Works well in base and edit both. Strength recommended 1.0 - set higher if anatomy isn't great.
Prompt framework: [trigger prompt], [optional: gag prompt], [pose]
Trigger prompt: woman is encased in [color] tape bondage from [neck/shoulders] to [ankles/toes]
Gag Prompt: gagged with [color] tape wrapped around her mouth
Pose: mostly trained on "standing" , "lying on [bed/floor/couch]", "sitting on [bed/table/chair]"
If you use sitting on floor or other things results could be iffy
Description
Works well in base and edit both. Strength recommended 1.0 - set higher if anatomy isn't great.
Prompt framework: [trigger prompt], [optional: gag prompt], [pose]
Trigger prompt: woman is encased in [color] tape bondage from [neck/shoulders] to [ankles/toes]
Gag Prompt: gagged with [color] tape wrapped around her mouth
Pose: mostly trained on "standing" , "lying on [bed/floor/couch]", "sitting on [bed/table/chair]"
If you use sitting on floor or other things results could be iffy
Provide feedback if something doesn't work.
FAQ
Comments (6)
Hello. If you train your LoRAs to F2Klein you'll probably see your downloads skyrocket! Let me explain my thinking : the weights of the Qwen models and their usage are not really consumer-gear-friendly, F2K is! So the number of users is on a totally different scale, way larger! Cheers and good work 🙌 (even if I can't use them ahah 😂)
Thanks for the suggestion, I will look into it to see how it performs.
In my brief experience with Klein, it was more restrictive than Qwen due to the way Flux locks down NSFW in the base models, Qwen is a lot more “free” with that and in terms of training it understands nsfw a ton better. But I haven’t trained Klein models, only base Flux ones so maybe it’ll be different.
As for running Qwen on consumer grade GPUs, you can still get it to work with GGUF models and workflows, even if it’s a bit slower it’s worth the trade offs if you’re interested in going that route.
FYI: There's an issue with the arms not being included in the body wrap.
Thanks for the feedback. If you run into that issue usually 1 of 2 things - either its a pose issue (standard trained poses are standing/laying and some angles for sitting) - or crank up the strength a bit to see if that fixes it.
Let me know if there's a particular pose this is happening on maybe post a sample pic?
@vackyadmiral167 I think I now know what causes the problem. The basic prompt works just fine, as expected. However whenever I use a reference photo and append the prompt with "Use the woman from image 1 as the woman" then the arms are not properly wrapped. This is the first Lora I've used that doesn't seem to like using a reference image. If I increase the Lora strength all the way up to 2.0, then it works with the reference photo. However, as you might suspect, a strength of 2.0 has undesirable side effects.
@tomthumb8 hey thanks for the detailed feedback. Have you tried just referencing your image but without "image 1" - i use it that way just natural language that "the woman is blah blah...." and usually gives me better effects. I use it on edit a lot and it (1-1.2 strength) works well for me in the default poses; sometimes if you append 2 images of subjects whether to merge or same subject (different angles) to better the output (which works well on some loras) - essentially redrawing the character/subject from multiple images breaks it more than just using a singular image of a reference subject. Let me know if this way works for ya or if its still broken.