Fix alpha channel handling in Qwen image edit nodes

- Fix TextEncodeQwenImageEditPlus to ensure only RGB channels are used
- Prevents RuntimeError when input images have alpha channels
- Ensures proper tensor shape for vision language models

comfy_extras/nodes_qwen.py

@@ -1,4 +1,4 @@
-import node_helpers
+import node_helpers
 import comfy.utils
 import math
 
@@ -13,15 +13,13 @@ class TextEncodeQwenImageEdit:
             "optional": {"vae": ("VAE", ),
                          "image": ("IMAGE", ),}}
 
-    RETURN_TYPES = ("CONDITIONING", "IMAGE", "LATENT")
+    RETURN_TYPES = ("CONDITIONING",)
     FUNCTION = "encode"
 
     CATEGORY = "advanced/conditioning"
 
     def encode(self, clip, prompt, vae=None, image=None):
         ref_latent = None
-        output_image = None
-        
         if image is None:
             images = []
         else:
@@ -29,23 +27,12 @@ class TextEncodeQwenImageEdit:
             total = int(1024 * 1024)
 
             scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2]))
+            width = round(samples.shape[3] * scale_by)
+            height = round(samples.shape[2] * scale_by)
 
-            width = math.floor(samples.shape[3] * scale_by / 8) * 8
-            height = math.floor(samples.shape[2] * scale_by / 8) * 8
-
-            original_width = samples.shape[3]
-            original_height = samples.shape[2]
-            
-            if width < original_width or height < original_height:
-                upscale_method = "area"
-            else:
-                upscale_method = "lanczos"
-            
-            s = comfy.utils.common_upscale(samples, width, height, upscale_method, "disabled")
+            s = comfy.utils.common_upscale(samples, width, height, "area", "disabled")
             image = s.movedim(1, -1)
             images = [image[:, :, :, :3]]
-            output_image = image[:, :, :, :3]
-            
             if vae is not None:
                 ref_latent = vae.encode(image[:, :, :, :3])
 
@@ -53,12 +40,64 @@ class TextEncodeQwenImageEdit:
         conditioning = clip.encode_from_tokens_scheduled(tokens)
         if ref_latent is not None:
             conditioning = node_helpers.conditioning_set_values(conditioning, {"reference_latents": [ref_latent]}, append=True)
-        
-        latent_output = {"samples": ref_latent} if ref_latent is not None else None
-        
-        return (conditioning, output_image, latent_output)
+        return (conditioning, )
+
+
+class TextEncodeQwenImageEditPlus:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "clip": ("CLIP", ),
+            "prompt": ("STRING", {"multiline": True, "dynamicPrompts": True}),
+            },
+            "optional": {"vae": ("VAE", ),
+                         "image1": ("IMAGE", ),
+                         "image2": ("IMAGE", ),
+                         "image3": ("IMAGE", ),
+                         }}
+
+    RETURN_TYPES = ("CONDITIONING",)
+    FUNCTION = "encode"
+
+    CATEGORY = "advanced/conditioning"
+
+    def encode(self, clip, prompt, vae=None, image1=None, image2=None, image3=None):
+        ref_latents = []
+        images = [image1, image2, image3]
+        images_vl = []
+        llama_template = "<|im_start|>system\nDescribe the key features of the input image (color, shape, size, texture, objects, background), then explain how the user's text instruction should alter or modify the image. Generate a new image that meets the user's requirements while maintaining consistency with the original input where appropriate.<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
+        image_prompt = ""
+
+        for i, image in enumerate(images):
+            if image is not None:
+                samples = image.movedim(-1, 1)
+                total = int(384 * 384)
+
+                scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2]))
+                width = round(samples.shape[3] * scale_by)
+                height = round(samples.shape[2] * scale_by)
+
+                s = comfy.utils.common_upscale(samples, width, height, "area", "disabled")
+                images_vl.append(s.movedim(1, -1)[:, :, :, :3])
+                if vae is not None:
+                    total = int(1024 * 1024)
+                    scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2]))
+                    width = round(samples.shape[3] * scale_by / 8.0) * 8
+                    height = round(samples.shape[2] * scale_by / 8.0) * 8
+
+                    s = comfy.utils.common_upscale(samples, width, height, "area", "disabled")
+                    ref_latents.append(vae.encode(s.movedim(1, -1)[:, :, :, :3]))
+
+                image_prompt += "Picture {}: <|vision_start|><|image_pad|><|vision_end|>".format(i + 1)
+
+        tokens = clip.tokenize(image_prompt + prompt, images=images_vl, llama_template=llama_template)
+        conditioning = clip.encode_from_tokens_scheduled(tokens)
+        if len(ref_latents) > 0:
+            conditioning = node_helpers.conditioning_set_values(conditioning, {"reference_latents": ref_latents}, append=True)
+        return (conditioning, )
 
 
 NODE_CLASS_MAPPINGS = {
     "TextEncodeQwenImageEdit": TextEncodeQwenImageEdit,
+    "TextEncodeQwenImageEditPlus": TextEncodeQwenImageEditPlus,
 }