Current pose estimation solutions for T2i Keypose model

[AugmentedRealityCat]

To work with the Keypose model for T2i you need to feed it an image that looks like the one below, with colored bones over a black background, and nothing else. The colors and the overall structure according to which the bones are attached together is essential for the system to understand the drawn pose.

As there is no Keypose pre-processor included with the ControlNet extension, the user must use one of the examples available online, or create his own either by drawing them manually or by arranging a preset 3d skeleton in Blender using the wonderful preset made by @toyxyz . You can access his free blender preset over here - it supports the Keypose format since version 4.5.

If you want to use a pre-processor to extract poses from pictures the only fast, one-click, easy solution I've found is the T2i online app demo on Huggingface. Here is the link:
https://huggingface.co/spaces/Adapter/T2I-Adapter

Go try it now, it works very well !

For more information about Keypose and the difference between it and Openpose when used with Stable Diffusion, do not miss @toyxyz twitter thread about this - it's gold !

https://twitter.com/toyxyz3/status/1630604817566691329?t=vUPS_4GhCwmlV9B6WqCDjg&s=19