XL and derivative models.

The biggest issue I have with it is that fingers are less consistent then I would like, but I prefer the range of this, then I did with a more "baked" version with higher consistency.

v8 vs v10

v10 is an attempt to improve pose range so that it doesn't pull the actor back a standard "sitting pose" I think it improves at this, I think that v10 also has less resistance to styles and other checkpoints. as I get burn less often then I did with v8. The bad, I think v10 produces more multiple kiseru's more commonly, and its hardly immune to conflicts with other loras. My attempts to teach the lora handedness utterly failed, this might be because Pony itself doesn't really understand the concept, or because tagging of the base model conflates image position, and hand choice. I do think its an upgrade, but feel free to let me know what you think.

v10 Trigger words:

"Holding Kiseru" - primary word, you can try adding "in left hand" or "in right hand" and prefixing with "woman" or "man" using natural language, hand selection is not as functional as I would like and can lead towards backwards hands in some cases.

try "Holding smoking Kiseru" and/or "Pipe Smoke" to get smoke from kiseru,

"Blowing smoke" to improve chances of the actor blowing smoke.

v8 Trigger words:
"holding kiseru pose" - primary word.
"kiseru-smoke" - smoke emerging from the kiseru, as apposed to just being around the image.
"blowing smoke" - weaker then I would like, but it can work.

Duel Wielding can be an issue, try telling it what to do with the other hand.

(All Previews were generated with ComyUI)