PuLID
PuLID copied to clipboard
[NeurIPS 2024] Official code for PuLID: Pure and Lightning ID Customization via Contrastive Alignment
Thanks for the great work! I'm confused about the align loss mentioned in Sec 3.4 of the paper. As you adopt a 2-norm loss, the loss value can be large...
Hello, thanks for your incredible work! In the 'Accurate ID Loss' section in the bottom right corner of Figure 2 of the paper, there are two generated images both denoted...
 
Lid = CosSim(/phi(Cid), /phi(...))
including the train from scratch (tfs) and fine-tuning (ft), as well as the training dataset? Looking forward to hearing about your plans and reply. Your efforts and contributions to the...
Thanks for your work, result ganggangdi ! I'm doing research related on this right now, can you please upload the stage2(maximum ID sim) model?
Hi, I'm a developer working on SD and its relevant pipelines. PulID is a great tool for maintaining fidelity when generating new images. However, I find the code difficult to...
Thank you for sharing. Do you have plans for the multi ID input and controllnet?
Hello, when you calculated the layout loss and layout-sem loss, which cross-attention layers are the QKV features from? Do you use the features from all the cross-attention layers? I'm looking...