[Feature] Infer pose from an image

[flesler]

The other pose editor extension, fkunn1326/openpose-editor, supports pasting/uploading an image and it infers the pose skeleton from it (not very successfully most of the time). Same is done, kind of, by the openpose preprocessor from ControlNet (quite successfully).

It would be a killer feature to add that here. IMO it's hard and time-consuming to arrange a pose from scratch. Hopefully you can take code from somewhere already, you won't be inventing this from scratch

Personally my main workflow would be:

1. Upload image, get a pose
2. Fix it
3. Move/Zoom it around
4. Use it

[RoyRiv3r]

I endorse this

[RJSprod]

This is also on my highly desired list.

Also the ability to send image to open pose to extract the pose from generated TXT2IMG/IMG2IMG results.

I often find that I capture the essence of what I want with a seed, but want to just change the pose a little bit and run again with the same seed. Example, generating an image with everything perfect but one limb is doing something strange, I just want to nudge the output to be more exactly what I want. Little bit of pose editing should fix.

Control net already has a feature to import an image and preview its pose map, so should be more efficient to take from control net than create another process. When in the webui, something like "send to control net" and inversely "send to pose-x" so the two extension can share to each other.

[aschmitz13]

Can I use the pose estimation from control net in posex? I have tried to save the picture, but that seems not to work. it says img name is missing.

[Slight0]

I would say this tool is nearly useless without this feature. Why go through all this effort to pose by hand when there are external web apps that have much better tooling to do that in half the time?

This is useful as a "slight adjustment" sort of workflow when starting from an existing pose image you have.