gray input image

app_masked.py

@@ -48,12 +48,15 @@ def image_to_image(
     time_cost_str = ''
     run_task_time, time_cost_str = get_time_cost(run_task_time, time_cost_str)
 
+    gray_image = input_image.convert("L")
+    gray_image = Image.merge("RGB", (gray_image, gray_image, gray_image))
+
     generator = torch.Generator(device=DEVICE).manual_seed(seed)
     generated_image = basepipeline(
         generator=generator,
         prompt=edit_prompt,
         negative_prompt=DEFAULT_NEGATIVE_PROMPT,
-        original_image=input_image,
+        original_image=gray_image,
         mask=mask_image,
         guidance_scale=guidance_scale,
         num_inference_steps=num_steps,

pipelines/masked_stable_diffusion_xl_img2img.py

@@ -213,7 +213,7 @@ class MaskedStableDiffusionXLImg2ImgPipeline(StableDiffusionXLImg2ImgPipeline):
         if pil_mask.mode != "L":
             pil_mask = pil_mask.convert("L")
         mask_blur = self.blur_mask(pil_mask, blur)
-        mask_compose = self.blur_mask(mask_blur, blur_compose)
+        mask_compose = self.blur_mask(pil_mask, blur_compose)
         if original_image is None:
             original_image = image
         if prompt is not None and isinstance(prompt, str):