commit e13e1aeedea084642782407cd1e14bcbbb4cc8cb Author: pandy <1272175942@qq.com> Date: Sun Mar 8 11:10:30 2026 +0800 Initial commit from comfyui-PD_comfy api node diff --git a/.cursorrules b/.cursorrules new file mode 100644 index 0000000..c06a209 --- /dev/null +++ b/.cursorrules @@ -0,0 +1,4 @@ +1. 你是一个经验丰富的python开发者，擅长制作commfyui的节点，现在需要创建一个新的节点 +2. 确保输入输出图像或者遮罩的张量形状分别为B H W C ,B H W +3. 需要包含导入，节点注册等构建节点所需的完整代码 + diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..d50b56d Binary files /dev/null and b/.gitignore differ diff --git a/README.md b/README.md new file mode 100644 index 0000000..bb1c778 --- /dev/null +++ b/README.md @@ -0,0 +1,182 @@ +# ComfyUI PD Comfy API Node + +ComfyUI 自定义节点套件，集成 Gemini 和 GPT 图像生成 API，支持两种认证方式。 + +## ✨ 功能特性 + +- 🎨 **多模型支持**：Gemini 2.5 Flash Image、GPT Image 1/1.5 +- 🔑 **双认证方式**：支持 API Key 和 Auth Token 两种调用方式 +- 💰 **成本追踪**：实时显示 API 调用成本 +- 🖼️ **图像处理**：支持文生图、图生图、图像编辑 +- ⚙️ **灵活配置**：多种宽高比、质量、分辨率选项 + +## 📦 安装方法 + +```bash +cd ComfyUI/custom_nodes +git clone https://github.com/7BEII/comfyui-PD_comfy-api-node.git +pip install -r comfyui-PD_comfy-api-node/requirements.txt +``` + +重启 ComfyUI 后，在节点菜单的 `PD_Tools/Image_Generation` 分类中找到节点。 + +## 🔑 两种认证方式 + +### 方式一：API Key（推荐） + +使用 ComfyUI 官方 API Key，简单直接。 + +**获取方法**： +1. 访问 [ComfyUI API](https://www.comfy.org/api-keys) +2. 登录并创建 API Key +3. 复制 Key 并粘贴到节点中 + +![API Key 节点示例](image/comfyui-node-apikey.png) + +**优点**： +- ✅ 官方支持，稳定可靠 +- ✅ 配置简单，一个 Key 搞定 +- ✅ 支持所有模型 + +### 方式二：Auth Token + +使用浏览器 Auth Token，适合高级用户。 + +**获取方法**： +1. 安装油猴脚本（Tampermonkey）：具体B站有教程：小熊猫pandy +2. 安装 Token 抓取脚本 +3. 访问 ComfyUI 网站并登录 +4. 脚本会自动抓取 Token +5. 复制 Token 并粘贴到节点中 + +![Auth Token 节点示例](image/comfyui-node-auth.png) + +**优点**： +- ✅ 无需申请 API Key +- ✅ 使用现有账号权限 +- ✅ 适合测试和开发 + +## 📝 更新日志 + +### v2.0.0 -2026.0112 +- 新增 Auth Token 认证方式 +- 新增 GPT Image 节点 +- 优化成本计算 +- 改进错误处理 + +### v1.0.0 +- 初始版本 +- 支持 Gemini 图像生成 +- 支持 API Key 认证 + +## 📋 节点说明 + +### 1. PD: Gemini Image (apikey) + +使用 API Key 调用 Gemini 图像生成。 + +**主要参数**： +- `api_key`：ComfyUI API Key +- `prompt`：图像生成提示词 +- `model`：模型选择（gemini-2.5-flash-image / gemini-2.5-flash-image-preview） +- `aspect_ratio`：宽高比（auto / 1:1 / 16:9 / 9:16 等） +- `image_ref`：参考图像（可选） + +**输出**： +- `image`：生成的图像 +- `info`：成本和状态信息 + +### 2. PD: Gemini Image (comfyui_AuthToken) + +使用 Auth Token 调用 Gemini 图像生成。 + +**主要参数**： +- `auth_token`：ComfyUI Auth Token +- `prompt`：图像生成提示词 +- `model`：模型选择 +- `aspect_ratio`：宽高比 +- `resolution`：分辨率（auto / 1K / 2K / 4K） +- `image_ref`：参考图像（可选） + +**输出**： +- `image`：生成的图像 +- `info`：成本和状态信息 + +### 3. PD: GPT Image (apikey) + +使用 API Key 调用 OpenAI GPT 图像生成。 + +**主要参数**： +- `api_key`：ComfyUI API Key +- `prompt`：图像生成提示词 +- `model`：模型选择（gpt-image-1 / gpt-image-1.5） +- `quality`：质量（low / medium / high） +- `size`：尺寸（auto / 1024x1024 / 1024x1536 / 1536x1024） +- `background`：背景（auto / opaque / transparent） +- `image`：输入图像（可选，用于编辑） +- `mask`：遮罩（可选，用于局部编辑） + +**输出**： +- `image`：生成的图像 +- `info`：成本和状态信息 + +**特性**： +- 支持文生图 +- 支持图生图编辑 +- 支持遮罩局部编辑 + +## 💡 使用示例 + +### 示例 1：使用 Gemini 生成图像（API Key） + +``` +1. 添加 "PD: Gemini Image (apikey)" 节点 +2. 输入 API Key +3. 输入提示词："A futuristic city with flying cars" +4. 选择宽高比：16:9 +5. 连接输出到预览节点 +6. 执行工作流 +``` + +### 示例 2：使用 GPT 编辑图像 + +``` +1. 添加 "PD: GPT Image (apikey)" 节点 +2. 输入 API Key +3. 连接输入图像到 image 端口 +4. 输入编辑提示词："Add a sunset sky" +5. 设置质量为 high +6. 执行工作流 +``` + +## ⚠️ 注意事项 + +1. **API Key 安全**：请妥善保管 API Key，不要分享给他人 +2. **成本控制**：注意查看成本信息，控制 API 调用次数 +3. **网络连接**：需要稳定的网络连接 +4. **生成时间**：高质量图像可能需要较长时间 + +## 🐛 故障排除 + +### 节点未显示 +- 检查是否正确安装到 `custom_nodes` 目录 +- 重启 ComfyUI +- 查看控制台错误信息 + +### API 调用失败 +- 验证 API Key 或 Auth Token 是否正确 +- 检查网络连接 +- 查看控制台详细错误信息 + +### 图像格式错误 +- 确保输入图像格式为 ComfyUI 标准格式（B H W C） +- 检查图像通道数是否为 3 (RGB) + +## 📄 许可证 + +MIT License + +## 🤝 贡献 +欢迎提交 Issue 和 Pull Request！ +--- + diff --git a/__init__.py b/__init__.py new file mode 100644 index 0000000..9b9e0f8 --- /dev/null +++ b/__init__.py @@ -0,0 +1,74 @@ +""" +Comfyui_PDuse - PD ComfyUI 自定义节点套件 +包含多种实用的图像处理和工具节点 +""" + +import importlib +import os +import sys + +# 初始化节点映射 +NODE_CLASS_MAPPINGS = {} +NODE_DISPLAY_NAME_MAPPINGS = {} + +def get_ext_dir(subpath=None, mkdir=False): + """获取扩展目录路径""" + dir = os.path.dirname(__file__) + if subpath is not None: + dir = os.path.join(dir, subpath) + return os.path.abspath(dir) + +def safe_import_module(module_path, file_name): + """安全导入模块，避免导入错误导致崩溃""" + try: + imported_module = importlib.import_module(module_path) + + # 检查模块是否包含所需的映射 + if hasattr(imported_module, 'NODE_CLASS_MAPPINGS'): + NODE_CLASS_MAPPINGS.update(imported_module.NODE_CLASS_MAPPINGS) + print(f"✅ 成功加载模块: {file_name}") + + if hasattr(imported_module, 'NODE_DISPLAY_NAME_MAPPINGS'): + NODE_DISPLAY_NAME_MAPPINGS.update(imported_module.NODE_DISPLAY_NAME_MAPPINGS) + + return True + + except Exception as e: + print(f"❌ 加载模块 {file_name} 失败: {e}") + return False + +# 动态扫描并加载 py/ 目录下的所有模块 +py_dir = get_ext_dir("py") + +if os.path.exists(py_dir): + files = os.listdir(py_dir) + + for file in files: + if not file.endswith(".py") or file.startswith("_"): + continue + + module_name = os.path.splitext(file)[0] + module_path = f"{__name__}.py.{module_name}" + + safe_import_module(module_path, file) +else: + print(f"警告: py目录不存在: {py_dir}") + +# Web界面目录 +WEB_DIRECTORY = "./js" + +# 显示加载信息 +if NODE_CLASS_MAPPINGS: + print("=" * 50) + print("🎨 Comfyui_PDuse 节点套件加载完成") + print("=" * 50) + print(f"📦 总计加载节点数量: {len(NODE_CLASS_MAPPINGS)}") + print("📋 已加载的节点:") + for name, display_name in NODE_DISPLAY_NAME_MAPPINGS.items(): + print(f" • {display_name}") + print("=" * 50) +else: + print("⚠️ 警告: 没有找到任何可用的节点") + +# 导出变量 +__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS", "WEB_DIRECTORY"] diff --git a/gittools.py b/gittools.py new file mode 100644 index 0000000..20239c3 --- /dev/null +++ b/gittools.py @@ -0,0 +1,590 @@ +#!/usr/bin/env python3 +# -*- coding: utf-8 -*- +""" +Git 自动同步工具 +用法: python gitrun.py [命令] + +命令说明: + push - 强制推送本地到远程（直接覆盖，保留历史记录） + pull - 强制拉取远程到本地（直接覆盖，保留历史记录） + sync - 智能同步（先拉取合并，再推送） + log - 查看提交历史并支持快速回退版本 + history - 查看完整提交历史 + config - 配置远程仓库地址 + + + + +重要说明: + • push/pull 都会保留完整的 Git 历史记录，可以随时回退 + • push 是纯粹的本地覆盖远程，不会先拉取远程内容 + • pull 是纯粹的远程覆盖本地，不会保留本地未推送的修改 + • sync 才会进行智能合并 + • log 命令可以查看历史并直接回退版本（输入数字选择，0或回车跳过） + • 所有历史提交都保存在 .git 目录中，永不丢失 +""" + +import os +import sys +import subprocess +import json +from pathlib import Path +from datetime import datetime + +CONFIG_FILE = Path.home() / '.gitrun_config.json' + + +class GitSyncTool: + def __init__(self): + self.config = self.load_config() + self.repo_path = os.getcwd() + + def load_config(self): + """加载配置文件""" + if CONFIG_FILE.exists(): + with open(CONFIG_FILE, 'r', encoding='utf-8') as f: + return json.load(f) + return {} + + def save_config(self): + """保存配置文件""" + with open(CONFIG_FILE, 'w', encoding='utf-8') as f: + json.dump(self.config, f, indent=2, ensure_ascii=False) + + def run_command(self, cmd, check=True, silent=False): + """执行命令并返回结果""" + try: + # Windows 上使用 GBK 编码，其他系统使用 UTF-8 + encoding = 'gbk' if sys.platform == 'win32' else 'utf-8' + + result = subprocess.run( + cmd, + shell=True, + cwd=self.repo_path, + capture_output=True, + text=True, + encoding=encoding, + errors='ignore' # 忽略无法解码的字符 + ) + if check and result.returncode != 0: + if not silent: + print(f"❌ 命令执行失败: {cmd}") + if result.stderr: + print(f"错误信息: {result.stderr}") + return None + return result + except Exception as e: + if not silent: + print(f"❌ 执行命令时出错: {e}") + return None + + def is_git_repo(self): + """检查当前目录是否是 Git 仓库""" + result = self.run_command("git rev-parse --is-inside-work-tree", check=False, silent=True) + return result and result.returncode == 0 + + def get_current_branch(self): + """获取当前分支名""" + result = self.run_command("git branch --show-current", silent=True) + return result.stdout.strip() if result else None + + def has_changes(self): + """检查是否有未提交的更改""" + result = self.run_command("git status --porcelain", silent=True) + return bool(result and result.stdout.strip()) + + def get_remote_url(self): + """获取远程仓库地址""" + result = self.run_command("git remote get-url origin", check=False, silent=True) + return result.stdout.strip() if result and result.returncode == 0 else None + + def config_remote(self): + """配置远程仓库地址""" + print("🔧 配置远程仓库地址") + + # 检查是否是 Git 仓库，如果不是则初始化 + if not self.is_git_repo(): + print("⚠️ 当前目录不是 Git 仓库") + init = input("是否初始化为 Git 仓库? (y/n): ").strip().lower() + if init == 'y': + result = self.run_command("git init") + if result: + print("✅ Git 仓库初始化成功") + else: + print("❌ Git 仓库初始化失败") + return + else: + print("❌ 操作已取消") + return + + current_url = self.get_remote_url() + if current_url: + print(f"当前远程仓库: {current_url}") + change = input("是否修改? (y/n): ").strip().lower() + if change != 'y': + return + + remote_url = input("请输入远程仓库地址 (如: https://github.com/username/repo.git): ").strip() + + if not remote_url: + print("❌ 远程仓库地址不能为空") + return + + # 检查是否已有 origin + if current_url: + result = self.run_command(f"git remote set-url origin {remote_url}") + else: + result = self.run_command(f"git remote add origin {remote_url}") + + if result: + print(f"✅ 远程仓库配置成功: {remote_url}") + + # 保存到配置文件 + repo_name = os.path.basename(self.repo_path) + if repo_name not in self.config: + self.config[repo_name] = {} + self.config[repo_name]['remote_url'] = remote_url + self.save_config() + + # 询问是否设置默认分支 + print("\n💡 提示: 首次推送前需要设置默认分支") + set_branch = input("是否设置默认分支为 main? (y/n): ").strip().lower() + if set_branch == 'y': + # 检查当前分支 + current_branch = self.get_current_branch() + if not current_branch: + # 创建初始提交 + print("\n📝 创建初始提交...") + self.run_command("git add .") + self.run_command('git commit -m "Initial commit" --allow-empty') + self.run_command("git branch -M main") + print("✅ 默认分支已设置为 main") + elif current_branch != 'main': + self.run_command("git branch -M main") + print("✅ 分支已重命名为 main") + else: + print("✅ 当前分支已是 main") + + def commit_changes(self, message=None): + """提交本地更改""" + if not self.has_changes(): + print("ℹ️ 没有需要提交的更改") + return True + + print("\n📝 检测到以下更改:") + self.run_command("git status --short") + + if message is None: + message = input("\n请输入提交信息: ").strip() + + if not message: + print("❌ 提交信息不能为空") + return False + + # 添加所有更改 + print("\n📦 正在暂存所有更改...") + result = self.run_command("git add -A") + if not result: + return False + + # 提交更改 + print(f"💾 正在提交: {message}") + result = self.run_command(f'git commit -m "{message}"') + if result: + print(f"✅ 提交成功") + return True + return False + + def force_push(self): + """纯粹的本地覆盖远程（不拉取，直接强制推送）""" + print("🚀 强制推送本地到远程") + + if not self.is_git_repo(): + print("❌ 当前目录不是 Git 仓库") + return + + branch = self.get_current_branch() + if not branch: + print("❌ 无法获取当前分支") + return + + print(f"📍 当前分支: {branch}") + + # 检查远程仓库 + remote_url = self.get_remote_url() + if not remote_url: + print("⚠️ 