Add files via upload

更新 ComfyUI-ModelScope-API 支持Qwen3-VL系列模型反推
2025-10-22 03:01:58 +08:00
parent b6468fb270
commit 2e3634fe6c
3 changed files with 375 additions and 5 deletions
@@ -0,0 +1,182 @@
+{
+  "id": "6d37ea84-d0ab-4123-8c8c-965b62a18230",
+  "revision": 0,
+  "last_node_id": 19,
+  "last_link_id": 31,
+  "nodes": [
+    {
+      "id": 17,
+      "type": "easy showAnything",
+      "pos": [
+        2732.3588101299074,
+        397.60778722933657
+      ],
+      "size": [
+        482.4299999999994,
+        553.9499999999997
+      ],
+      "flags": {},
+      "order": 2,
+      "mode": 0,
+      "inputs": [
+        {
+          "label": "输入任何",
+          "name": "anything",
+          "shape": 7,
+          "type": "*",
+          "link": 31
+        }
+      ],
+      "outputs": [
+        {
+          "name": "output",
+          "type": "*",
+          "links": null
+        }
+      ],
+      "properties": {
+        "cnr_id": "comfyui-easy-use",
+        "ver": "1.3.3",
+        "Node name for S&R": "easy showAnything"
+      },
+      "widgets_values": [
+        "一位年轻美丽的亚洲女性，身着精致的淡紫色丝绸旗袍，旗袍上绣有金色祥云与花卉纹样，领口为高立领配盘扣设计，前襟开衩至大腿中部，展现优雅曲线。她佩戴半透明薄纱长袖，袖口拼接红色织锦并饰以金色龙纹刺绣，搭配同色系红色过膝袜。发型为古典盘发，插有金质流苏发簪与珠饰，刘海轻柔垂落，妆容精致，红唇微启，眼神温柔含情，直视镜头。背景为中式室内场景，墙面挂有水墨风格的山水或花鸟画作，色调柔和偏暖，营造静谧典雅氛围。整体构图采用中景人像视角，人物居中站立，略微侧身，突出服装细节与身材比例。光影柔和均匀，强调丝绸面料的光泽感与刺绣纹理，皮肤质感细腻光滑，呈现数字绘画的超写实美感。风格参考新中式美学与现代商业人像摄影，融合传统元素与时尚表现力，色彩主调为淡紫、金、红三色，辅以米白与浅棕背景，形成高级和谐的视觉层次。画面干净无杂物，焦点集中于人物神态与服饰工艺，适合用于AI绘图生成高质量东方美人肖像。"
+      ]
+    },
+    {
+      "id": 5,
+      "type": "LoadImage",
+      "pos": [
+        1869.8370214843746,
+        350.8748266601562
+      ],
+      "size": [
+        290.8019714355469,
+        570.7747192382812
+      ],
+      "flags": {},
+      "order": 0,
+      "mode": 0,
+      "inputs": [],
+      "outputs": [
+        {
+          "label": "图像",
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": [
+            30
+          ]
+        },
+        {
+          "label": "遮罩",
+          "name": "MASK",
+          "type": "MASK"
+        }
+      ],
+      "properties": {
+        "cnr_id": "comfy-core",
+        "ver": "0.3.49",
+        "Node name for S&R": "LoadImage",
+        "ue_properties": {
+          "version": "7.0.1",
+          "widget_ue_connectable": {
+            "image": true,
+            "upload": true
+          }
+        }
+      },
+      "widgets_values": [
+        "ebe510f1a069ba899eec0f82fefd3746d26f2f95ec2378a10ec1cd56cbc53642.png",
+        "image"
+      ],
+      "color": "#232",
+      "bgcolor": "#353"
+    },
+    {
+      "id": 19,
+      "type": "ModelScopeImageCaptionNode",
+      "pos": [
+        2266.803982765229,
+        414.0277931000381
+      ],
+      "size": [
+        400,
+        214
+      ],
+      "flags": {},
+      "order": 1,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "image",
+          "type": "IMAGE",
+          "link": 30
+        }
+      ],
+      "outputs": [
+        {
+          "name": "description",
+          "type": "STRING",
+          "links": [
+            31
+          ]
+        }
+      ],
+      "properties": {
+        "Node name for S&R": "ModelScopeImageCaptionNode"
+      },
+      "widgets_values": [
+        "***已保存1个Token***",
+        "帮我拆解这张图片的提示词，要求从主体内容、场景设定、风格参考、色条色彩、构图视角、细节补充等这些角度来用文字描述图片，并汇总生成一个能让我用 ai绘画工具的文生图提示词，字数在 800 字以内，只要求输出汇总后的最终提示词，不需要无用信息。如：提示词标题，总字数等信息。",
+        "Qwen/Qwen3-VL-235B-A22B-Instruct",
+        1000,
+        0.7
+      ]
+    }
+  ],
+  "links": [
+    [
+      30,
+      5,
+      0,
+      19,
+      0,
+      "IMAGE"
+    ],
+    [
+      31,
+      19,
+      0,
+      17,
+      0,
+      "*"
+    ]
+  ],
+  "groups": [
+    {
+      "id": 2,
+      "title": "魔搭-图片编辑",
+      "bounding": [
+        1663.396484375,
