search

HL-hanlin / Ctrl-Adapter

Official implementation of Ctrl-Adapter: An Efficient and Versatile Framework for Adapting Diverse Controls to Any Diffusion Model
https://ctrl-adapter.github.io/
Apache License 2.0
381 stars 16 forks source link

SDXL checkpoints #2

Closed scarbain closed 5 months ago

scarbain commented 5 months ago

Hello, thanks for the amazing paper and the release of i2vgenxl_depth !

When do you plan on releasing the models for SDXL ? Which one have you already trained ?

Also, do you have an idea on when you'll release the training code for SDXL ? :)

HL-hanlin commented 5 months ago

Thanks for your interest in our work!

Yeah I'll working on cleaning the code for SDXL and other backbones in the next few days. Hopefully I can release the inference code and model checkpoint for SDXL this week, and I'll see if I can also finish cleaning the training code.

1585511010 commented 5 months ago

release this week？But I do not find it .

HL-hanlin commented 5 months ago

Working on it now. Just a few hours. Thanks!

HL-hanlin commented 5 months ago

SDXL inference code & checkpoint uploaded. Thanks for your patience!

1585511010 commented 5 months ago

could you please ask you when you release training code and training guideline to adapt our Ctrl-Adapter to new image/video diffusion models.

HL-hanlin commented 5 months ago

Hi all, Thanks for your patience! The training code for SDXL has been uploaded. Please let me know if you find any issues!

scarbain commented 5 months ago

That's amazing, thanks !!

Can you please provide the non-default hyperparameters you used for training the depth SDXL CTRL-Adapter ? Such as LR, warmup, precision, decay, ...

HL-hanlin commented 5 months ago

Hi, the hyper-parameters we used are all contained in the configuration files and training scripts. So no change needed! You can just run the scripts directly :)

scarbain commented 5 months ago

That's perfect, thanks Han!