When the model version is updated, I'll put new prompt in "About this version" if necessary.
V1.1
Prompt: lying,on stomach,on bed
Note:
OpenPose+Depth is necessary.
Hundreds image I tried, and only one acceptable. It would be easier to make this pose by TI/Lora or making openpose + depth by existing images. SD 1.5 just not working properly in this pose.
V1.0
OpenPose bone structure and example image with prompt information.
Prompt: lying on bed
Welcome to share your generation here!
Note:
Subject is usually posing "lying on stomach" while using prompt "lying on bed" in AnalogMadness V60.
"lying on stomach" looks like mission impossible for text to image and openpose. No single acceptable image after 800 examples.
Description
3rd lying pose.
Prompt: lying,on stomach,on bed,
OpenPose + Depth is necessary.
FAQ
Comments (2)
Haha, flying too close to the sun, eh? Maybe if a lora was just trained enough to help this specific pose, you could get them both working together, it would have to be a system. With a depth pose, your body types would have no variety.
This is the white whale of SD, no one has yet reliably achieved it, even in XL. It's actually a testament to how amazing our bodies are, even AI can't conceive of any more than the most basic physiological concepts that we can do.
Other than lying, SD is not good at "cross fingers/arms/legs", too. My experience is we cannot easily reproduce all these poses with controlnet.