You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello. Initially, I wanted to create a pr for this task. However, in the process, I realized that the amount of changes needed is more than what should be in the public pr. Anyway, I have started the process and hope this pieces of code will be useful.
I redesigned cross-attention blocks from the author's code to diffusers and made the weights loading. I added changes to transformer, pipeline. Here is the test on one person, in my work I did the test on several persons, and everything worked as expected.
In my opinion, we need to create a 'LoaderMixin' like for IP adapter. The main problem is that cross attention with id_emb in PuLID is done differently compared to IP adapter it is separated from cross attention with text, and done after the whole block. Because of this, I do not know how best to proceed, to create a separate class like 'IPAdapterLoaderMixin', only for PuLID or to embed in the logic of the IP adapter.
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.
Please note that issues that do not follow the contributing guidelines are likely to be ignored.
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.
Please note that issues that do not follow the contributing guidelines are likely to be ignored.
Activity
maxs-kan commentedon Dec 19, 2024
Hello. Initially, I wanted to create a pr for this task. However, in the process, I realized that the amount of changes needed is more than what should be in the public pr. Anyway, I have started the process and hope this pieces of code will be useful.
I redesigned cross-attention blocks from the author's code to diffusers and made the weights loading. I added changes to transformer, pipeline. Here is the test on one person, in my work I did the test on several persons, and everything worked as expected.
In my opinion, we need to create a 'LoaderMixin' like for IP adapter. The main problem is that cross attention with id_emb in PuLID is done differently compared to IP adapter it is separated from cross attention with text, and done after the whole block. Because of this, I do not know how best to proceed, to create a separate class like 'IPAdapterLoaderMixin', only for PuLID or to embed in the logic of the IP adapter.
newideas99 commentedon Dec 31, 2024
Thank you for doing this!! Any luck with adding lora support?
github-actions commentedon Jan 25, 2025
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.
Please note that issues that do not follow the contributing guidelines are likely to be ignored.
nitinmukesh commentedon Jan 25, 2025
This is interesting feature to add in diffusers
github-actions commentedon Feb 19, 2025
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.
Please note that issues that do not follow the contributing guidelines are likely to be ignored.
bghira commentedon Apr 22, 2025
not stale