search

RedAIGC / StoryMaker

StoryMaker: Towards consistent characters in text-to-image generation
411 stars 35 forks source link

about position embedding #13

Closed ChdDongyang closed 3 days ago

ChdDongyang commented 4 days ago

Hi, thanks for your excellent work! Your paper mentioned “These embeddings are concatenated and augmented with positional embeddings ,i.e., Epos, which serve to distinguish different characters. ”，I want to know how the positional embeddings are designed here and how I can get it. Thanks very much!

RedAIGC commented 3 days ago

Thanks for your attention, we just utilize a learnable parameter to represent the positional embeddings to distinguish different characters, please ref the code at https://github.com/RedAIGC/StoryMaker/blob/main/ip_adapter/ip_adapter_faceid.py#L73

ChdDongyang commented 3 days ago

Thanks for your attention, we just utilize a learnable parameter to represent the positional embeddings to distinguish different characters, please ref the code at https://github.com/RedAIGC/StoryMaker/blob/main/ip_adapter/ip_adapter_faceid.py#L73

Thank you very much!