NoobAI-XL ControlNet

flip the image?

So is it solved, bro?

Depends on what UI you use. But a lot of UI's will feature either directly or as an comunity made extension some form of additional input for controllnets.

In ComfyUI I like to use the "Clip Text Encode Controllnet" node that allows to add some further conditioning/promting to the conditioning/prompt that goes in the "Apply Controllnet" node besides the pose image input. That way some more cocise information can be added to support the controllnet in generating the targeted result. So for a pose controllnet I add stuff that describes the upside down pose in question in that way if I want to generate one, because Controllnets struggling with these kinds of poses is not a rare phenomen. Especially when the Checkpoint used already might have not the best training data or understanding of the concept.

It also will help to improove results in certain fringe cases when limbs overlap in certain ways and the controllnet gets confused and starts to switch what tzpe of limb it is or contorts them in weird ways. A simple "*insert Limb* behind insert other Limb" added in like described has resolved that problem in many cases, or at least mitigated it.

It seems it works for me in Easy Diffusion (v3.5 engine) with this exact CN model.

Any idea if it works?

No idea what you're using, but you don't need any yaml files for this to work in ComfyUI