未配置远程仓库") + self.config_remote() + remote_url = self.get_remote_url() + if not remote_url: + return + + print(f"🌐 远程仓库: {remote_url}") + + # 提交本地更改 + print() + if not self.commit_changes(): # 改为交互式提交，不自动 + print("❌ 提交失败，推送中止") + return + + # 强制推送（不拉取远程） + print(f"\n⬆️ 正在强制推送到远程 ({branch})...") + result = self.run_command(f"git push -f origin {branch}") + + if result: + print(f"\n✅ 推送成功!") + else: + print("\n❌ 推送失败") + + def force_pull(self): + """纯粹的远程覆盖本地（不保留本地未推送的修改）""" + print("⬇️ 强制拉取远程到本地") + + if not self.is_git_repo(): + print("❌ 当前目录不是 Git 仓库") + return + + branch = self.get_current_branch() + if not branch: + print("❌ 无法获取当前分支") + return + + print(f"📍 当前分支: {branch}") + + # 检查远程仓库 + remote_url = self.get_remote_url() + if not remote_url: + print("⚠️ 未配置远程仓库") + self.config_remote() + return + + print(f"🌐 远程仓库: {remote_url}") + + # 检查本地更改 + if self.has_changes(): + print("\n⚠️ 检测到未提交的本地更改，这些更改将会被丢弃") + self.run_command("git status --short") + + # 先获取远程最新信息 + print(f"\n📡 正在获取远程仓库信息...") + self.run_command("git fetch origin") + + # 强制重置到远程版本 + print(f"⬇️ 正在强制拉取远程 ({branch})...") + self.run_command(f"git reset --hard origin/{branch}") + self.run_command("git clean -fd") + + print(f"\n✅ 拉取成功!") + + def smart_sync(self): + """智能同步（先拉取合并，再推送）""" + print("🔄 智能同步") + + if not self.is_git_repo(): + print("❌ 当前目录不是 Git 仓库") + return + + branch = self.get_current_branch() + if not branch: + print("❌ 无法获取当前分支") + return + + print(f"📍 当前分支: {branch}") + + # 检查远程仓库 + remote_url = self.get_remote_url() + if not remote_url: + print("⚠️ 未配置远程仓库") + self.config_remote() + return + + print(f"🌐 远程仓库: {remote_url}") + + # 提交本地更改 + print() + if self.has_changes(): + if not self.commit_changes(): # 改为交互式提交 + print("❌ 提交失败") + return + + # 拉取远程更改 + print(f"\n⬇️ 正在拉取远程更新 ({branch})...") + result = self.run_command(f"git pull origin {branch}", check=False) + + if result and result.returncode != 0: + # 如果有冲突 + if "conflict" in result.stderr.lower() or "conflict" in result.stdout.lower(): + print("\n⚠️ 检测到合并冲突!") + print("\n冲突文件:") + self.run_command("git diff --name-only --diff-filter=U") + + choice = input("\n选择操作:\n1. 保留本地版本\n2. 保留远程版本\n3. 手动解决后继续\n请选择 (1/2/3): ").strip() + + if choice == '1': + self.run_command("git checkout --ours .") + self.run_command("git add -A") + self.run_command('git commit --no-edit') + print("✅ 已保留本地版本") + elif choice == '2': + self.run_command("git checkout --theirs .") + self.run_command("git add -A") + self.run_command('git commit --no-edit') + print("✅ 已保留远程版本") + else: + print("ℹ️ 请手动解决冲突后执行: git add . && git commit") + return + else: + print("❌ 拉取失败") + return + else: + print("✅ 拉取成功") + + # 推送到远程 + print(f"\n⬆️ 正在推送到远程 ({branch})...") + push_result = self.run_command(f"git push origin {branch}") + + if push_result: + print(f"\n✅ 同步成功!") + else: + print("⚠️ 推送失败或无新内容需要推送") + + def show_log(self, num=10): + """查看提交历史并支持直接回退""" + print(f"📋 最近 {num} 次提交历史") + + if not self.is_git_repo(): + print("❌ 当前目录不是 Git 仓库") + return + + branch = self.get_current_branch() + print(f"📍 当前分支: {branch}\n") + + # 使用 git log 的简单格式，然后解析 + result = self.run_command(f'git log -{num} --format="COMMIT_START%nHASH:%H%nSUBJECT:%s%nAUTHOR:%an%nDATE:%ar%nCOMMIT_END"') + + if not result: + return + + commits = result.stdout.strip().split('COMMIT_START') + commit_hashes = [] + idx = 1 + + for commit in commits: + if not commit.strip(): + continue + + lines = commit.strip().split('\n') + commit_info = {} + for line in lines: + if ':' in line: + key, value = line.split(':', 1) + commit_info[key] = value.strip() + + if 'HASH' in commit_info: + hash_full = commit_info['HASH'] + hash_short = hash_full[:7] + subject = commit_info.get('SUBJECT', '') + date = commit_info.get('DATE', '') + + commit_hashes.append(hash_full) + print(f"{idx}. [{hash_short}] {subject} -- {date}") + idx += 1 + + # 询问是否回退 + print() + choice = input("是否回退版本? (输入数字1-10回退，直接回车或输入0跳过): ").strip() + + if not choice or choice == '0': + print("✅ 未进行回退") + return + + # 判断是数字还是哈希值 + if choice.isdigit(): + idx = int(choice) - 1 + if 0 <= idx < len(commit_hashes): + commit_hash = commit_hashes[idx][:7] + else: + print(f"❌ 无效的数字，请输入 1-{len(commit_hashes)}") + return + else: + commit_hash = choice + + # 验证提交是否存在 + verify = self.run_command(f"git cat-file -t {commit_hash}", check=False, silent=True) + if not verify or verify.returncode != 0: + print(f"❌ 提交 {commit_hash} 不存在") + return + + # 显示目标提交信息 + print(f"\n📍 即将回退到: {commit_hash}") + + # 选择回退方式 + print("\n回退方式:") + print("1. 软回退 (保留工作区和暂存区的更改)") + print("2. 混合回退 (保留工作区更改，清空暂存区) [推荐]") + print("3. 硬回退 (完全回退，丢弃所有更改)") + + mode_choice = input("\n请选择 (1/2/3，直接回车默认选2): ").strip() + + if not mode_choice: + mode_choice = '2' + + mode_map = { + '1': ('--soft', '软回退'), + '2': ('--mixed', '混合回退'), + '3': ('--hard', '硬回退') + } + + if mode_choice not in mode_map: + print("❌ 无效选择") + return + + mode, mode_name = mode_map[mode_choice] + + # 执行回退 + print(f"\n⏮️ 正在执行{mode_name}...") + result = self.run_command(f"git reset {mode} {commit_hash}") + + if result: + print(f"\n✅ {mode_name}成功!") + print(f"📍 已回退到: {commit_hash}\n") + self.run_command("git status --short") + else: + print(f"\n❌ 回退失败") + + def show_full_history(self): + """查看完整提交历史""" + print("📚 完整提交历史") + + if not self.is_git_repo(): + print("❌ 当前目录不是 Git 仓库") + return + + branch = self.get_current_branch() + print(f"📍 当前分支: {branch}\n") + + # 显示完整历史（简化格式以兼容 Windows） + self.run_command("git log --graph --oneline --decorate --all") + + def rollback(self): + """回退到指定版本""" + print("⏮️ 版本回退") + + if not self.is_git_repo(): + print("❌ 当前目录不是 Git 仓库") + return + + # 显示最近10次提交 + print("📋 最近10次提交:\n") + result = self.run_command("git log -10 --pretty=format:%H|%s|%an|%ar") + + if not result: + return + + lines = result.stdout.strip().split('\n') + commits = [] + + for i, line in enumerate(lines, 1): + if '|' in line: + hash_full, subject, author, date = line.split('|', 3) + hash_short = hash_full[:7] + commits.append(hash_full) + print(f"{i}. [{hash_short}] {subject}") + print(f" 作者: {author} | 时间: {date}\n") + + print() + + # 获取用户输入 + choice = input("请输入要回退到的版本 (输入数字1-10，或输入提交哈希值，q 取消): ").strip() + + if not choice or choice.lower() == 'q': + print("❌ 操作已取消") + return + + # 判断是数字还是哈希值 + if choice.isdigit(): + idx = int(choice) - 1 + if 0 <= idx < len(commits): + commit_hash = commits[idx][:7] + else: + print(f"❌ 无效的数字，请输入 1-{len(commits)}") + return + else: + commit_hash = choice + + # 验证提交是否存在 + verify = self.run_command(f"git cat-file -t {commit_hash}", check=False, silent=True) + if not verify or verify.returncode != 0: + print(f"❌ 提交 {commit_hash} 不存在") + return + + # 显示目标提交信息 + print(f"\n📍 即将回退到:") + self.run_command(f"git log -1 {commit_hash} --pretty=format:[%h] %s%n作者: %an | 时间: %ar") + + # 选择回退方式 + print("\n\n回退方式:") + print("1. 软回退 (保留工作区和暂存区的更改)") + print("2. 混合回退 (保留工作区更改，清空暂存区) [推荐]") + print("3. 硬回退 (完全回退，丢弃所有更改)") + + mode_choice = input("\n请选择 (1/2/3，输入 q 取消): ").strip() + + if mode_choice.lower() == 'q': + print("❌ 操作已取消") + return + + mode_map = { + '1': ('--soft', '软回退'), + '2': ('--mixed', '混合回退'), + '3': ('--hard', '硬回退') + } + + if mode_choice not in mode_map: + print("❌ 无效选择") + return + + mode, mode_name = mode_map[mode_choice] + + # 执行回退 + print(f"\n⏮️ 正在执行{mode_name}...") + result = self.run_command(f"git reset {mode} {commit_hash}") + + if result: + print(f"\n✅ {mode_name}成功!") + print(f"📍 已回退到: {commit_hash}\n") + self.run_command("git status --short") + else: + print(f"\n❌ 回退失败") + + +def main(): + tool = GitSyncTool() + + if len(sys.argv) < 2: + print(__doc__) + return + + command = sys.argv[1].lower() + + if command == 'push': + tool.force_push() + elif command == 'pull': + tool.force_pull() + elif command == 'sync': + tool.smart_sync() + elif command == 'log': + tool.show_log(5) + elif command == 'history': + tool.show_full_history() + elif command == 'rollback': + tool.rollback() + elif command == 'config': + tool.config_remote() + else: + print(f"❌ 未知命令: {command}") + print(__doc__) + + +if __name__ == '__main__': + main() \ No newline at end of file diff --git a/image/comfyui-node-apikey.png b/image/comfyui-node-apikey.png new file mode 100644 index 0000000..75c2d06 Binary files /dev/null and b/image/comfyui-node-apikey.png differ diff --git a/image/comfyui-node-auth.png b/image/comfyui-node-auth.png new file mode 100644 index 0000000..98aa1a8 Binary files /dev/null and b/image/comfyui-node-auth.png differ diff --git a/py/IMAGE_GPT_apikey.py b/py/IMAGE_GPT_apikey.py new file mode 100644 index 0000000..317f203 --- /dev/null +++ b/py/IMAGE_GPT_apikey.py @@ -0,0 +1,312 @@ +""" +PD: OpenAI GPT Image Node (ComfyUI API Key) +使用 ComfyUI API Key 调用 ComfyUI 代理 API +参数和官方节点完全一样 +""" + +import torch +import requests +import time +import io +import base64 +import numpy as np +from PIL import Image + + +def tensor_to_bytes(tensor, max_pixels=2048*2048): + """将 Tensor 转换为 PNG BytesIO，并进行下采样""" + if tensor.ndim == 4: + tensor = tensor[0] + + # 下采样大图片 + height, width = tensor.shape[:2] + current_pixels = height * width + if current_pixels > max_pixels: + scale = (max_pixels / current_pixels) ** 0.5 + new_height = int(height * scale) + new_width = int(width * scale) + + # 使用 PIL 进行高质量缩放 + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + image = image.resize((new_width, new_height), Image.Resampling.LANCZOS) + else: + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + + buffered = io.BytesIO() + image.save(buffered, format="PNG") + buffered.seek(0) + return buffered + + +def load_image_to_tensor(image_data): + """将图片数据（URL 或 bytes）转换为 Tensor""" + try: + if isinstance(image_data, str) and image_data.startswith("http"): + response = requests.get(image_data, timeout=30) + response.raise_for_status() + img = Image.open(io.BytesIO(response.content)) + else: + img = Image.open(io.BytesIO(image_data)) + + img = img.convert("RGB") + image_np = np.array(img).astype(np.float32) / 255.0 + return torch.from_numpy(image_np).unsqueeze(0) + except Exception as e: + print(f"[IMAGE_GPT_APIKEY] Error loading image: {e}") + return torch.zeros((1, 512, 512, 3)) + + +class PDOpenAIGPTImageAPIKey: + """ + PD: OpenAI GPT Image (API Key) + 使用 ComfyUI API Key 调用 ComfyUI 代理 API + """ + + @classmethod + def INPUT_TYPES(cls): + return { + "required": { + "api_key": ("STRING", { + "default": "", + "multiline": False, + "placeholder": "Your ComfyUI API Key" + }), + "prompt": ("STRING", { + "multiline": True, + "default": "", + "tooltip": "Text prompt for GPT Image" + }), + "seed": ("INT", { + "default": 0, + "min": 0, + "max": 2**31 - 1, + "step": 1, + "tooltip": "not implemented yet in backend" + }), + "quality": (["low", "medium", "high"], { + "default": "low", + "tooltip": "Image quality, affects cost and generation time" + }), + "background": (["auto", "opaque", "transparent"], { + "default": "auto", + "tooltip": "Return