+        122.82080078125,
+        1826.526123046875,
+        960.4658203125
+      ],
+      "color": "#3f789e",
+      "font_size": 24,
+      "flags": {}
+    }
+  ],
+  "config": {},
+  "extra": {
+    "ds": {
+      "scale": 0.620921323059155,
+      "offset": [
+        -1125.4324556670972,
+        206.079530787193
+      ]
+    },
+    "frontendVersion": "1.30.1"
+  },
+  "version": 0.4
+}
@@ -1,11 +1,22 @@
+# 原有导入
 from .modelscope_image_node import NODE_CLASS_MAPPINGS as IMAGE_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS as IMAGE_DISPLAY_MAPPINGS
 from .modelscope_vision_node import NODE_CLASS_MAPPINGS as VISION_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS as VISION_DISPLAY_MAPPINGS
 from .modelscope_text_node import NODE_CLASS_MAPPINGS as TEXT_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS as TEXT_DISPLAY_MAPPINGS
- 
-NODE_CLASS_MAPPINGS = {**IMAGE_MAPPINGS, **VISION_MAPPINGS, **TEXT_MAPPINGS}
-NODE_DISPLAY_NAME_MAPPINGS = {**IMAGE_DISPLAY_MAPPINGS, **VISION_DISPLAY_MAPPINGS, **TEXT_DISPLAY_MAPPINGS}
- 
+
+# 新增：导入图像描述节点的映射
+from .modelscope_image_caption_node import NODE_CLASS_MAPPINGS as IMAGE_CAPTION_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS as IMAGE_CAPTION_DISPLAY_MAPPINGS
+
+# 合并所有节点映射（包含新增的图像描述节点）
+NODE_CLASS_MAPPINGS = {
+    **IMAGE_MAPPINGS,** VISION_MAPPINGS,
+    **TEXT_MAPPINGS,** IMAGE_CAPTION_MAPPINGS  # 新增
+}
+NODE_DISPLAY_NAME_MAPPINGS = {
+    **IMAGE_DISPLAY_MAPPINGS,** VISION_DISPLAY_MAPPINGS,
+    **TEXT_DISPLAY_MAPPINGS,** IMAGE_CAPTION_DISPLAY_MAPPINGS  # 新增
+}
+
 __all__ = ['NODE_CLASS_MAPPINGS', 'NODE_DISPLAY_NAME_MAPPINGS']
- 
+
 WEB_DIRECTORY = "./js"
 __version__ = "1.0.0"
@@ -0,0 +1,177 @@
+import requests
+import json
+import time
+import torch
+import numpy as np
+from PIL import Image
+from io import BytesIO
+import os
+import base64
+import re
+from .modelscope_image_node import load_config, load_api_tokens, save_api_tokens, tensor_to_base64_url
+
+# 检查openai库是否可用
+try:
+    from openai import OpenAI
+    OPENAI_AVAILABLE = True
+except ImportError:
+    OPENAI_AVAILABLE = False
+
+class ModelScopeImageCaptionNode:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        if not OPENAI_AVAILABLE:
+            return {
+                "required": {
+                    "error_message": ("STRING", {
+                        "default": "请先安装openai库: pip install openai",
+                        "multiline": True
+                    }),
+                }
+            }
+        saved_tokens = load_api_tokens()
+        # 定义支持的模型列表
+        supported_models = [
+            "Qwen/Qwen3-VL-8B-Instruct",
+            "Qwen/Qwen3-VL-235B-A22B-Instruct"
+        ]
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "api_tokens": ("STRING", {
+                    "default": f"***已保存{len(saved_tokens)}个Token***" if saved_tokens else "",
+                    "placeholder": "请输入API Token（支持多个，用逗号/换行分隔）",
+                    "multiline": True
+                }),
+            },
+            "optional": {
+                "prompt": ("STRING", {
+                    "multiline": True,
+                    "default": "详细描述这张图片的内容，包括主体、背景、颜色、风格等信息"
+                }),
+                # 添加模型下拉选择
+                "model": (supported_models, {
+                    "default": "Qwen/Qwen3-VL-8B-Instruct"  # 默认选中原模型
+                }),
+                "max_tokens": ("INT", {
+                    "default": 1000,
+                    "min": 100,
+                    "max": 4000
+                }),
+                "temperature": ("FLOAT", {
+                    "default": 0.7,
+                    "min": 0.1,
+                    "max": 2.0,
+                    "step": 0.1
+                }),
+            }
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("description",)
