Replies: 2 comments 2 replies
-
That's an interesting suggestion indeed. Handstands and the like are |
Beta Was this translation helpful? Give feedback.
0 replies
-
I am conducting some experiments, but they appear to be challenging. Rotating the Latent midway results in images consisting solely of noise. There may be a crucial oversight, or it could fundamentally be unfeasible. |
Beta Was this translation helpful? Give feedback.
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
When faces are not in normal orientation, stable diffusion tends to mess them up.
I suggest if there is a way to feed a rotated latent to SD depending on the mask. This way, the base prompt or the controlNet will create the rotated face/hand/etc.., while the masks or the ADDCOL will have the face presented in the best suited SD orientation.
Example prompt:
faces from below. Friends circle
Beta Was this translation helpful? Give feedback.
All reactions