image with or without background" + }), + "size": (["auto", "1024x1024", "1024x1536", "1536x1024"], { + "default": "auto", + "tooltip": "Image size" + }), + "n": ("INT", { + "default": 1, + "min": 1, + "max": 8, + "step": 1, + "tooltip": "How many images to generate" + }), + }, + "optional": { + "image": ("IMAGE", { + "tooltip": "Optional reference image for image editing." + }), + "mask": ("MASK", { + "tooltip": "Optional mask for inpainting (white areas will be replaced)" + }), + "model": (["gpt-image-1", "gpt-image-1.5"], { + "default": "gpt-image-1" + }), + }, + "hidden": { + "unique_id": "UNIQUE_ID", + } + } + + RETURN_TYPES = ("IMAGE", "STRING") + RETURN_NAMES = ("image", "info") + FUNCTION = "generate_image" + CATEGORY = "PD_Tools/Image_Generation" + + def generate_image(self, api_key, prompt, seed, quality, background, size, n, unique_id, image=None, mask=None, model="gpt-image-1"): + """生成或编辑图片""" + + # 验证 API Key + if not api_key or not api_key.strip(): + error_msg = "Error: ComfyUI API Key is required" + print(f"[IMAGE_GPT_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 清理 API Key + api_key = api_key.strip() + + # 判断模式：编辑 vs 生成 + is_edit_mode = image is not None + + # ComfyUI 代理 API URL + if is_edit_mode: + base_url = "https://api.comfy.org/proxy/openai/images/edits" + else: + base_url = "https://api.comfy.org/proxy/openai/images/generations" + + start_time = time.time() + + try: + if is_edit_mode: + # 图片编辑模式 - 使用 multipart/form-data + # 准备文件 + files = {} + img_bytes = tensor_to_bytes(image) + files['image'] = ('image.png', img_bytes, 'image/png') + + # 处理 mask + if mask is not None: + m_tensor = mask + if m_tensor.ndim == 3: + m_tensor = m_tensor[0] + + mask_h, mask_w = m_tensor.shape + rgba = np.zeros((mask_h, mask_w, 4), dtype=np.uint8) + + mask_np = m_tensor.cpu().numpy() + alpha = ((1.0 - mask_np) * 255).astype(np.uint8) + rgba[..., 3] = alpha + + mask_pil = Image.fromarray(rgba) + mask_buffer = io.BytesIO() + mask_pil.save(mask_buffer, format="PNG") + mask_buffer.seek(0) + files['mask'] = ('mask.png', mask_buffer, 'image/png') + + # 表单数据 - multipart/form-data 模式下所有值都应该是字符串 + data = { + "model": model, + "prompt": prompt, + "n": str(n), + "size": size, + "quality": quality, + "background": background, + "moderation": "low" + } + + # 请求头 + headers = { + "X-API-KEY": api_key, + "Accept": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + print(f"[IMAGE_GPT_APIKEY] Sending edit request") + print(f"[IMAGE_GPT_APIKEY] Model: {model}, Size: {size}, Quality: {quality}") + print(f"[IMAGE_GPT_APIKEY] Image size: {image.shape}, Mask: {'Yes' if mask is not None else 'No'}") + + response = requests.post(base_url, headers=headers, files=files, data=data, timeout=300) + + else: + # 文生图模式 - 使用 JSON + payload = { + "model": model, + "prompt": prompt, + "n": n, + "size": size, + "quality": quality, + "background": background, + "response_format": "url" + } + + headers = { + "X-API-KEY": api_key, + "Accept": "application/json", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + print(f"[IMAGE_GPT_APIKEY] Sending generation request") + print(f"[IMAGE_GPT_APIKEY] Model: {model}, Size: {size}, Quality: {quality}") + print(f"[IMAGE_GPT_APIKEY] Prompt length: {len(prompt)} chars") + + response = requests.post(base_url, headers=headers, json=payload, timeout=300) + + # 检查响应 + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[IMAGE_GPT_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + result = response.json() + data_list = result.get("data", []) + + if not data_list: + error_msg = "Error: No image data in response" + print(f"[IMAGE_GPT_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 提取图片 + image_tensors = [] + for item in data_list: + if "url" in item: + img_tensor = load_image_to_tensor(item["url"]) + image_tensors.append(img_tensor) + elif "b64_json" in item: + img_data = base64.b64decode(item["b64_json"]) + img_tensor = load_image_to_tensor(img_data) + image_tensors.append(img_tensor) + + if not image_tensors: + error_msg = "Error: Failed to process output images" + print(f"[IMAGE_GPT_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + + # 计算耗时和成本 + end_time = time.time() + duration = end_time - start_time + + # 估算成本 + cost_map = { + "gpt-image-1": {"low": 0.02, "medium": 0.04, "high": 0.08}, + "gpt-image-1.5": {"low": 0.03, "medium": 0.06, "high": 0.12} + } + cost = cost_map.get(model, {}).get(quality, 0.04) + total_cost = cost * n + + info_str = ( + f"Model: {model}\n" + f"Mode: {'Edit' if is_edit_mode else 'Generate'}\n" + f"Status: Success\n" + f"Time: {duration:.2f}s\n" + f"Images: {len(image_tensors)}\n" + f"Size: {size}\n" + f"Quality: {quality}\n" + f"Background: {background}\n" + f"Est. Cost: ${total_cost:.3f}" + ) + + print(f"[IMAGE_GPT_APIKEY] Success! Generated {len(image_tensors)} images in {duration:.2f}s") + return (final_image, info_str) + + except requests.exceptions.Timeout: + error_msg = "Error: Request timeout (300s)" + print(f"[IMAGE_GPT_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except requests.exceptions.ConnectionError as e: + error_msg = f"Error: Connection failed - {str(e)}" + print(f"[IMAGE_GPT_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except Exception as e: + error_msg = f"Error: {str(e)}" + print(f"[IMAGE_GPT_APIKEY] {error_msg}") + import traceback + traceback.print_exc() + return (torch.zeros((1, 512, 512, 3)), error_msg) + + +# 节点注册 +NODE_CLASS_MAPPINGS = { + "PDOpenAIGPTImageAPIKey": PDOpenAIGPTImageAPIKey +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDOpenAIGPTImageAPIKey": "PD: GPT Image (apikey)" +} diff --git a/py/IMAGE_GPT_authtoken.py b/py/IMAGE_GPT_authtoken.py new file mode 100644 index 0000000..d8d392e --- /dev/null +++ b/py/IMAGE_GPT_authtoken.py @@ -0,0 +1,313 @@ +""" +PD: OpenAI GPT Image Node (ComfyUI Auth Token) +基于 gemini_image_api 的 Auth Token 认证方式 +支持文生图和图片编辑 +""" + +import torch +import requests +import time +import io +import base64 +import numpy as np +from PIL import Image + + +def tensor_to_bytes(tensor, max_pixels=2048*2048): + """将 Tensor 转换为 PNG BytesIO，并进行下采样""" + if tensor.ndim == 4: + tensor = tensor[0] + + # 下采样大图片 + height, width = tensor.shape[:2] + current_pixels = height * width + if current_pixels > max_pixels: + scale = (max_pixels / current_pixels) ** 0.5 + new_height = int(height * scale) + new_width = int(width * scale) + + # 使用 PIL 进行高质量缩放 + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + image = image.resize((new_width, new_height), Image.Resampling.LANCZOS) + else: + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + + buffered = io.BytesIO() + image.save(buffered, format="PNG") + buffered.seek(0) + return buffered + + +def load_image_to_tensor(image_data): + """将图片数据（URL 或 bytes）转换为 Tensor""" + try: + if isinstance(image_data, str) and image_data.startswith("http"): + response = requests.get(image_data, timeout=30) + response.raise_for_status() + img = Image.open(io.BytesIO(response.content)) + else: + img = Image.open(io.BytesIO(image_data)) + + img = img.convert("RGB") + image_np = np.array(img).astype(np.float32) / 255.0 + return torch.from_numpy(image_np).unsqueeze(0) + except Exception as e: + print(f"[IMAGE_GPT_AUTHTOKEN] Error loading image: {e}") + return torch.zeros((1, 512, 512, 3)) + + +class PDOpenAIGPTImageAuthToken: + """ + PD: OpenAI GPT Image (Auth Token) + 使用 ComfyUI Auth Token 调用 GPT Image API + """ + + @classmethod + def INPUT_TYPES(cls): + return { + "required": { + "auth_token": ("STRING", { + "default": "", + "multiline": False, + "placeholder": "Paste your Auth Token here" + }), + "prompt": ("STRING", { + "multiline": True, + "default": "", + "tooltip": "Describe what you want to generate or edit" + }), + "model": (["gpt-image-1", "gpt-image-1.5"], { + "default": "gpt-image-1", + "tooltip": "GPT Image model version" + }), + "quality": (["low", "medium", "high"], { + "default": "low", + "tooltip": "Image quality, affects cost and generation time" + }), + "background": (["auto", "opaque", "transparent"], { + "default": "auto", + "tooltip": "Background transparency (GPT Image only)" + }), + "size": (["auto", "1024x1024", "1024x1536", "1536x1024"], { + "default": "auto", + "tooltip": "Image size" + }), + "n": ("INT", { + "default": 1, + "min": 1, + "max": 8, + "step": 1, + "tooltip": "Number of images to generate" + }), + "seed": ("INT", { + "default": 0, + "min": 0, + "max": 0xFFFFFFFFFFFFFFFF + }), + }, + "optional": { + "image": ("IMAGE", {"default": None, "tooltip": "Input image for editing"}), + "mask": ("MASK", {"default": None, "tooltip": "Mask for inpainting"}), + }, + "hidden": { + "unique_id": "UNIQUE_ID", + } + } + + RETURN_TYPES = ("IMAGE", "STRING") + RETURN_NAMES = ("image", "info") + FUNCTION = "generate_image" + CATEGORY = "PD_Tools/Image_Generation" + + def generate_image(self, auth_token, prompt, model, quality, background, size, n, seed, unique_id, image=None, mask=None): + """生成或编辑图片""" + + # 验证 Token + if not auth_token or not auth_token.strip(): + error_msg = "Error: Auth Token is required" + print(f"[IMAGE_GPT_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 清理 Token + auth_token = auth_token.strip() + if auth_token.lower().startswith("bearer "): + auth_token = auth_token[7:].strip() + + # API 基础地址 + base_url = "https://api.comfy.org" + + # 判断模式：编辑 vs 生成 + is_edit_mode = image is not None + + start_time = time.time() + + try: + if is_edit_mode: + # 图片编辑模式 - 使用 multipart/form-data + url = f"{base_url}/proxy/openai/images/edits" + + # 准备文件 + files = {} + img_bytes = tensor_to_bytes(image) + files['image'] = ('image.png', img_bytes, 'image/png') + + # 处理 mask + if mask is not None: + m_tensor = mask + if m_tensor.ndim == 3: + m_tensor = m_tensor[0] + + mask_h, mask_w = m_tensor.shape + rgba = np.zeros((mask_h, mask_w, 4), dtype=np.uint8) + + mask_np = m_tensor.cpu().numpy() + alpha = ((1.0 - mask_np) * 255).astype(np.uint8) + rgba[..., 3] = alpha + + mask_pil = Image.fromarray(rgba) + mask_buffer = io.BytesIO() + mask_pil.save(mask_buffer, format="PNG") + mask_buffer.seek(0) + files['mask'] = ('mask.png', mask_buffer, 'image/png') + + # 表单数据 - multipart/form-data 模式下所有值都应该是字符串 + data = { + "model": model, + "prompt": prompt, + "n": str(n), + "size": size, + "quality": quality, + "background": background, + "moderation": "low" + } + + # 请求头 + headers = { + "Authorization": f"Bearer {auth_token}", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + print(f"[IMAGE_GPT_AUTHTOKEN] Sending edit request to {url}") + print(f"[IMAGE_GPT_AUTHTOKEN] Model: {model}, Size: {size}, Quality: {quality}, Background: {background}") + print(f"[IMAGE_GPT_AUTHTOKEN] Image size: {image.shape}, Mask: {'Yes' if mask is not None else 'No'}") + + response = requests.post(url, headers=headers, files=files, data=data, timeout=300) + + else: + # 文生图模式 - 使用 JSON + url = f"{base_url}/proxy/openai/images/generations" + + payload = { + "model": model, + "prompt": prompt, + "n": n, + "size": size, + "quality": quality, + "background": background, + "moderation": "low", + "response_format": "url" + } + + headers = { + "Authorization": f"Bearer {auth_token}", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + print(f"[IMAGE_GPT_AUTHTOKEN] Sending generation request to {url}") + print(f"[IMAGE_GPT_AUTHTOKEN] Model: {model}, Size: {size}, Quality: {quality}, Background: {background}") + print(f"[IMAGE_GPT_AUTHTOKEN] Prompt length: {len(prompt)} chars") + + response = requests.post(url, headers=headers, json=payload, timeout=300) + + # 检查响应 + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[IMAGE_GPT_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + result = response.json() + data_list = result.get("data", []) + + if not data_list: + error_msg = "Error: No image data in response" + print(f"[IMAGE_GPT_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 提取图片 + image_tensors = [] + for item in data_list: + if "url" in item: + img_tensor = load_image_to_tensor(item["url"]) + image_tensors.append(img_tensor) + elif "b64_json" in item: + img_data = base64.b64decode(item["b64_json"]) + img_tensor = load_image_to_tensor(img_data) + image_tensors.append(img_tensor) + + if not image_tensors: + error_msg = "Error: Failed to process output images" + print(f"[IMAGE_GPT_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + + # 计算耗时和成本 + end_time = time.time() + duration = end_time - start_time + + # 估算成本 + if model == "gpt-image-1": + # GPT Image 1: $10/1M input tokens, $40/1M output tokens + # 估算：约 1000 tokens input, 2000 tokens output per image + cost = (1000 * 10 + 2000 * 40) / 1_000_000 + elif model == "gpt-image-1.5": + # GPT Image 1.5: $8/1M input tokens, $32/1M output tokens + cost = (1000 * 8 + 2000 * 32) / 1_000_000 + else: + cost = 0.10 # 默认估算 + total_cost = cost * n + + info_str = ( + f"Model: {model}\n" + f"Mode: {'Edit' if is_edit_mode else 'Generate'}\n" + f"Status: Success\n" + f"Time: {duration:.2f}s\n" + f"Images: {len(image_tensors)}\n" + f"Size: {size}\n" + f"Quality: {quality}\n" + f"Background: {background}\n" + f"Est. Cost: ${total_cost:.4f}" + ) + + print(f"[IMAGE_GPT_AUTHTOKEN] Success! Generated {len(image_tensors)} images in {duration:.2f}s") + return (final_image, info_str) + + except requests.exceptions.Timeout: + error_msg = "Error: Request timeout (300s)" + print(f"[IMAGE_GPT_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except requests.exceptions.ConnectionError as e: + error_msg = f"Error: Connection failed - {str(e)}" + print(f"[IMAGE_GPT_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except Exception as e: + error_msg = f"Error: {str(e)}" + print(f"[IMAGE_GPT_AUTHTOKEN] {error_msg}") + import traceback + traceback.print_exc() + return (torch.zeros((1, 512, 512, 3)), error_msg) + + +# 节点注册 +NODE_CLASS_MAPPINGS = { + "PDOpenAIGPTImageAuthToken": PDOpenAIGPTImageAuthToken +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDOpenAIGPTImageAuthToken": "PD: GPT Image (comfyui_AuthToken)" +} diff --git a/py/IMAGE_GeminiNanoBanana2.py b/py/IMAGE_GeminiNanoBanana2.py new file mode 100644 index 0000000..5d82445 --- /dev/null +++ b/py/IMAGE_GeminiNanoBanana2.py @@ -0,0 +1,279 @@ +""" +PD: Nano Banana 2 (ComfyUI Key) +基于 Google Gemini 3.1 Flash Image 模型的图像生成/编辑节点 +支持进度反馈、Token 成本计算、及其独立输出接口 +""" + +import torch +import base64 +import json +import time +import uuid +import requests +import numpy as np +import io +from PIL import Image +from typing import Optional, List +from server import PromptServer +import comfy.utils + +# 默认系统提示词 +GEMINI_IMAGE_SYS_PROMPT = ( + "You are an expert image-generation engine. You must ALWAYS produce an image.\n" + "Interpret all user input—regardless of " + "format, intent, or abstraction—as literal visual directives for image composition.\n" + "If a prompt is conversational or lacks specific visual details, " + "you must creatively invent a concrete visual scenario that depicts the concept.\n" + "Prioritize generating the visual representation above any text, formatting, or conversational requests." +) + +def tensor_to_base64(tensor): + """将 Tensor 转换为 base64 字符串""" + if tensor.ndim == 4: + tensor = tensor[0] + + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + buffered = io.BytesIO() + image.save(buffered, format="PNG") + return base64.b64encode(buffered.getvalue()).decode("utf-8") + +def bytes_to_tensor(data): + """将图片字节数据转换为 Tensor""" + try: + img = Image.open(io.BytesIO(data)).convert("RGB") + image_np = np.array(img).astype(np.float32) / 255.0 + return torch.from_numpy(image_np).unsqueeze(0) + except Exception as e: + print(f"[PD_NB2] Image decode error: {e}") + return torch.zeros((1, 512, 512, 3)) + +class PDGeminiNanoBanana2: + @classmethod + def INPUT_TYPES(cls): + return { + "required": { + "comfy_api_key": ("STRING", { + "default": "", + "multiline": False, + "placeholder": "Paste your ComfyUI API Key here" + }), + "prompt": ("STRING", { + "multiline": True, + "default": "A futuristic cityscape with flying cars and neon lights", + }), + "model": (["Nano Banana 2 (Gemini 3.1 Flash Image)"], { + "default": "Nano Banana 2 (Gemini 3.1 Flash Image)" + }), + "seed": ("INT", { + "default": 42, + "min": 0, + "max": 0x7FFFFFFF, + }), + "aspect_ratio": (["auto", "1:1", "2:3", "3:2", "3:4", "4:3", "4:5", "5:4", "9:16", "16:9", "21:9"], { + "default": "auto", + }), + "resolution": (["1K", "2K", "4K"], { + "default": "1K", + }), + "response_modalities": (["IMAGE", "IMAGE+TEXT"], { + "default": "IMAGE", + }), + "thinking_level": (["MINIMAL", "HIGH"], { + "default": "MINIMAL" + }), + }, + "optional": { + "images": ("IMAGE", ), + "files": ("GEMINI_INPUT_FILES", ), + "system_prompt": ("STRING", { + "multiline": True, + "default": GEMINI_IMAGE_SYS_PROMPT, + }), + }, + "hidden": { + "unique_id": "UNIQUE_ID", + } + } + + RETURN_TYPES = ("IMAGE", "STRING", "STRING") + RETURN_NAMES = ("image", "info", "price") + FUNCTION = "generate_image" + CATEGORY = "PD_Tools/Image_Generation" + + def generate_image(self, comfy_api_key, prompt, model, seed, aspect_ratio, resolution, response_modalities, thinking_level, unique_id, images=None, files=None, system_prompt=""): + pbar = comfy.utils.ProgressBar(100) + pbar.update(10) + + if not comfy_api_key or not comfy_api_key.strip(): + return (torch.zeros((1, 512, 512, 3)), "Error: Missing ComfyUI API Key", "$0.00") + + api_key = comfy_api_key.strip() + api_model = "gemini-3.1-flash-image-preview" + + # 发送进度文字到前端 + self._send_status(unique_id, "Preparing tokens...") + + # 构建请求内容 + parts = [{"text": prompt}] + + if images is not None: + for i in range(min(images.shape[0], 14)): + b64_str = tensor_to_base64(images[i]) + parts.append({ + "inlineData": { + "mimeType": "image/png", + "data": b64_str + } + }) + + # Files 输入处理 + if files: + for file_part in files: + if hasattr(file_part, "model_dump"): + parts.append(file_part.model_dump(exclude_none=True)) + else: + parts.append(file_part) + + # 核心参数修正：匹配官方 Pydantic 模型的 camelCase 结构 + # 且不发送 seed (官方 GeminiNanoBanana2.execute 并没有发送 seed 到 API) + image_config = { + "imageSize": resolution, + "imageOutputOptions": {"mimeType": "image/png"} + } + if aspect_ratio != "auto": + image_config["aspectRatio"] = aspect_ratio + + generation_config = { + "responseModalities": (["IMAGE"] if response_modalities == "IMAGE" else ["TEXT", "IMAGE"]), + "imageConfig": image_config, + "thinkingConfig": {"thinkingLevel": thinking_level} + } + + payload = { + "contents": [{ + "role": "user", + "parts": parts + }], + "generationConfig": generation_config + } + + if system_prompt and system_prompt.strip(): + payload["systemInstruction"] = { + "parts": [{"text": system_prompt.strip()}] + } + + headers = { + "X-API-KEY": api_key, + "Accept": "application/json", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + base_url = f"https://api.comfy.org/proxy/vertexai/gemini/{api_model}" + + self._send_status(unique_id, "Sending request to Google...") + pbar.update(30) + + start_time = time.time() + try: + # 打印发送的 Payload 调试 (可选) + # print(f"[PD_NB2] Payload: {json.dumps(payload)}") + + response = requests.post(base_url, headers=headers, json=payload, timeout=120) + pbar.update(80) + + if response.status_code != 200: + error_body = response.text + error_msg = f"API Error {response.status_code}: {error_body}" + self._send_status(unique_id, "Error occurred") + return (torch.zeros((1, 512, 512, 3)), error_msg, "N/A") + + result = response.json() + image_tensors = [] + text_parts = [] + + self._send_status(unique_id, "Processing response...") + + candidates = result.get("candidates", []) + for cand in candidates: + content = cand.get("content", {}) + parts_out = content.get("parts", []) + for part in parts_out: + inline_data = part.get("inline_data") or part.get("inlineData") + if inline_data: + data_b64 = inline_data.get("data") + if data_b64: + raw_data = base64.b64decode(data_b64) + image_tensors.append(bytes_to_tensor(raw_data)) + + if "text" in part and part["text"]: + text_parts.append(part["text"]) + + if not image_tensors: + return (torch.zeros((1, 512, 512, 3)), "Error: No images found in response", "N/A") + + final_image = torch.cat(image_tensors, dim=0) + final_text = "\n".join(text_parts) if text_parts else "No description" + + # 计算成本 + cost_info = self._calculate_cost(result, api_model, time.time() - start_time) + + pbar.update(100) + self._send_status(unique_id, "Success") + + return (final_image, final_text, cost_info) + + except Exception as e: + self._send_status(unique_id, "Exception caught") + return (torch.zeros((1, 512, 512, 3)), f"Error: {str(e)}", "N/A") + + def _send_status(self, node_id, text): + """发送状态文字到节点下方显示""" + try: + PromptServer.instance.send_sync("progress_text", {"text": text, "node": node_id}) + except: + pass + + def _calculate_cost(self, result, model, duration): + """基于 API 响应计算成本""" + usage = result.get("usageMetadata", {}) + prompt_tokens = usage.get("promptTokenCount", 0) + candidates_tokens = usage.get("candidatesTokenCount", 0) + details = usage.get("candidatesTokensDetails", []) + thoughts_tokens = usage.get("thoughtsTokenCount", 0) + + # 价格配置 (3.1 Flash Image Preview) + prices = {"input": 0.5, "output_text": 3.0, "output_image": 60.0} + + text_tokens = thoughts_tokens + img_tokens = 0 + + for d in details: + modality = d.get("modality") + count = d.get("tokenCount", 0) + if modality == "IMAGE": + img_tokens += count + else: + text_tokens += count + + if not details and candidates_tokens > 0: + text_tokens += candidates_tokens + + input_cost = (prompt_tokens / 1_000_000) * prices["input"] + text_cost = (text_tokens / 1_000_000) * prices["output_text"] + img_cost = (img_tokens / 1_000_000) * prices["output_image"] + total_cost = input_cost + text_cost + img_cost + + info = [ + f"Model: {model}", + f"Status: Success", + f"Time: {duration:.2f}s", + f"Tokens: In:{prompt_tokens:,} | Out Text:{text_tokens:,} | Out Img:{img_tokens:,}", + f"Est. Cost: ${total_cost:.6f}" + ] + return "\n".join(info) + +# 注册 +NODE_CLASS_MAPPINGS = {"PDGeminiNanoBanana2": PDGeminiNanoBanana2} +NODE_DISPLAY_NAME_MAPPINGS = {"PDGeminiNanoBanana2": "PD: Nano Banana 2 (ComfyUI Key)"} diff --git a/py/IMGE_GeminiNode_apikey.py b/py/IMGE_GeminiNode_apikey.py new file mode 100644 index 0000000..b72c782 --- /dev/null +++ b/py/IMGE_GeminiNode_apikey.py @@ -0,0 +1,284 @@ +""" +PD: Gemini Image Generation Node (ComfyUI API Key) +使用 ComfyUI API Key 调用 ComfyUI 代理 API +参数和官方节点完全一样 +""" + +import torch +import requests +import time +import io +import base64 +import numpy as np +from PIL import Image + +# 默认系统提示词（来自 ComfyUI 官方 API 节点） +GEMINI_IMAGE_SYS_PROMPT = ( + "You are an expert image-generation engine. You must ALWAYS produce an image.