From 80e230bc9d948625ebf422644dd3dfb3d6106eed Mon Sep 17 00:00:00 2001
From: chenpipi0807 <144689950+chenpipi0807@users.noreply.github.com>
Date: Thu, 4 Sep 2025 18:45:24 +0800
Subject: [PATCH] =?UTF-8?q?=E5=AF=B9qwen-image=E4=B9=9F=E5=A2=9E=E5=8A=A0?=
 =?UTF-8?q?=E4=BA=86=E8=87=AA=E5=AE=9A=E4=B9=89=E6=8F=90=E7=A4=BA=E8=AF=8D?=
 =?UTF-8?q?=E7=9A=84=E5=8A=9F=E8=83=BD?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 nodes.py | 43 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 43 insertions(+)

diff --git a/nodes.py b/nodes.py
index e8436e3..b6691c4 100644
--- a/nodes.py
+++ b/nodes.py
@@ -82,11 +82,54 @@ Process: First identify all distinctive character features from the input image,
         return (conditioning, )
 
 
+class TextEncodeQwenImageT2ICustom:
+    @classmethod
+    def INPUT_TYPES(s):
+        # Default text-to-image system prompt template (official default)
+        default_t2i_template = (
+            "<|im_start|>system\n"
+            "Describe the image by detailing the color, shape, size, texture, quantity, text, "
+            "spatial relationships of the objects and background:<|im_end|>\n"
+            "<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
+        )
+
+        return {"required": {
+            "clip": ("CLIP", ),
+            "prompt": ("STRING", {"multiline": True, "dynamicPrompts": True}),
+            "system_template": ("STRING", {"multiline": True, "default": default_t2i_template}),
+        }}
+
+    RETURN_TYPES = ("CONDITIONING",)
+    FUNCTION = "encode"
+    CATEGORY = "advanced/conditioning"
+
+    def encode(self, clip, prompt, system_template):
+        # Create tokenizer dedicated to qwen-image and set custom text-only system template
+        tokenizer = QwenImageTokenizer()
+        tokenizer.llama_template = system_template
+
+        # Temporarily swap tokenizer
+        original_tokenizer = clip.tokenizer
+        clip.tokenizer = tokenizer
+
+        try:
+            # Text-to-image: no images passed
+            tokens = clip.tokenize(prompt, images=[])
+            conditioning = clip.encode_from_tokens_scheduled(tokens)
+        finally:
+            # Always restore original tokenizer
+            clip.tokenizer = original_tokenizer
+
+        return (conditioning, )
+
+
 # Node mappings for ComfyUI
 NODE_CLASS_MAPPINGS = {
     "TextEncodeQwenImageEditEnhanced": TextEncodeQwenImageEditEnhanced,
+    "TextEncodeQwenImageT2ICustom": TextEncodeQwenImageT2ICustom,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
     "TextEncodeQwenImageEditEnhanced": "Text Encode Qwen Image Edit (Enhanced)",
+    "TextEncodeQwenImageT2ICustom": "Text Encode Qwen Image (Qwen-only, Custom System)",
 }