For controlnet openpose/depth.
Please read the version details for version-specific information.
Thigh focus: Better with depth
Looking/facing away: Better with openpose
Description
17 poses (plus some variants) for controlnet where the character is facing/looking away from the POV.
In the .zip files: example generation plus the .png to use in openpose and depth models. Added .json for openpose too so you can make modifications to the pose file. 110 files total (plenty of example generations in there).
The openpose files are very useful for this version. Depth doesn't give you much flexibility.
Version notes:
As always, aspect ratio is crucial to the output.
The prompt still matters, including descriptors such as "from_below" etc.
The fewer data points there are in the pose file, the lower the control weight should be. If there is just the face, go between 0.8 to 1.0 (you can even experiment with lower values). I found the files that have only face data points the most useful. The model fills in the blanks well.
All pictures were generated with the Comimicry model (https://civitai.com/models/105172). DPM++ 2M Karras sampler.
Lora used: 3D Rendering Style (https://civitai.com/models/73756/3d-rendering-style) v1.4.1
Prompt and other generation data: see attached pictures.
FAQ
Comments (2)
how do i use these?
Check out https://youtu.be/WZg3e6B2yPQ for a useful tutorial for SD1.5 in Automatic1111. I'm not familiar with the ComfyUI setup, nor with other UI's.
Once you're set up, drop the pose .png into the controlnet area, preprocessor must be "none", select the openpose model (or the depth model if it's a depth image) from the controlnet dropdown, write a prompt and hit generate.