\n" + "Interpret all user input—regardless of " + "format, intent, or abstraction—as literal visual directives for image composition.\n" + "If a prompt is conversational or lacks specific visual details, " + "you must creatively invent a concrete visual scenario that depicts the concept.\n" + "Prioritize generating the visual representation above any text, formatting, or conversational requests." +) + + +def tensor_to_base64(tensor): + """将 Tensor 转换为 base64 字符串""" + if tensor.ndim == 4: + tensor = tensor[0] + + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + buffered = io.BytesIO() + image.save(buffered, format="PNG") + return base64.b64encode(buffered.getvalue()).decode("utf-8") + + +def bytes_to_tensor(data): + """将图片字节数据转换为 Tensor""" + try: + img = Image.open(io.BytesIO(data)).convert("RGB") + image_np = np.array(img).astype(np.float32) / 255.0 + return torch.from_numpy(image_np).unsqueeze(0) + except Exception as e: + print(f"[GEMINI_APIKEY] Image decode error: {e}") + return torch.zeros((1, 512, 512, 3)) + + +class PDGeminiImageGenAPIKey: + """ + PD: Gemini Image Generation (API Key) + 使用 ComfyUI API Key 调用 ComfyUI 代理 API + """ + + @classmethod + def INPUT_TYPES(cls): + return { + "required": { + "api_key": ("STRING", { + "default": "", + "multiline": False, + "placeholder": "Your ComfyUI API Key" + }), + "prompt": ("STRING", { + "multiline": True, + "default": "A futuristic city with flying cars", + "tooltip": "Describe what you want to generate" + }), + "model": ([ + "gemini-2.5-flash-image", + "gemini-2.5-flash-image-preview" + ], { + "default": "gemini-2.5-flash-image" + }), + "aspect_ratio": ([ + "auto", "1:1", "2:3", "3:2", "3:4", "4:3", "4:5", "5:4", "9:16", "16:9", "21:9" + ], { + "default": "auto" + }), + "response_modalities": (["IMAGE+TEXT", "IMAGE"], { + "default": "IMAGE+TEXT", + "tooltip": "Choose 'IMAGE' for image-only output, or 'IMAGE+TEXT' for both" + }), + "seed": ("INT", { + "default": 42, + "min": 0, + "max": 0xFFFFFFFFFFFFFFFF + }), + }, + "optional": { + "image_ref": ("IMAGE", { + "tooltip": "Reference image(s) for image-to-image generation" + }), + "system_prompt": ("STRING", { + "multiline": True, + "default": GEMINI_IMAGE_SYS_PROMPT, + "tooltip": "Optional system instructions" + }), + }, + "hidden": { + "unique_id": "UNIQUE_ID", + } + } + + RETURN_TYPES = ("IMAGE", "STRING") + RETURN_NAMES = ("image", "info") + FUNCTION = "generate_image" + CATEGORY = "PD_Tools/Image_Generation" + + def generate_image(self, api_key, prompt, model, aspect_ratio, response_modalities, seed, unique_id, image_ref=None, system_prompt=""): + """生成图片""" + + # 验证 API Key + if not api_key or not api_key.strip(): + error_msg = "Error: ComfyUI API Key is required" + print(f"[GEMINI_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 清理 API Key + api_key = api_key.strip() + + # ComfyUI 代理 API URL + base_url = f"https://api.comfy.org/proxy/vertexai/gemini/{model}" + + # 构建请求内容 + parts = [] + + # 添加用户提示词 + parts.append({"text": prompt}) + + # 添加参考图片（如果有） + if image_ref is not None: + for i in range(image_ref.shape[0]): + b64_str = tensor_to_base64(image_ref[i]) + parts.append({ + "inlineData": { + "mimeType": "image/png", + "data": b64_str + } + }) + print(f"[GEMINI_APIKEY] Added reference image {i+1}") + + # 生成配置 + generation_config = { + "responseModalities": ["IMAGE"] if response_modalities == "IMAGE" else ["TEXT", "IMAGE"] + } + + # 图片配置 + if aspect_ratio != "auto": + generation_config["imageConfig"] = { + "aspectRatio": aspect_ratio + } + + # 系统指令 + system_instruction = None + if system_prompt and system_prompt.strip(): + system_instruction = { + "parts": [{"text": system_prompt.strip()}] + } + + # 完整请求体 + payload = { + "contents": [{ + "role": "user", + "parts": parts + }], + "generationConfig": generation_config + } + + if system_instruction: + payload["systemInstruction"] = system_instruction + + # 请求头 + headers = { + "X-API-KEY": api_key, + "Accept": "application/json", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + start_time = time.time() + + try: + print(f"[GEMINI_APIKEY] Sending request to {model}") + print(f"[GEMINI_APIKEY] Aspect Ratio: {aspect_ratio}") + + response = requests.post(base_url, headers=headers, json=payload, timeout=120) + + # 检查响应 + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[GEMINI_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + result = response.json() + + # 提取图片和文本 + image_tensors = [] + text_parts = [] + candidates = result.get("candidates", []) + + if not candidates: + error_msg = "Error: No candidates in response" + print(f"[GEMINI_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + for cand in candidates: + content = cand.get("content", {}) + parts = content.get("parts", []) + + for part in parts: + # 提取图片 + inline_data = part.get("inline_data") or part.get("inlineData") + + if inline_data: + mime = inline_data.get("mime_type") or inline_data.get("mimeType") + data_b64 = inline_data.get("data") + + if mime in ["image/png", "image/jpeg"] and data_b64: + raw_data = base64.b64decode(data_b64) + image_tensors.append(bytes_to_tensor(raw_data)) + print(f"[GEMINI_APIKEY] Extracted image ({mime})") + + # 提取文本 + if "text" in part and part["text"]: + text_parts.append(part["text"]) + + if not image_tensors: + error_msg = f"Error: No images found in response" + print(f"[GEMINI_APIKEY] {error_msg}") + print(f"[GEMINI_APIKEY] Response: {str(result)[:500]}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + final_text = "\n".join(text_parts) if text_parts else "No text output" + + # 计算耗时和成本 + end_time = time.time() + duration = end_time - start_time + + # 估算成本 + cost_per_img = 0.04 + total_cost = cost_per_img * len(image_tensors) + + info_str = ( + f"Model: {model}\n" + f"Status: Success\n" + f"Time: {duration:.2f}s\n" + f"Images: {len(image_tensors)}\n" + f"Aspect Ratio: {aspect_ratio}\n" + f"Est. Cost: ${total_cost:.4f}\n" + f"---\n" + f"{final_text}" + ) + + print(f"[GEMINI_APIKEY] Success! Generated {len(image_tensors)} images in {duration:.2f}s") + return (final_image, info_str) + + except requests.exceptions.Timeout: + error_msg = "Error: Request timeout (120s)" + print(f"[GEMINI_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except requests.exceptions.ConnectionError as e: + error_msg = f"Error: Connection failed - {str(e)}" + print(f"[GEMINI_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except Exception as e: + error_msg = f"Error: {str(e)}" + print(f"[GEMINI_APIKEY] {error_msg}") + import traceback + traceback.print_exc() + return (torch.zeros((1, 512, 512, 3)), error_msg) + + +# 节点注册 +NODE_CLASS_MAPPINGS = { + "PDGeminiImageGenAPIKey": PDGeminiImageGenAPIKey +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDGeminiImageGenAPIKey": "PD: Gemini Image (apikey)" +} diff --git a/py/IMGE_GeminiNode_authtoken.py b/py/IMGE_GeminiNode_authtoken.py new file mode 100644 index 0000000..7a4597c --- /dev/null +++ b/py/IMGE_GeminiNode_authtoken.py @@ -0,0 +1,281 @@ +""" +PD: Gemini Image Generation Node (ComfyUI Auth Token) +基于 gemini_image_api 的 Auth Token 认证方式 +支持文生图和图生图 +""" + +import torch +import requests +import time +import io +import base64 +import numpy as np +from PIL import Image + +# 默认系统提示词（来自 ComfyUI 官方 API 节点） +GEMINI_IMAGE_SYS_PROMPT = ( + "You are an expert image-generation engine. You must ALWAYS produce an image.\n" + "Interpret all user input—regardless of " + "format, intent, or abstraction—as literal visual directives for image composition.\n" + "If a prompt is conversational or lacks specific visual details, " + "you must creatively invent a concrete visual scenario that depicts the concept.\n" + "Prioritize generating the visual representation above any text, formatting, or conversational requests." +) + + +def tensor_to_base64(tensor): + """将 Tensor 转换为 base64 字符串""" + if tensor.ndim == 4: + tensor = tensor[0] + + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + buffered = io.BytesIO() + image.save(buffered, format="PNG") + return base64.b64encode(buffered.getvalue()).decode("utf-8") + + +def bytes_to_tensor(data): + """将图片字节数据转换为 Tensor""" + try: + img = Image.open(io.BytesIO(data)).convert("RGB") + image_np = np.array(img).astype(np.float32) / 255.0 + return torch.from_numpy(image_np).unsqueeze(0) + except Exception as e: + print(f"[GEMINI_AUTHTOKEN] Image decode error: {e}") + return torch.zeros((1, 512, 512, 3)) + + +class PDGeminiImageGenAuthToken: + """ + PD: Gemini Image Generation (Auth Token) + 使用 ComfyUI Auth Token 调用 Gemini Image API + """ + + @classmethod + def INPUT_TYPES(cls): + return { + "required": { + "auth_token": ("STRING", { + "default": "", + "multiline": False, + "placeholder": "Paste your Auth Token here" + }), + "prompt": ("STRING", { + "multiline": True, + "default": "A futuristic city with flying cars", + "tooltip": "Describe what you want to generate" + }), + "model": ([ + "gemini-2.5-flash-image", + "gemini-2.5-flash-image-preview" + ], { + "default": "gemini-2.5-flash-image" + }), + "aspect_ratio": ([ + "auto", "1:1", "16:9", "9:16", "4:3", "3:4", "21:9", "9:21" + ], { + "default": "auto" + }), + "resolution": (["auto", "1K", "2K", "4K"], { + "default": "1K" + }), + "seed": ("INT", { + "default": 0, + "min": 0, + "max": 0xFFFFFFFFFFFFFFFF + }), + }, + "optional": { + "image_ref": ("IMAGE", { + "tooltip": "Reference image(s) for image-to-image generation" + }), + "system_prompt": ("STRING", { + "multiline": True, + "default": GEMINI_IMAGE_SYS_PROMPT, + "tooltip": "Optional system instructions" + }), + }, + "hidden": { + "unique_id": "UNIQUE_ID", + } + } + + RETURN_TYPES = ("IMAGE", "STRING") + RETURN_NAMES = ("image", "info") + FUNCTION = "generate_image" + CATEGORY = "PD_Tools/Image_Generation" + + def