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

1 files changed, 5 insertions, 1 deletions

diff --git a/modules/sd_samplers_cfg_denoiser.py b/modules/sd_samplers_cfg_denoiser.py
index 113425b2..bc9b97e4 100644
--- a/modules/sd_samplers_cfg_denoiser.py
+++ b/modules/sd_samplers_cfg_denoiser.py
@@ -56,6 +56,7 @@ class CFGDenoiser(torch.nn.Module):
         self.sampler = sampler

         self.model_wrap = None

         self.p = None

+        self.mask_before_denoising = False

 

     @property

     def inner_model(self):

@@ -104,7 +105,7 @@ class CFGDenoiser(torch.nn.Module):
 

         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)

@@ -206,6 +207,9 @@ class CFGDenoiser(torch.nn.Module):
         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":