-
|
I am trying to do something simple, inpaint a hand on the sholder of a character, so I have this image, I put it on the inpaint tab, masked the shape of a hand, increased the denoising strenght to 1, inpaint masked, and no hand appears, I tried like 20 times and nothing is inpainted, before I think I made this work, but now doesn't matter what I do, it just doesn't show what I need, I tried to change inpaint conditioning mask and also nothing changes. Does anyone knows if I am doing something wrong? The image What I get
|
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 5 replies
-
|
This model is extremely context-aware. I.e. it relies on the information on the picture, the closer it is to the edge of mask, the more important info it is. It sees only shirt all around itself and you cannot force it to draw a hand. Therefore it's very difficult to inpaint something that is completely not there and cannot be assumed by context with this model (You can inpaint nose on voldemort's face without any problems). it works great when you need to edit parts of already existing objects. Proposed workflow: add what you want as a sketch, refine it in some non-inpaint model so it looks normal, then fine-tune with inpainting. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for your answer, I understood, it's a bit sad that it can add this concept on the image without any input, but it's good to know better, thank you again! |
Beta Was this translation helpful? Give feedback.
-
|
@leonnn1 Inpainting model shows unreal results when used for outpainting because of context-awareness. |
Beta Was this translation helpful? Give feedback.
-
|
I see, it's still tricky sometimes to make it work as you want right, anyways thank you for your time and help! |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
I want to read the scribbled article |
Beta Was this translation helpful? Give feedback.
This model is extremely context-aware. I.e. it relies on the information on the picture, the closer it is to the edge of mask, the more important info it is. It sees only shirt all around itself and you cannot force it to draw a hand. Therefore it's very difficult to inpaint something that is completely not there and cannot be assumed by context with this model (You can inpaint nose on voldemort's face without any problems). it works great when you need to edit parts of already existing objects.
Proposed workflow: add what you want as a sketch, refine it in some non-inpaint model so it looks normal, then fine-tune with inpainting.