generate_image(self, auth_token, prompt, model, aspect_ratio, resolution, seed, unique_id, image_ref=None, system_prompt=""): + """生成图片""" + + print(f"[GEMINI_AUTHTOKEN] ========== 开始执行 ==========") + print(f"[GEMINI_AUTHTOKEN] Prompt: {prompt[:50]}...") + print(f"[GEMINI_AUTHTOKEN] Model: {model}") + + # 验证 Token + if not auth_token or not auth_token.strip(): + error_msg = "Error: Auth Token is required" + print(f"[GEMINI_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 清理 Token + auth_token = auth_token.strip() + if auth_token.lower().startswith("bearer "): + auth_token = auth_token[7:].strip() + + # 构建 API URL (不需要添加 :generateContent 后缀，API 会自动处理) + base_url = f"https://api.comfy.org/proxy/vertexai/gemini/{model}" + + # 构建请求内容 + parts = [] + + # 添加系统提示词（如果有） + if system_prompt and system_prompt.strip(): + parts.append({"text": system_prompt.strip()}) + + # 添加用户提示词 + parts.append({"text": prompt}) + + # 添加参考图片（如果有） + if image_ref is not None: + for i in range(image_ref.shape[0]): + b64_str = tensor_to_base64(image_ref[i]) + parts.append({ + "inlineData": { + "mimeType": "image/png", + "data": b64_str + } + }) + print(f"[GEMINI_AUTHTOKEN] Added reference image {i+1}") + + # 生成配置 + generation_config = { + "responseModalities": ["IMAGE"] + } + + # 图片配置 + image_config = {} + if aspect_ratio != "auto": + image_config["aspectRatio"] = aspect_ratio + if resolution != "auto": + image_config["resolution"] = resolution + + if image_config: + generation_config["imageConfig"] = image_config + + # 完整请求体 + payload = { + "contents": [{ + "role": "user", + "parts": parts + }], + "generationConfig": generation_config + } + + # 请求头 + headers = { + "Authorization": f"Bearer {auth_token}", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + start_time = time.time() + + try: + print(f"[GEMINI_AUTHTOKEN] Sending request to {model}") + print(f"[GEMINI_AUTHTOKEN] Aspect Ratio: {aspect_ratio}, Resolution: {resolution}") + + response = requests.post(base_url, headers=headers, json=payload, timeout=120) + + # 检查响应 + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[GEMINI_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + result = response.json() + + # 提取图片 + image_tensors = [] + candidates = result.get("candidates", []) + + if not candidates: + error_msg = "Error: No candidates in response" + print(f"[GEMINI_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + for cand in candidates: + content = cand.get("content", {}) + parts = content.get("parts", []) + + for part in parts: + # 支持两种字段名格式 + inline_data = part.get("inline_data") or part.get("inlineData") + + if inline_data: + mime = inline_data.get("mime_type") or inline_data.get("mimeType") + data_b64 = inline_data.get("data") + + if mime in ["image/png", "image/jpeg"] and data_b64: + raw_data = base64.b64decode(data_b64) + image_tensors.append(bytes_to_tensor(raw_data)) + print(f"[GEMINI_AUTHTOKEN] Extracted image ({mime})") + + if not image_tensors: + error_msg = f"Error: No images found in response" + print(f"[GEMINI_AUTHTOKEN] {error_msg}") + print(f"[GEMINI_AUTHTOKEN] Response: {str(result)[:500]}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + + # 计算耗时和成本 + end_time = time.time() + duration = end_time - start_time + + # 估算成本（根据模型） + cost_per_img = 0.04 + if "pro" in model.lower(): + cost_per_img = 0.08 + total_cost = cost_per_img * len(image_tensors) + + info_str = ( + f"Model: {model}\n" + f"Status: Success\n" + f"Time: {duration:.2f}s\n" + f"Images: {len(image_tensors)}\n" + f"Aspect Ratio: {aspect_ratio}\n" + f"Resolution: {resolution}\n" + f"Est. Cost: ${total_cost:.4f}" + ) + + print(f"[GEMINI_AUTHTOKEN] Success! Generated {len(image_tensors)} images in {duration:.2f}s") + return (final_image, info_str) + + except requests.exceptions.Timeout: + error_msg = "Error: Request timeout (120s)" + print(f"[GEMINI_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except requests.exceptions.ConnectionError as e: + error_msg = f"Error: Connection failed - {str(e)}" + print(f"[GEMINI_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except Exception as e: + error_msg = f"Error: {str(e)}" + print(f"[GEMINI_AUTHTOKEN] {error_msg}") + import traceback + traceback.print_exc() + return (torch.zeros((1, 512, 512, 3)), error_msg) + + +# 节点注册 +NODE_CLASS_MAPPINGS = { + "PDGeminiImageGenAuthToken": PDGeminiImageGenAuthToken +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDGeminiImageGenAuthToken": "PD: Gemini Image (comfyui_AuthToken)" +} diff --git a/py/IMGE_GeminiPro_apikey.py b/py/IMGE_GeminiPro_apikey.py new file mode 100644 index 0000000..0f4e955 --- /dev/null +++ b/py/IMGE_GeminiPro_apikey.py @@ -0,0 +1,294 @@ +""" +PD: Gemini Pro Image Generation Node (ComfyUI API Key) +Nano Banana Pro - 使用 ComfyUI API Key 调用 ComfyUI 代理 API +支持高分辨率图像生成（1K/2K/4K） +""" + +import torch +import requests +import time +import io +import base64 +import numpy as np +from PIL import Image + +# 默认系统提示词（来自 ComfyUI 官方 API 节点） +GEMINI_IMAGE_SYS_PROMPT = ( + "You are an expert image-generation engine. You must ALWAYS produce an image.\n" + "Interpret all user input—regardless of " + "format, intent, or abstraction—as literal visual directives for image composition.\n" + "If a prompt is conversational or lacks specific visual details, " + "you must creatively invent a concrete visual scenario that depicts the concept.\n" + "Prioritize generating the visual representation above any text, formatting, or conversational requests." +) + + +def tensor_to_base64(tensor): + """将 Tensor 转换为 base64 字符串""" + if tensor.ndim == 4: + tensor = tensor[0] + + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + buffered = io.BytesIO() + image.save(buffered, format="PNG") + return base64.b64encode(buffered.getvalue()).decode("utf-8") + + +def bytes_to_tensor(data): + """将图片字节数据转换为 Tensor""" + try: + img = Image.open(io.BytesIO(data)).convert("RGB") + image_np = np.array(img).astype(np.float32) / 255.0 + return torch.from_numpy(image_np).unsqueeze(0) + except Exception as e: + print(f"[GEMINI_PRO_APIKEY] Image decode error: {e}") + return torch.zeros((1, 512, 512, 3)) + + +class PDGeminiProImageGenAPIKey: + """ + PD: Gemini Pro Image Generation (API Key) + Nano Banana Pro - 使用 ComfyUI API Key 调用 ComfyUI 代理 API + """ + + @classmethod + def INPUT_TYPES(cls): + return { + "required": { + "api_key": ("STRING", { + "default": "", + "multiline": False, + "placeholder": "Your ComfyUI API Key" + }), + "prompt": ("STRING", { + "multiline": True, + "default": "A futuristic city with flying cars", + "tooltip": "Text prompt describing the image to generate or the edits to apply" + }), + "model": (["gemini-3-pro-image-preview"], { + "default": "gemini-3-pro-image-preview" + }), + "aspect_ratio": ([ + "auto", "1:1", "2:3", "3:2", "3:4", "4:3", "4:5", "5:4", "9:16", "16:9", "21:9" + ], { + "default": "auto", + "tooltip": "If set to 'auto', matches your input image's aspect ratio" + }), + "resolution": (["1K", "2K", "4K"], { + "default": "1K", + "tooltip": "Target output resolution. For 2K/4K the native Gemini upscaler is used" + }), + "response_modalities": (["IMAGE+TEXT", "IMAGE"], { + "default": "IMAGE+TEXT", + "tooltip": "Choose 'IMAGE' for image-only output, or 'IMAGE+TEXT' for both" + }), + "seed": ("INT", { + "default": 42, + "min": 0, + "max": 0xFFFFFFFFFFFFFFFF + }), + }, + "optional": { + "images": ("IMAGE", { + "tooltip": "Optional reference image(s) (up to 14)" + }), + "system_prompt": ("STRING", { + "multiline": True, + "default": GEMINI_IMAGE_SYS_PROMPT, + "tooltip": "Foundational instructions that dictate an AI's behavior" + }), + }, + "hidden": { + "unique_id": "UNIQUE_ID", + } + } + + RETURN_TYPES = ("IMAGE", "STRING") + RETURN_NAMES = ("image", "text") + FUNCTION = "generate_image" + CATEGORY = "PD_Tools/Image_Generation" + + def generate_image(self, api_key, prompt, model, aspect_ratio, resolution, response_modalities, seed, unique_id, images=None, system_prompt=""): + """生成图片""" + + # 验证 API Key + if not api_key or not api_key.strip(): + error_msg = "Error: ComfyUI API Key is required" + print(f"[GEMINI_PRO_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 清理 API Key + api_key = api_key.strip() + + # ComfyUI 代理 API URL + base_url = f"https://api.comfy.org/proxy/vertexai/gemini/{model}" + + # 构建请求内容 + parts = [] + + # 添加用户提示词 + parts.append({"text": prompt}) + + # 添加参考图片（如果有，最多14张） + if images is not None: + num_images = min(images.shape[0], 14) + for i in range(num_images): + b64_str = tensor_to_base64(images[i]) + parts.append({ + "inlineData": { + "mimeType": "image/png", + "data": b64_str + } + }) + print(f"[GEMINI_PRO_APIKEY] Added reference image {i+1}/{num_images}") + + if images.shape[0] > 14: + print(f"[GEMINI_PRO_APIKEY] Warning: Only first 14 images will be used (provided {images.shape[0]})") + + # 生成配置 + generation_config = { + "responseModalities": ["IMAGE"] if response_modalities == "IMAGE" else ["TEXT", "IMAGE"] + } + + # 图片配置 + image_config = { + "imageSize": resolution + } + if aspect_ratio != "auto": + image_config["aspectRatio"] = aspect_ratio + + generation_config["imageConfig"] = image_config + + # 系统指令 + system_instruction = None + if system_prompt and system_prompt.strip(): + system_instruction = { + "parts": [{"text": system_prompt.strip()}] + } + + # 完整请求体 + payload = { + "contents": [{ + "role": "user", + "parts": parts + }], + "generationConfig": generation_config + } + + if system_instruction: + payload["systemInstruction"] = system_instruction + + # 请求头 + headers = { + "X-API-KEY": api_key, + "Accept": "application/json", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + start_time = time.time() + + try: + print(f"[GEMINI_PRO_APIKEY] Sending request to {model}") + print(f"[GEMINI_PRO_APIKEY] Aspect Ratio: {aspect_ratio}, Resolution: {resolution}") + + response = requests.post(base_url, headers=headers, json=payload, timeout=180) + + # 检查响应 + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[GEMINI_PRO_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + result = response.json() + + # 提取图片和文本 + image_tensors = [] + text_parts = [] + candidates = result.get("candidates", []) + + if not candidates: + error_msg = "Error: No candidates in response" + print(f"[GEMINI_PRO_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + for cand in candidates: + content = cand.get("content", {}) + parts = content.get("parts", []) + + for part in parts: + # 提取图片 + inline_data = part.get("inline_data") or part.get("inlineData") + + if inline_data: + mime = inline_data.get("mime_type") or inline_data.get("mimeType") + data_b64 = inline_data.get("data") + + if mime in ["image/png", "image/jpeg"] and data_b64: + raw_data = base64.b64decode(data_b64) + image_tensors.append(bytes_to_tensor(raw_data)) + print(f"[GEMINI_PRO_APIKEY] Extracted image ({mime})") + + # 提取文本 + if "text" in part and part["text"]: + text_parts.append(part["text"]) + + if