Support my work on Patreon and Ko-Fi and get access to tutorials and exclusive models.
Rate and leave a like if you enjoyed it, and follow for new models!
My base model Yawarakai is now available on my Patreon to Hardcore tier patrons.
How to use. Trigger: upside-down
Description
FAQ
Comments (10)
How did you do this? In general, it is so difficult for AI to understand these kinds of pose,such as lying on back from side, upside-down.
我也想問,能做到真的很神奇!!
yeah! how the hell? psoft obviously met the devil at the crossroads
yeah, I decided to have a go at it given how hard it is in general for SD to generate this kind of pose. It still isn't perfect, but I don't think any LoRA can fix the underlying issues with SD.
having a large enough, but also consistent enough dataset, that allows the repeated pattern to be learned (i.e. repeated pose, varied aesthetics and details) is the best way to achieve this kind of results.
great question, this issue really bothers me. I can't understand why stable diffusion can only "see" right side up. certainly there's no physics model built in. and it cannot be the case that SD had no training in images that contained upside down things. this is an intriguing problem indeed. it's also irritating. why should SD care if things are right side up or not? it's even a scary problem... how does SD even know what direction is "up" or "down" at all?
next up: psoft's The Upside Down from stranger things.
Every time I tried upside-down pose, the faces are always hell-like, AI just cant understand
yes, this LoRA helps with that a bit, but doesn't solve the underlying issues with SD
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.