instruct-pix2pix
instruct-pix2pix copied to clipboard
Why two same negative prompt
Hi,
I'm wondering the purpose of using two same negative prompt embedding.
prompt_embeds = torch.cat([prompt_embeds, negative_prompt_embeds, negative_prompt_embeds])
And you set
noise_pred = ( noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_image) + self.image_guidance_scale * (noise_pred_image - noise_pred_uncond) )
In this case why not merge the scale factor?
I don't know how to set it up.