Fixed mask+masked_image in sd inpaint pipeline (huggingface#1516)

* Fixed mask+masked_image in sd inpaint pipeline

Those were left unset when inputs are not PIL images

* Fixed formatting

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -632,6 +632,9 @@ def __call__(
         # 4. Preprocess mask and image
         if isinstance(image, PIL.Image.Image) and isinstance(mask_image, PIL.Image.Image):
             mask, masked_image = prepare_mask_and_masked_image(image, mask_image)
+        else:
+            mask = mask_image
+            masked_image = image * (mask < 0.5)
 
         # 5. set timesteps
         self.scheduler.set_timesteps(num_inference_steps, device=device)