Spaces:
Running
on
Zero
Weird posing behavior
I have tried adding poses both in Gradio environment and in my own cloud environment, and in all cases posing only changes the face placement, not the actual pose of the character in the generated image. Without the pose, the face is always placed at the same place in an image, which produces a weird fish-eye effect in some images as if the character is very close to the viewer. The pose is always randomized, even if a full body photo is used for reference.
Here are some examples of images without posing:
And here are examples of images with the same pose:
Only the face stays in the same position, everything else changes, even if the prompt is unmodified and doesn't contain any information on the positioning of the character.
The model I'm using is Animagine XL 3.0.
its pose just as a kps, not a real pose. you need another pose controlnet yo control it
Yes. We should make it clearer that the pose only stands for facial landmarks. For body pose, you should use another ControlNet.