[REQUEST] Release of control weights without SD1.5 model built in

[2blackbar]

Can You guys release just weights for controlling ? In webuit it pretty much ignores builting 1.5 fomr your models and is using our own so the issue is it loads SD1.5 twice and wasting VRAM...
I think weights with just control withoyut additional SD weights would still work, but would need to test it first...
Or is there a way i can extract just control weights on my own and reduce fiesize ?

[2blackbar]

Fixed with this but would still be nice to upload them for everyone else , the weights are at 1.4GB, it saves time and VRAM
Mikubill/sd-webui-controlnet#35

[lllyasviel]

Hi 2blackbar, the base model of in some ControlNet contains small offset with unlocked training:
See also

1. The point 1 in [Experiment] Transfer Control to Other SD1.X Models #12
2. The ending part of https://github.com/lllyasviel/ControlNet/blob/main/docs/train.md
3. The section 3.4.2 of https://arxiv.org/pdf/2302.05543.pdf

[2blackbar]

thanks! the code is pretty amazing and revolutionary for SD, was it trained at 1024? Can we use higher res ?

[Mikubill]

btw, model trained in unlocked mode could also be extracted as normal but it should pair with the original model to get intent/best result.

[2blackbar]

Well i used dreambooth model with likeness of a person and from what i see so far , extracted controlnet weights work better with that extension than controlnet+sd1.5, maybe it was affecting somethin not sure so it might be desired at times and at tother times it might interfere too much, i extracted most of the weights, cant believe how amazing this is, just bought 3090 cause of it .
Thanks a lot !!!

[FurkanGozukara]

@Mikubill your extension is just amazing. my kudos. made a tutorial on your extension i hope you accept my pull request

[geekyayush]

Hey @2blackba
How did you use dreambooth model with controlnet?

Could you help me to do the same please?
if you have any tutorial or any kind of guide I can follow, can you share here?

Thanks!

[FurkanGozukara]

Hey @2blackba How did you use dreambooth model with controlnet?

Could you help me to do the same please? if you have any tutorial or any kind of guide I can follow, can you share here?

Thanks!

it is easy. train your dreambooth model and select it when using control net
here my tutorial
https://youtu.be/vhqqmkTBMlU