-
Notifications
You must be signed in to change notification settings - Fork 44
Open
Description
It is difficult to accurately describe the multi-reference image model when writing prompt words. For example, the model cannot understand the person on the left and the person on the right very well.
Nowadays, many commercial generative models can already achieve subject identity control, such as the picture below.
You can give each reference image an identity name and use the identity name directly in the prompt word. I wonder if XVerse has such a capability?
Reference image identity name, prompt word example:
小明 拿着手机与 林黛玉 悟空 图1 坐在跑车上自拍, 图1 驾驶着车
Metadata
Metadata
Assignees
Labels
No labels