+    FUNCTION = "generate_caption"
+    CATEGORY = "ModelScopeAPI"
+
+    def parse_api_tokens(self, token_input):
+        """解析输入的多个API Token（支持逗号、分号、换行分隔）"""
+        if not token_input or token_input.strip() in ["", f"***已保存{len(load_api_tokens())}个Token***"]:
+            return load_api_tokens()
+        
+        # 支持多种分隔符拆分Token
+        tokens = re.split(r'[,;\n]+', token_input)
+        return [token.strip() for token in tokens if token.strip()]
+
+    # 调整参数顺序，加入model参数（与INPUT_TYPES顺序匹配）
+    def generate_caption(self, image=None, api_tokens="", prompt="详细描述这张图片的内容", model="Qwen/Qwen3-VL-8B-Instruct", max_tokens=1000, temperature=0.7):
+        if not OPENAI_AVAILABLE:
+            return ("请先安装openai库: pip install openai",)
+        
+        # 解析Token列表（支持多个）
+        tokens = self.parse_api_tokens(api_tokens)
+        if not tokens:
+            raise Exception("请提供至少一个有效的API Token")
+        
+        # 保存新Token（如果有变化）
+        saved_tokens = load_api_tokens()
+        if api_tokens.strip() not in ["", f"***已保存{len(saved_tokens)}个Token***"]:
+            if save_api_tokens(tokens):
+                print(f"✅ 已保存 {len(tokens)} 个API Token")
+            else:
+                print("⚠️ API Token保存失败，但不影响当前使用")
+        
+        try:
+            print(f"🔍 开始生成图像描述...")
+            print(f"📝 提示词: {prompt}")
+            print(f"🤖 模型: {model}")  # 显示选中的模型
+            print(f"🔑 可用Token数量: {len(tokens)}")
+            
+            # 转换图像为base64格式
+            image_url = tensor_to_base64_url(image)
+            print(f"🖼️ 图像已转换为base64格式")
+            
+            # 构建消息体
+            messages = [{
+                'role': 'user',
+                'content': [{
+                    'type': 'text',
+                    'text': prompt,
+                }, {
+                    'type': 'image_url',
+                    'image_url': {
+                        'url': image_url,
+                    },
+                }],
+            }]
+            
+            # 轮询尝试每个Token
+            last_exception = None
+            for i, token in enumerate(tokens):
+                try:
+                    print(f"🔄 尝试使用第 {i+1}/{len(tokens)} 个Token...")
+                    
+                    # 初始化OpenAI客户端
+                    client = OpenAI(
+                        base_url='https://api-inference.modelscope.cn/v1',
+                        api_key=token
+                    )
+                    
+                    # 调用API（使用选中的模型）
+                    response = client.chat.completions.create(
+                        model=model,
+                        messages=messages,
+                        max_tokens=max_tokens,
+                        temperature=temperature,
+                        stream=False
+                    )
+                    
+                    # 成功获取结果
+                    description = response.choices[0].message.content
+                    print(f"✅ 第 {i+1} 个Token调用成功!")
+                    print(f"📄 结果预览: {description[:100]}...")
+                    return (description,)
+                    
+                except Exception as e:
+                    last_exception = e
+                    print(f"❌ 第 {i+1} 个Token调用失败: {str(e)}")
+                    if i < len(tokens) - 1:
+                        print(f"⏳ 准备尝试下一个Token...")
+            
+            # 所有Token都失败
+            raise Exception(f"所有Token均调用失败: {str(last_exception)}")
+            
+        except Exception as e:
+            error_msg = f"图像描述生成失败: {str(e)}"
+            print(f"❌ {error_msg}")
+            return (error_msg,)
+
+# 节点映射
+NODE_CLASS_MAPPINGS = {
+    "ModelScopeImageCaptionNode": ModelScopeImageCaptionNode
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "ModelScopeImageCaptionNode": "ModelScope 图像描述生成"
+}