15 Replies
DDIM 50 steps
clearly euler A is much better
DDIM has to put more steps to compete, like 100. on 50 steps the different parts of the prompt doesn't mix well yet
DDIM 100 steps
@Dr. Furkan Gözükara
How to use civit ai models for training in shivam's colab?
In your previous YouTube you showed how to use civit ai model for image generation in Shivam colab not for training.
i freaking love this image 😆
on
Settings and run
you can check the save_to_gdrive option so the drive is mounted in /content/drive, and then specify the path of the model in MODEL_NAME@felixsanz ok do I need to put .safetensor or .ckpt or weights or what?
probably all will work. try it and see
diffusers weights or safetensors
Should I comment this line or keep as it is --pretrained_vae_name_or_path=
does anyone know if A1111, kohya_ss and other Projects perform (in terms of speed or memory efficiency) well in WSL2 rather than windows? would WSL2 be better for doing Dreambooth finetunings and Lora trainings?
you can use another vae, or not. that's up to you. i would use the default that comes with the model
@felixsanz
Ok i put .safetensor file in my drive but now it says no config.json file found in directory
so convert it to diffusers format
Sorry but how to do that is that same explained in video
Is there any other way?
what model you're trying to train?
Realistic v3
It is available on both place civit ai and huggingface