ali-vilab / AnyDoor

Official implementations for paper: Anydoor: zero-shot object-level image customization
https://ali-vilab.github.io/AnyDoor-Page/
MIT License
4.01k stars 366 forks source link

Its not working #108

Open srossitto79 opened 1 month ago

srossitto79 commented 1 month ago

Hello, i installed AnyDoor, lets not speak about the fact that all needed models where already on my system but it still downloaded them all and in the readme you describe this as a feature. I tried to pur the suit on a person in a photo on another person in another photo, simple case, like in the examples, the image was 1200x800+- the result was just a mix of colors. Am i doing something wrong? are the images on the website reproducible because i cant reach the same results. thanks in advance.

XavierCHEN34 commented 1 month ago

Could you give me your examples of inputs? A common problem is that some users forget to input the segmentation masks of the reference object, you should provide both of the reference mask and target mask