Anybody got a loose idea on how the the developer is generating his art in the way he does? I mean it in how he generate the busts of the characters, I just wasted all of my credits on civitai trying to do the same but I can never achieve a similar result in terms of quality. I'm not sure it's the Lora or prompt but it always comes wrong, maybe he used a image as reference, but I wouldn't know.
the simplest way would just be to get stable diffusion a1111. It’s old but it’s by far the most user friendly and you don’t need credits to use and runs locally off your own computer. Then his models just look like wai-illustrious generations with basic character Lora’s. Probably just make it a grey background so it’s easier to contrast and apply a remove background function to it. Not using an illustrious model would probably be your issue if you’re struggling.
Positive prompts probably the generic: best quality, high quality, masterpiece, “character name”, solo, whole body, “pose”, “body descriptors”, “hair descriptors”, “clothes”, , grey background.
Negative prompts:
Worst quality, low quality, censored, watermark, text.
If you’re struggling with hands just add it to the negative prompt.
He probably then just in-paints different facial expressions. Or keeps the same seed and changes the expression descriptor.