Daz to SD (img2img) - 4 Portaits!

[KonoDioDa]

Not really except you are using an old version commint of the automatic1111, if you use newer versions (month ago onwards should be enough) the inference of your LORA training (even LyCoris too) should be ok to put all to models/lora, etc. No need for extension except very old commits.

[Frank21]

Thanks @Davizca87. I should just go ahead and install Kohya and see if it works. If I have SD already installed and running OK do I need to install anything other than Kohya for it to work? I'm using SD 1.5 which I installed about 2 weeks ago. Python 3.10.6.

[KonoDioDa]

@frank22: Theorically with a git clone install and installing all the dependencies should be enough. I think minimim. reqs. were Python 3.8 and I think Xformers (but this was included on Kohya installation).

For SDXL should be the same. I'm waiting for them to implement GLORA training (LORA on differente dimensions and actuations ranks) into Kohya to update my installation since should be very soon. GLORA > LORA in terms of aesthetics/quality.

[Frank21]

Thanks again @Davizca87. I already have xformers installed for SD. I'll have a go at installing Kohya tomorrow. Thanks for all your help mate😛.

[Author]

Posts