allow batches of kps images, fix cubiq#80

InstantID.py

@@ -376,7 +376,7 @@ def INPUT_TYPES(s):
     CATEGORY = "InstantID"
 
     def preprocess_image(self, faceanalysis, image):
-        face_kps = extractFeatures(faceanalysis, image[0].unsqueeze(0), extract_kps=True)
+        face_kps = extractFeatures(faceanalysis, image, extract_kps=True)
 
         if face_kps is None:
             face_kps = torch.zeros_like(image)
@@ -437,7 +437,8 @@ def apply_instantid(self, instantid, insightface, control_net, image, model, pos
         if face_embed is None:
             raise Exception('Reference Image: No face detected.')
 
-        face_kps = extractFeatures(insightface, image_kps[0].unsqueeze(0) if image_kps is not None else image[0].unsqueeze(0), extract_kps=True)
+        # if no keypoints image is provided, use the image itself (only the first one in the batch)
+        face_kps = extractFeatures(insightface, image_kps if image_kps is not None else image[0].unsqueeze(0), extract_kps=True)
 
         if face_kps is None:
             face_kps = torch.zeros_like(image) if image_kps is None else image_kps

README.md

@@ -58,7 +58,7 @@ The person is posed based on the keypoints generated from the reference image. Y
 
 ## Noise Injection
 
-The default InstantID implementation seems to really burn the image, I find that by injecting noise to the negative embeds we can mitigate the effect and also increase the likeliness to the reference. The default Apply InstantID node automatically injects 35% noise, if you want to fine tune the effect you use the Advanced InstantID node.
+The default InstantID implementation seems to really burn the image, I find that by injecting noise to the negative embeds we can mitigate the effect and also increase the likeliness to the reference. The default Apply InstantID node automatically injects 35% noise, if you want to fine tune the effect you can use the Advanced InstantID node.
 
 This is still experimental and may change in the future.