16
Carbon Flame Saber
(image.civitai.com)
What model do you use to generate the descriptions?
The descriptions are heavily edited JoyCaption Alpha Two descriptions.
Thank you
It gets a lot of things wrong, so you have to do a lot of corrections. It's impressive how well VLM captioned models work if they're trained on raw generated captions.
Share art created with Stable Diffusion
Also see: