JackAILab/ConsistentID

Expression editability

Opened this issue · 1 comments

During model training and testing, does the fine-grained description of facial local regions include expression information from the reference face? And this seems to weaken the expression editability of the model.

Hi, @tyrink Yes, the model is not currently optimized for expressions. We will release a version of the multi-image input after the paper is accepted. The new version parameters will be improved in terms of expression editability.