Use Blender to pose your model and drive the pose in stablediffusion ControlNet - Pose.
By using b3d and rigify to generate the hands and finger data, it is possible to re-pose a prompt -generated character.
The control net model is so weird. I get why it has two spines, to indicate spine twist on a 2D surface, but why an unsubdivded spine? Surely, clavicles are less important than a subdivided spine.
Sorry, just learning about SD recently. I was originally thinking about how I could integrate it into Blender, but now I’m more thinking about how I can integrate Blender into it. (Depth, normal, openpose are all so error prone.)