[Stable Diffusion Inpaint] Fix incorrect noise addition

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -32,6 +32,9 @@ def preprocess_mask(mask):
     mask = np.array(mask).astype(np.float32) / 255.0
     mask = np.tile(mask, (4, 1, 1))
     mask = mask[None].transpose(0, 1, 2, 3)  # what does this step do?
+    # make sure mask is properly valid
+    mask[mask < 0.5] = 0.0
+    mask[mask >= 0.5] = 1.0
     mask = 1 - mask  # repaint white, keep black
     mask = torch.from_numpy(mask)
     return mask
@@ -66,7 +69,7 @@ def __call__(
         prompt: Union[str, List[str]],
         init_image: Union[torch.FloatTensor, PIL.Image.Image],
         mask_image: Union[torch.FloatTensor, PIL.Image.Image],
-        strength: float = 0.8,
+        strength: float = 1.0,
         num_inference_steps: Optional[int] = 50,
         guidance_scale: Optional[float] = 7.5,
         eta: Optional[float] = 0.0,
@@ -177,8 +180,12 @@ def __call__(
             latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)["prev_sample"]
 
             # masking
-            init_latents_proper = self.scheduler.add_noise(init_latents_orig, noise, t)
-            latents = (init_latents_proper * mask) + (latents * (1 - mask))
+            if t > 1:
+              t_noise = torch.randn(latents.shape, generator=generator, device=self.device)
+              init_latents_proper = self.scheduler.add_noise(init_latents_orig, t_noise, t-1)
+              latents = init_latents_proper * mask + latents * (1-mask)
+            else:
+              latents = init_latents_orig * mask + latents * (1-mask)
 
         # scale and decode the image latents with vae
         latents = 1 / 0.18215 * latents