revert to applying mask before denoising for k-diffusion, like it was before

        self.sampler = sampler

        self.model_wrap = None

        self.p = None

        self.mask_before_denoising = False

    @property

    def inner_model(self):

        assert not is_edit_model or all(len(conds) == 1 for conds in conds_list), "AND is not supported for InstructPix2Pix checkpoint (unless using Image CFG scale = 1.0)"

        if self.mask is not None:

        if self.mask_before_denoising and self.mask is not None:

            x = self.init_latent * self.mask + self.nmask * x

        batch_size = len(conds_list)

        else:

            denoised = self.combine_denoised(x_out, conds_list, uncond, cond_scale)

        if not self.mask_before_denoising and self.mask is not None:

            denoised = self.init_latent * self.mask + self.nmask * denoised

        self.sampler.last_latent = self.get_pred_x0(torch.cat([x_in[i:i + 1] for i in denoised_image_indexes]), torch.cat([x_out[i:i + 1] for i in denoised_image_indexes]), sigma)

        if opts.live_preview_content == "Prompt":

        super().__init__(sampler)

        self.alphas = shared.sd_model.alphas_cumprod

        self.mask_before_denoising = True

    def get_pred_x0(self, x_in, x_out, sigma):

        ts = sigma.to(dtype=int)