updated the MaskFromFaceModel node.

py/node_gemini_enhance_prompt.py

@@ -137,7 +137,9 @@ class GeminiPromptEnhance:
                         max_output_tokens=8192, temperature=0.4, gender_prior="",gender_alternative="", enabled=True, request_exception_handle="bypass"):
         if not enabled:
             return (text_input,)
-        
+        if prompt is None or prompt.strip() == "":
+            prompt = self.default_prompt
+            
         gender = gender_prior if gender_prior else gender_alternative
         # 生成缓存键
         cache_key = f"{text_input or ''}_{gender}"

py/nodes.py

@@ -22,7 +22,7 @@ import cv2
 from comfy_extras.nodes_upscale_model import ImageUpscaleWithModel
 from math import dist
 import folder_paths
-from .utils import tensor2np,np2tensor
+from .utils import tensor2cv
 import hashlib
 
 config_dir = os.path.join(folder_paths.base_path, "config")
@@ -581,7 +581,8 @@ class MaskFromFaceModel:
             }
         }
 
-    RETURN_TYPES = ("MASK",)
+    RETURN_TYPES = ("MASK","STRING")
+    RETURN_NAMES = ("mask","genders")
     FUNCTION = 'mask_get'
     CATEGORY = 'utils/mask'
 
@@ -601,7 +602,7 @@ class MaskFromFaceModel:
             raise Exception("both faceanalysis and face_model are none!")
         
         if face_model is None:
-            image_np = tensor2np(image)
+            image_np = tensor2cv(image)
             image_np = image_np[0] if isinstance(image_np, list) else image_np
             face_model = self.analyze_faces(faceanalysis, image_np)
 
@@ -619,8 +620,11 @@ class MaskFromFaceModel:
         if max_face_number !=-1 and len(face_model) > max_face_number:
             face_models = self.remove_unavaible_face_models(face_models=face_models,max_people_number=max_face_number) 
 
+        face_models = sorted(face_models, key=lambda x:x.bbox[0])
         result = np.zeros((h, w), dtype=np.uint8)
+        genders = []
         for face in face_models:
+            genders.append(face.sex)
             points = face.landmark_2d_106.astype(np.int32)  # Convert landmarks to integer format            
             if add_bbox_upper_points:
                   # 获取bbox的坐标
@@ -649,7 +653,7 @@ class MaskFromFaceModel:
 
         result = torch.unsqueeze(torch.tensor(np.clip(result/255, 0, 1)), 0)
 
-        return (result,)
+        return (result, ",".join(genders))
     
     def remove_unavaible_face_models(self, face_models, max_people_number):
         max_lengths = []

py/utils.py

@@ -75,3 +75,8 @@ def draw_rect(image:Image, x:int, y:int, width:int, height:int, line_color:str,
     draw = ImageDraw.Draw(image)
     draw.rectangle((x, y, x + width, y + height), fill=box_color, outline=line_color, width=line_width, )
     return image
+
+# Tensor to cv2
+def tensor2cv(image):
+    image_np = np.clip(255. * image.cpu().numpy().squeeze(), 0, 255).astype(np.uint8)      
+    return cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR)