Recommended generating at 640x896.
Trained at TF32 precision and saved as float. Still 128 dim size.
Use AOM2
Note: Feel free to see the metadata of my lora I included the full training cmdline if you wish to try and make it yourself.
Description
Well it's really 1 Epoch but 27 repeats.
Dataset used all Takamichi's Comic LO covers numbering 229.
Trained on AOM2 hardcore as that was the most popular.
This was a test train. But it turned out quite well.
Edit: I guess civitai does not save file names when you upload it. The file was supposed to be called CoLOv5t_fp32.safetensors as shown in the example images.
FAQ
Comments (2)
how to use? i tried putting it on the LORA folder and i get this message "Couldn't find Lora with name CoLOv5t_fp32"
CoLOv5t_fp32 from the images here? That's what I had it called. It's still the same one you download here. Also I thought civitai kept file names... I guess not. Anyways just rename it to that and you should be good