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
Hey @Zeju1997 , first of all thanks for a very interesting paper, haven't seen such a good one for long time.
I think i don't clearly understand what you do in your Controlnet experiments. in the paper you mentioned that ContorlNet conditions are passed through CN encoder and concatenated to SD input, according to the code this is what happens in oft/oft-control/oldm/oft_ldm.py but in original CN the conditions are added as residuals during forward rather that concatenated to SD. Why did you deviated from original CN implementation? Is it only to be able to actually fine-tune SD weights to follow control and be able to apply OFT to this process? (since in original CN implementation i think your method couldn't be applied)
The text was updated successfully, but these errors were encountered:
Hey @Zeju1997 , first of all thanks for a very interesting paper, haven't seen such a good one for long time.
I think i don't clearly understand what you do in your Controlnet experiments. in the paper you mentioned that ContorlNet conditions are passed through CN encoder and concatenated to SD input, according to the code this is what happens in
oft/oft-control/oldm/oft_ldm.py
but in original CN the conditions are added as residuals during forward rather that concatenated to SD. Why did you deviated from original CN implementation? Is it only to be able to actually fine-tune SD weights to follow control and be able to apply OFT to this process? (since in original CN implementation i think your method couldn't be applied)The text was updated successfully, but these errors were encountered: