add combine embeds option

InstantID.py

@@ -6,10 +6,9 @@
 import math
 import cv2
 import PIL.Image
-from comfy.ldm.modules.attention import optimized_attention
 from .resampler import Resampler
 from .CrossAttentionPatch import CrossAttentionPatch
-from .utils import tensor_to_size, tensor_to_image, image_to_tensor
+from .utils import tensor_to_image
 
 from insightface.app import FaceAnalysis
 
@@ -96,12 +95,12 @@ def get_image_embeds(self, clip_embed, clip_embed_zeroed):
 class ImageProjModel(torch.nn.Module):
     def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024, clip_extra_context_tokens=4):
         super().__init__()
-        
+
         self.cross_attention_dim = cross_attention_dim
         self.clip_extra_context_tokens = clip_extra_context_tokens
         self.proj = torch.nn.Linear(clip_embeddings_dim, self.clip_extra_context_tokens * cross_attention_dim)
         self.norm = torch.nn.LayerNorm(cross_attention_dim)
-        
+
     def forward(self, image_embeds):
         embeds = image_embeds
         clip_extra_context_tokens = self.proj(embeds).reshape(-1, self.clip_extra_context_tokens, self.cross_attention_dim)
@@ -151,7 +150,7 @@ def load_model(self, instantid_file):
                 elif key.startswith("ip_adapter."):
                     st_model["ip_adapter"][key.replace("ip_adapter.", "")] = model[key]
             model = st_model
-        
+
         return (model,)
 
 def extractFeatures(insightface, image, extract_kps=False):
@@ -234,7 +233,7 @@ def add_noise(image, factor):
     mask = (torch.rand_like(image) < factor).float()
     noise = torch.rand_like(image)
     noise = torch.zeros_like(image) * (1-mask) + noise * mask
-    
+
     return factor*noise
 
 class ApplyInstantID:
@@ -264,7 +263,7 @@ def INPUT_TYPES(s):
     FUNCTION = "apply_instantid"
     CATEGORY = "InstantID"
 
-    def apply_instantid(self, instantid, insightface, control_net, image, model, positive, negative, start_at, end_at, weight=.8, ip_weight=None, cn_strength=None, noise=0.35, image_kps=None, mask=None):
+    def apply_instantid(self, instantid, insightface, control_net, image, model, positive, negative, start_at, end_at, weight=.8, ip_weight=None, cn_strength=None, noise=0.35, image_kps=None, mask=None, combine_embeds='average'):
         self.dtype = torch.float16 if comfy.model_management.should_use_fp16() else torch.float32
         self.device = comfy.model_management.get_torch_device()
 
@@ -290,7 +289,10 @@ def apply_instantid(self, instantid, insightface, control_net, image, model, pos
         clip_embed = face_embed
         # InstantID works better with averaged embeds (TODO: needs testing)
         if clip_embed.shape[0] > 1:
-            clip_embed = torch.mean(clip_embed, dim=0).unsqueeze(0)
+            if combine_embeds == 'average':
+                clip_embed = torch.mean(clip_embed, dim=0).unsqueeze(0)
+            elif combine_embeds == 'norm average':
+                clip_embed = torch.mean(clip_embed / torch.norm(clip_embed, dim=0, keepdim=True), dim=0).unsqueeze(0)
 
         if noise > 0:
             seed = int(torch.sum(clip_embed).item()) % 1000000007
@@ -413,6 +415,7 @@ def INPUT_TYPES(s):
                 "start_at": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.001, }),
                 "end_at": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.001, }),
                 "noise": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0, "step": 0.1, }),
+                "combine_embeds": (['average', 'norm average', 'concat'], {"default": 'average'}),
             },
             "optional": {
                 "image_kps": ("IMAGE",),
@@ -443,7 +446,7 @@ def INPUT_TYPES(s):
     FUNCTION = "patch_attention"
     CATEGORY = "InstantID"
 
-    def patch_attention(self, instantid, insightface, image, model, weight, start_at, end_at, noise=0.0, mask=None):       
+    def patch_attention(self, instantid, insightface, image, model, weight, start_at, end_at, noise=0.0, mask=None):
         self.dtype = torch.float16 if comfy.model_management.should_use_fp16() else torch.float32
         self.device = comfy.model_management.get_torch_device()
 
@@ -530,7 +533,7 @@ def patch_attention(self, instantid, insightface, image, model, weight, start_at
             for index in range(10):
                 _set_model_patch_replace(work_model, patch_kwargs, ("middle", 0, index))
                 patch_kwargs["number"] += 1
-        
+
         return(work_model, { "cond": image_prompt_embeds, "uncond": uncond_image_prompt_embeds }, )
 
 class ApplyInstantIDControlNet:
@@ -565,7 +568,7 @@ def apply_controlnet(self, face_embeds, control_net, image_kps, positive, negati
 
         if mask is not None:
             mask = mask.to(self.device)
-        
+
         if mask is not None and len(mask.shape) < 3:
             mask = mask.unsqueeze(0)