not image_tensors: + error_msg = f"Error: No images found in response" + print(f"[GEMINI_PRO_APIKEY] {error_msg}") + print(f"[GEMINI_PRO_APIKEY] Response: {str(result)[:500]}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + final_text = "\n".join(text_parts) if text_parts else "No text output" + + # 计算耗时和成本 + end_time = time.time() + duration = end_time - start_time + + # 估算成本（Gemini 3 Pro Image） + cost_per_img = 0.12 # 估算 + total_cost = cost_per_img * len(image_tensors) + + info_str = ( + f"Model: {model}\n" + f"Status: Success\n" + f"Time: {duration:.2f}s\n" + f"Images: {len(image_tensors)}\n" + f"Aspect Ratio: {aspect_ratio}\n" + f"Resolution: {resolution}\n" + f"Est. Cost: ${total_cost:.4f}\n" + f"---\n" + f"{final_text}" + ) + + print(f"[GEMINI_PRO_APIKEY] Success! Generated {len(image_tensors)} images in {duration:.2f}s") + return (final_image, info_str) + + except requests.exceptions.Timeout: + error_msg = "Error: Request timeout (180s)" + print(f"[GEMINI_PRO_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except requests.exceptions.ConnectionError as e: + error_msg = f"Error: Connection failed - {str(e)}" + print(f"[GEMINI_PRO_APIKEY] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except Exception as e: + error_msg = f"Error: {str(e)}" + print(f"[GEMINI_PRO_APIKEY] {error_msg}") + import traceback + traceback.print_exc() + return (torch.zeros((1, 512, 512, 3)), error_msg) + + +# 节点注册 +NODE_CLASS_MAPPINGS = { + "PDGeminiProImageGenAPIKey": PDGeminiProImageGenAPIKey +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDGeminiProImageGenAPIKey": "PD: Gemini Pro Image (apikey)" +} diff --git a/py/IMGE_GeminiPro_authtoken.py b/py/IMGE_GeminiPro_authtoken.py new file mode 100644 index 0000000..2a606f2 --- /dev/null +++ b/py/IMGE_GeminiPro_authtoken.py @@ -0,0 +1,300 @@ +""" +PD: Gemini Pro Image Generation Node (ComfyUI Auth Token) +Nano Banana Pro - 基于 Gemini 3 Pro Image +支持高分辨率图像生成（1K/2K/4K） +""" + +import torch +import requests +import time +import io +import base64 +import numpy as np +from PIL import Image + +# 默认系统提示词（来自 ComfyUI 官方 API 节点） +GEMINI_IMAGE_SYS_PROMPT = ( + "You are an expert image-generation engine. You must ALWAYS produce an image.\n" + "Interpret all user input—regardless of " + "format, intent, or abstraction—as literal visual directives for image composition.\n" + "If a prompt is conversational or lacks specific visual details, " + "you must creatively invent a concrete visual scenario that depicts the concept.\n" + "Prioritize generating the visual representation above any text, formatting, or conversational requests." +) + + +def tensor_to_base64(tensor): + """将 Tensor 转换为 base64 字符串""" + if tensor.ndim == 4: + tensor = tensor[0] + + array = np.clip(255. * tensor.cpu().numpy(), 0, 255).astype(np.uint8) + image = Image.fromarray(array) + buffered = io.BytesIO() + image.save(buffered, format="PNG") + return base64.b64encode(buffered.getvalue()).decode("utf-8") + + +def bytes_to_tensor(data): + """将图片字节数据转换为 Tensor""" + try: + img = Image.open(io.BytesIO(data)).convert("RGB") + image_np = np.array(img).astype(np.float32) / 255.0 + return torch.from_numpy(image_np).unsqueeze(0) + except Exception as e: + print(f"[GEMINI_PRO_AUTHTOKEN] Image decode error: {e}") + return torch.zeros((1, 512, 512, 3)) + + +class PDGeminiProImageGenAuthToken: + """ + PD: Gemini Pro Image Generation (Auth Token) + Nano Banana Pro - 使用 ComfyUI Auth Token 调用 Gemini 3 Pro Image API + """ + + @classmethod + def INPUT_TYPES(cls): + return { + "required": { + "auth_token": ("STRING", { + "default": "", + "multiline": False, + "placeholder": "Paste your Auth Token here" + }), + "prompt": ("STRING", { + "multiline": True, + "default": "A futuristic city with flying cars", + "tooltip": "Text prompt describing the image to generate or the edits to apply" + }), + "model": (["gemini-3-pro-image-preview"], { + "default": "gemini-3-pro-image-preview" + }), + "aspect_ratio": ([ + "auto", "1:1", "2:3", "3:2", "3:4", "4:3", "4:5", "5:4", "9:16", "16:9", "21:9" + ], { + "default": "auto", + "tooltip": "If set to 'auto', matches your input image's aspect ratio" + }), + "resolution": (["1K", "2K", "4K"], { + "default": "1K", + "tooltip": "Target output resolution. For 2K/4K the native Gemini upscaler is used" + }), + "response_modalities": (["IMAGE+TEXT", "IMAGE"], { + "default": "IMAGE+TEXT", + "tooltip": "Choose 'IMAGE' for image-only output, or 'IMAGE+TEXT' for both" + }), + "seed": ("INT", { + "default": 42, + "min": 0, + "max": 0xFFFFFFFFFFFFFFFF + }), + }, + "optional": { + "images": ("IMAGE", { + "tooltip": "Optional reference image(s) (up to 14)" + }), + "system_prompt": ("STRING", { + "multiline": True, + "default": GEMINI_IMAGE_SYS_PROMPT, + "tooltip": "Foundational instructions that dictate an AI's behavior" + }), + }, + "hidden": { + "unique_id": "UNIQUE_ID", + } + } + + RETURN_TYPES = ("IMAGE", "STRING") + RETURN_NAMES = ("image", "text") + FUNCTION = "generate_image" + CATEGORY = "PD_Tools/Image_Generation" + + def generate_image(self, auth_token, prompt, model, aspect_ratio, resolution, response_modalities, seed, unique_id, images=None, system_prompt=""): + """生成图片""" + + print(f"[GEMINI_PRO_AUTHTOKEN] ========== 开始执行 ==========") + print(f"[GEMINI_PRO_AUTHTOKEN] Prompt: {prompt[:50]}...") + print(f"[GEMINI_PRO_AUTHTOKEN] Model: {model}") + + # 验证 Token + if not auth_token or not auth_token.strip(): + error_msg = "Error: Auth Token is required" + print(f"[GEMINI_PRO_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + # 清理 Token + auth_token = auth_token.strip() + if auth_token.lower().startswith("bearer "): + auth_token = auth_token[7:].strip() + + # 构建 API URL + base_url = f"https://api.comfy.org/proxy/vertexai/gemini/{model}" + + # 构建请求内容 + parts = [] + + # 添加用户提示词 + parts.append({"text": prompt}) + + # 添加参考图片（如果有，最多14张） + if images is not None: + num_images = min(images.shape[0], 14) + for i in range(num_images): + b64_str = tensor_to_base64(images[i]) + parts.append({ + "inlineData": { + "mimeType": "image/png", + "data": b64_str + } + }) + print(f"[GEMINI_PRO_AUTHTOKEN] Added reference image {i+1}/{num_images}") + + if images.shape[0] > 14: + print(f"[GEMINI_PRO_AUTHTOKEN] Warning: Only first 14 images will be used (provided {images.shape[0]})") + + # 生成配置 + generation_config = { + "responseModalities": ["IMAGE"] if response_modalities == "IMAGE" else ["TEXT", "IMAGE"] + } + + # 图片配置 + image_config = { + "imageSize": resolution + } + if aspect_ratio != "auto": + image_config["aspectRatio"] = aspect_ratio + + generation_config["imageConfig"] = image_config + + # 系统指令 + system_instruction = None + if system_prompt and system_prompt.strip(): + system_instruction = { + "parts": [{"text": system_prompt.strip()}] + } + + # 完整请求体 + payload = { + "contents": [{ + "role": "user", + "parts": parts + }], + "generationConfig": generation_config + } + + if system_instruction: + payload["systemInstruction"] = system_instruction + + # 请求头 + headers = { + "Authorization": f"Bearer {auth_token}", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.0" + } + + start_time = time.time() + + try: + print(f"[GEMINI_PRO_AUTHTOKEN] Sending request to {model}") + print(f"[GEMINI_PRO_AUTHTOKEN] Aspect Ratio: {aspect_ratio}, Resolution: {resolution}") + + response = requests.post(base_url, headers=headers, json=payload, timeout=180) + + # 检查响应 + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[GEMINI_PRO_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + result = response.json() + + # 提取图片和文本 + image_tensors = [] + text_parts = [] + candidates = result.get("candidates", []) + + if not candidates: + error_msg = "Error: No candidates in response" + print(f"[GEMINI_PRO_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + for cand in candidates: + content = cand.get("content", {}) + parts = content.get("parts", []) + + for part in parts: + # 提取图片 + inline_data = part.get("inline_data") or part.get("inlineData") + + if inline_data: + mime = inline_data.get("mime_type") or inline_data.get("mimeType") + data_b64 = inline_data.get("data") + + if mime in ["image/png", "image/jpeg"] and data_b64: + raw_data = base64.b64decode(data_b64) + image_tensors.append(bytes_to_tensor(raw_data)) + print(f"[GEMINI_PRO_AUTHTOKEN] Extracted image ({mime})") + + # 提取文本 + if "text" in part and part["text"]: + text_parts.append(part["text"]) + + if not image_tensors: + error_msg = f"Error: No images found in response" + print(f"[GEMINI_PRO_AUTHTOKEN] {error_msg}") + print(f"[GEMINI_PRO_AUTHTOKEN] Response: {str(result)[:500]}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + final_text = "\n".join(text_parts) if text_parts else "No text output" + + # 计算耗时和成本 + end_time = time.time() + duration = end_time - start_time + + # 估算成本（Gemini 3 Pro Image） + # $2/1M input tokens, $12/1M output text tokens, $120/1M output image tokens + cost_per_img = 0.12 # 估算 + total_cost = cost_per_img * len(image_tensors) + + info_str = ( + f"Model: {model}\n" + f"Status: Success\n" + f"Time: {duration:.2f}s\n" + f"Images: {len(image_tensors)}\n" + f"Aspect Ratio: {aspect_ratio}\n" + f"Resolution: {resolution}\n" + f"Est. Cost: ${total_cost:.4f}\n" + f"---\n" + f"{final_text}" + ) + + print(f"[GEMINI_PRO_AUTHTOKEN] Success! Generated {len(image_tensors)} images in {duration:.2f}s") + return (final_image, info_str) + + except requests.exceptions.Timeout: + error_msg = "Error: Request timeout (180s)" + print(f"[GEMINI_PRO_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except requests.exceptions.ConnectionError as e: + error_msg = f"Error: Connection failed - {str(e)}" + print(f"[GEMINI_PRO_AUTHTOKEN] {error_msg}") + return (torch.zeros((1, 512, 512, 3)), error_msg) + + except Exception as e: + error_msg = f"Error: {str(e)}" + print(f"[GEMINI_PRO_AUTHTOKEN] {error_msg}") + import traceback + traceback.print_exc() + return (torch.zeros((1, 512, 512, 3)), error_msg) + + +# 节点注册 +NODE_CLASS_MAPPINGS = { + "PDGeminiProImageGenAuthToken": PDGeminiProImageGenAuthToken +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDGeminiProImageGenAuthToken": "PD: Gemini Pro Image (comfyui_AuthToken)" +} diff --git a/py/PD_comfyplus_image.py b/py/PD_comfyplus_image.py new file mode 100644 index 0000000..16b5138 --- /dev/null +++ b/py/PD_comfyplus_image.py @@ -0,0 +1,78 @@ +import torch +from comfy_api_nodes.apis import GeminiPart, GeminiInlineData, GeminiMimeType +from comfy_api_nodes.util.conversions import tensor_to_base64_string + +class PD_ComfyPlusImage: + """ + 将最多9个尺寸不一的图片按顺序整合打包为 GEMINI_INPUT_FILES， + 可以直接连接给支持 files 的大模型节点 (如 Nano Banana Pro 等)， + 从而完美突破 ComfyUI Batch 必须相同尺寸的限制。 + """ + @classmethod + def INPUT_TYPES(cls): + return { + # 第1张图片强制必填 + "required": { + "image1": ("IMAGE", {"tooltip": "必须输入的第1张图片"}), + }, + # 第 2-9 张及 previous_files 作为可选填 + "optional": { + "image2": ("IMAGE", {"tooltip": "可选输入的第2张图片"}), + "image3": ("IMAGE", {"tooltip": "可选输入的第3张图片"}), + "image4": ("IMAGE", {"tooltip": "可选输入的第4张图片"}), + "image5": ("IMAGE", {"tooltip": "可选输入的第5张图片"}), + "image6": ("IMAGE", {"tooltip": "可选输入的第6张图片"}), + "image7": ("IMAGE", {"tooltip": "可选输入的第7张图片"}), + "image8": ("IMAGE", {"tooltip": "可选输入的第8张图片"}), + "image9": ("IMAGE", {"tooltip": "可选输入的第9张图片"}), + "previous_files": ("GEMINI_INPUT_FILES", {"tooltip": "如超过9张图需继续打包，可连接上一个该节点输出的files"}), + } + } + + RETURN_TYPES = ("GEMINI_INPUT_FILES",) + RETURN_NAMES = ("files",) + FUNCTION = "pack_images" + CATEGORY = "PD_Tools" + + def pack_images(self, image1, image2=None, image3=None, image4=None, + image5=None, image6=None, image7=None, image8=None, + image9=None, previous_files=None): + + files_list = [] + + # 如果需要无缝串联别的节点输出的 files ，优先加进去 + if previous_files is not None: + files_list.extend(previous_files) + + images = [image1, image2, image3, image4, image5, image6, image7, image8, image9] + + for img in images: + if img is not None: + # 兼容万一单条线里输入了 Batch(N) 图片的特殊情况，分别解开处理 + for i in range(img.shape[0]): + try: + # 转换成 Base64 + img_b64 = tensor_to_base64_string(img[i].unsqueeze(0)) + # 构造为 Gemini 认证的 Part + new_part = GeminiPart( + inlineData=GeminiInlineData( + mimeType=GeminiMimeType.image_png, + data=img_b64 + ) + ) + # 转成字典防止 Pydantic 跨模块传递对象时校验失败 + files_list.append(new_part.model_dump(exclude_none=True)) + except Exception as e: + print(f"[Error] PD_comfyplus_image pack failed for an image: {e}") + + return (files_list,) + + +# 注册节点 +NODE_CLASS_MAPPINGS = { + "PD_comfyplus_image": PD_ComfyPlusImage +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PD_comfyplus_image": "PD_comfyplus_image (Multi-Size Pack)" +} diff --git a/py/README.md b/py/README.md new file mode 100644 index 0000000..f6382c1 --- /dev/null +++ b/py/README.md @@ -0,0 +1,352 @@ +# PD ComfyUI API Nodes + +ComfyUI 自定义节点集合，用于调用各种 AI 图像生成 API（GPT Image、Gemini Image、Flux Kontext）。 + +## 📦 节点列表 + +### ComfyUI AuthToken 版本（通过 ComfyUI 代理） +ComfyUI AuthToken 版本（通过 ComfyUI 代理） + - PD: GPT Image (comfyui_AuthToken) + - PD: Gemini Image (comfyui_AuthToken) + - PD: Flux Kontext (comfyui_AuthToken) + +### 直接 API Key 版本直接调用官方comfyui的key + +API Key 版本： +- 只保留 ComfyUI API Key 输入 +- 参数和调用逻辑跟 ComfyUI 官方自带的 API 节点一样 - 只是在调用时可以手动填入 key + +| 节点名称 | 文件 | 说明 | +|---------|------|------| +| **PD: GPT Image (apikey)** | `IMAGE_GPT_apikey.py` | OpenAI GPT Image 生成/编辑 | +| **PD: Gemini Image (apikey)** | `IMGE_GeminiNode_apikey.py` | Google Gemini 图像生成 | +| **PD: Gemini Pro Image (apikey)** | `IMGE_GeminiPro_apikey.py` | Google Gemini 3 Pro 高分辨率图像生成 | +| **PD: Flux Kontext (apikey)** | `IMGE_kontext_apikey.py` | Flux Kontext 图像编辑 | + +--- + +## 🔑 两种版本的区别 + +### ComfyUI AuthToken 版本 (comfyui_AuthToken) + +**认证方式**：使用 ComfyUI Auth Token，pandy自制油猴插件获取。 + +**认证头格式**：`Authorization: Bearer {auth_token}` + +**使用示例**： +``` +auth_token: eyJhbGciOiJSUzI1NiIsImtpZCI6ImEzOGVhNmEw... +``` + +### API Key 版本 (apikey) + +**认证方式**：使用 ComfyUI 官方 API Key + +**认证头格式**：`X-API-KEY: {api_key}` （与 ComfyUI 官方 API 节点完全一致） + +**使用示例**： +``` +api_key: comfy_xxxxxxxxxxxxxxxxxxxxx +``` + +**特点**： +- 参数和调用逻辑跟 ComfyUI 官方自带的 API 节点一样 +- 只是在调用时可以手动填入 key +- 直接使用官方账号的 key 申请下来填入使用 + + +## 📖 节点详细说明 + +### 1. GPT Image 节点 + +**功能**： +- 文生图（Text-to-Image） +- 图片编辑（Image Editing） +- 图片修复（Inpainting with Mask） + +**支持模型**： + +**AuthToken 版本**： +- `gpt-image-1`：OpenAI GPT Image 1（推荐） +- `gpt-image-1.5`：OpenAI GPT Image 1.5（更高质量） + + +### 3. Gemini Pro Image 节点（Nano Banana Pro） + +**功能**： +- 高分辨率图像生成（1K/2K/4K） +- 文生图（Text-to-Image） +- 图生图（Image-to-Image） +- 支持多张参考图片（最多14张） +- 同时输出图像和文本说明 + +**支持模型**： +- `gemini-3-pro-image-preview`：Gemini 3 Pro Image（最高质量） + +**参数说明**： +- `prompt`：文本提示词，描述要生成的图像或编辑指令 +- `model`：gemini-3-pro-image-preview +- `aspect_ratio`：宽高比 + - auto：自动匹配输入图片比例，无图片时默认 16:9 + - 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 +- `resolution`：目标输出分辨率 + - 1K：标准分辨率 + - 2K：使用原生 Gemini 放大器 + - 4K：使用原生 Gemini 放大器（最高质量） +- `response_modalities`：输出模式 + - IMAGE+TEXT：同时返回图像和文本说明 + - IMAGE：仅返回图像 +- `images`（可选）：参考图片（最多14张，使用 Batch Images 节点） +- `system_prompt`（可选）：系统提示词，默认使用官方优化的图像生成提示词 + +**输出**： +- `image`：生成的图像 +- `text`：文本说明（包含模型的思考过程和描述） + +**成本估算**： +- Gemini 3 Pro Image: 约 $0.12/张 + - 基于 token 使用量：$2/1M input + $12/1M output text + $120/1M output image + +**特殊说明（apikey 版本）**： +- 需要提供 `project_id`（Google Cloud 项目 ID） +- 需要提供 `location`（区域，默认 us-central1） +- 超时时间：180秒（因为高分辨率生成需要更长时间） + +**使用建议**： +- 适合需要高质量、高分辨率图像的场景 +- 2K/4K 分辨率会使用原生放大器，质量更好但成本更高 +- IMAGE+TEXT 模式可以看到模型的思考过程，有助于理解生成结果 +- 可以提供多张参考图片来引导生成风格 + +--- + +### 4. Flux Kontext 节点 + +**功能**： +- 图片编辑（Image Editing） +- 基于上下文的智能修改 + +**支持模型**： +- `flux-kontext-pro`：快速版本 +- `flux-kontext-max`：高质量版本 + +**参数说明**： +- `prompt`：编辑指令 +- `aspect_ratio`：宽高比 +- `guidance`：提示词遵循度（0.1-99.0） +- `steps`：生成步数（1-150） +- `prompt_upsampling`：AI 增强提示词 +- `input_image`（可选）：输入图片 + +**成本估算**： +- Kontext Pro: $0.05/张 +- Kontext Max: $0.10/张 + +**特点**： +- 异步生成，需要轮询结果 +- 最长等待时间：10 分钟 +- 轮询间隔：2 秒 + +--- + +## 🚀 使用建议 + +### 选择 ComfyUI AuthToken 版本，如果你： +- ✅ 已经有 ComfyUI Auth Token +- ✅ 想要统一管理多个 AI 模型 +- ✅ 喜欢通过 ComfyUI 生态系统 +- ✅ 不想配置多个平台的 API Key + +### 选择直接 API Key 版本，如果你： +- ✅ 已经有各个平台的 API Key +- ✅ 想要直接控制 API 调用 +- ✅ 不想依赖 ComfyUI 代理服务 +- ✅ 需要更灵活的配置（如 Gemini 的 project_id） + +--- + +## 📝 使用示例 + +### 示例 1：使用 ComfyUI AuthToken 生成图片（GPT Image） + +``` +节点：PD: GPT Image (comfyui_AuthToken) + +参数： +- auth_token: eyJhbGciOiJSUzI1NiIsImtpZCI6ImEzOGVhNmEw... +- prompt: A futuristic city with flying cars at sunset +- model: gpt-image-1 +- quality: medium +- background: auto +- size: 1024x1024 +- n: 1 +``` + +### 示例 2：使用直接 API Key 生成图片（DALL-E） + +``` +节点：PD: GPT Image (apikey) + +参数： +- api_key: sk-proj-abc123... +- prompt: A futuristic city with flying cars at sunset +- model: dall-e-3 +- quality: hd +- size: 1024x1792 +``` + +### 示例 3：使用 Gemini 生成图片（apikey 版本） + +``` +节点：PD: Gemini Image (apikey) + +参数： +- api_key: AIzaSyD... +- project_id: my-project-123 +- location: us-central1 +- prompt: A beautiful landscape with mountains +- model: gemini-2.5-flash-image +- aspect_ratio: 16:9 +- resolution: 2K +``` + +### 示例 4：使用 Flux Kontext 编辑图片 + +``` +节点：PD: Flux Kontext (comfyui_AuthToken) + +参数： +- auth_token: eyJhbGciOiJSUzI1NiIsImtpZCI6ImEzOGVhNmEw... +- prompt: Change the sky to sunset colors +- model: flux-kontext-pro +- aspect_ratio: 16:9 +- guidance: 3.0 +- steps: 50 +- input_image: [连接输入图片] +``` + +--- + +## ⚠️ 注意事项 + +1. **API Key 安全**： + - 不要在公开的工作流中暴露 API Key + - 建议使用环境变量或配置文件管理 + +2. **成本控制**： + - 所有节点都会显示预估成本 + - 实际成本以各平台账单为准 + - 建议设置 API 使用限额 + +3. **超时设置**： + - GPT/Gemini: 120 秒超时 + - Flux Kontext: 10 分钟超时（包含轮询） + +4. **错误处理**： + - 所有节点都有详细的错误日志 + - 检查 ComfyUI 控制台查看详细信息 + +5. **网络要求**： + - 需要稳定的网络连接 + - 某些地区可能需要代理访问 + +--- + +## 🔧 故障排除 + +### 问题 1：认证失败（401 错误） + +**ComfyUI AuthToken 版本**： +- 检查 Auth Token 是否正确 +- Token 可能已过期，重新获取 +- 认证头格式：`Authorization: Bearer {auth_token}` + +**直接 API Key 版本**： +- 检查 API Key 格式是否正确 +- 确认 API Key 有效且有余额 +- 认证头格式：`X-API-KEY: {api_key}` （与 ComfyUI 官方 API 节点一致） +- Gemini 需要确认 Project ID 和 Location 正确 + +### 问题 2：请求超时 + +- 检查网络连接 +- 尝试增加超时时间 +- Flux Kontext 可能需要更长时间，属于正常现象 + +### 问题 3：图片无法生成 / 返回黑图 + +**症状**：节点返回 512x512 的黑色图片 + +**可能原因**： +1. **API 端点错误**（已修复）： + - 旧版本的 Gemini AuthToken 节点使用了错误的端点格式 + - 错误：`https://api.comfy.org/proxy/vertexai/gemini/{model}:generateContent` + - 正确：`https://api.comfy.org/proxy/vertexai/gemini/{model}` + - **解决方案**：更新到最新版本（v2.1+） + +2. **模型名称不支持**： + - 检查控制台是否有 "Unsupported model" 错误 + - 确认使用的模型在支持列表中 + - Gemini AuthToken 版本支持：`gemini-2.5-flash-image`, `gemini-2.5-flash-image-preview` + +3. **提示词违规**： + - 检查提示词是否符合内容政策 + - 查看控制台日志获取详细错误信息 + +4. **API 配额不足**： + - 确认账户有足够的配额/余额 + - 检查 API 使用限制 + +**调试步骤**： +1. 打开 ComfyUI 控制台 +2. 查找 `[GEMINI_AUTHTOKEN]` 或 `[GEMINI_APIKEY]` 开头的日志 +3. 检查是否有 API Error 信息 +4. 根据错误信息采取相应措施 + +### 问题 4：认证失败（Gemini 节点） + +- 确认已启用 Vertex AI API +- 确认 Project ID 正确 +- 确认 Location 支持 Gemini 模型 +- 检查 API Key 权限 + +--- + +## 📚 相关链接 + +- [ComfyUI 官网](https://www.comfy.org) +- [OpenAI Platform](https://platform.openai.com) +- [Google Cloud Console](https://console.cloud.google.com) +- [Black Forest Labs](https://blackforestlabs.ai) +- [ComfyUI GitHub](https://github.com/comfyanonymous/ComfyUI) + +--- + +## 📄 许可证 + +本项目遵循 MIT 许可证。 + +--- + +## 🤝 贡献 + +欢迎提交 Issue 和 Pull Request！ + +--- + +## 📧 联系方式 + +如有问题或建议，请通过 GitHub Issues 联系。 + +--- + +**最后更新**：2025-01-12 +**版本**：2.5 + +**更新日志**： +- 2025-01-12 v2.5: 修复 API Key 版本认证头问题（401 错误），使用 `X-API-KEY` 与官方一致 +- 2025-01-12 v2.4: 新增 Gemini Pro Image (Nano Banana Pro) 节点，支持高分辨率生成 +- 2025-01-12 v2.3: 更新 GPT Image AuthToken 节点参数，匹配官方 ComfyUI API 节点 +- 2025-01-12 v2.2: 添加官方 ComfyUI API 节点的默认 system prompt +- 2025-01-12 v2.1: 修复 Gemini AuthToken 版本黑图问题，更新模型列表 +- 2025-01-11 v2.0: 初始版本，支持 AuthToken 和 API Key 两种认证方式 diff --git a/temperay/PD_comfyui_token_grabber.user.js b/temperay/PD_comfyui_token_grabber.user.js new file mode 100644 index 0000000..817242c --- /dev/null +++ b/temperay/PD_comfyui_token_grabber.user.js @@ -0,0 +1,506 @@ +// ==UserScript== +// @name Pandy-ComfyUI Token Grabber +// @namespace http://tampermonkey.net/ +// @version 3.5 +// @description 直接从 ComfyUI 提取 auth_token_comfy_org 并同步到 Pandy +// @author You +// @match *://localhost:8188/* +// @match *://127.0.0.1:8188/* +// @match *://127.0.0.1:8180/* +// @grant GM_setValue +// @grant GM_getValue +// @grant GM_setClipboard +// @grant GM_xmlhttpRequest +// @grant unsafeWindow +// @connect 127.0.0.1 +// @connect localhost +// @run-at document-idle +// ==/UserScript== + +(function () { + 'use strict'; + + const STORAGE_KEY = 'comfyui_auth_token'; + const POS_KEY = 'token_grabber_pos'; + const COLLAPSED_KEY = 'token_grabber_collapsed'; + + let foundToken = ''; + let isCollapsed = GM_getValue(COLLAPSED_KEY, false); + let savedPos = GM_getValue(POS_KEY, { x: null, y: 10 }); + + // 确保位置在可视范围内 + function clampPosition() { + const panel = document.getElementById('token-grabber-panel'); + if (!panel) return; + + const rect = panel.getBoundingClientRect(); + let needUpdate = false; + + // 如果面板超出右边界 + if (rect.right > window.innerWidth) { + panel.style.left = 'auto'; + panel.style.right = '10px'; + needUpdate = true; + } + // 如果面板超出下边界 + if (rect.bottom > window.innerHeight) { + panel.style.top = Math.max(10, window.innerHeight - rect.height - 10) + 'px'; + needUpdate = true; + } + // 如果面板超出左边界 + if (rect.left < 0) { + panel.style.left = '10px'; + panel.style.right = 'auto'; + needUpdate = true; + } + // 如果面板超出上边界 + if (rect.top < 0) { + panel.style.top = '10px'; + needUpdate = true; + } + } + + // 搜索 token + function searchToken() { + // 1. localStorage + for (let i = 0; i < localStorage.length; i++) { + const key = localStorage.key(i); + const value = localStorage.getItem(key); + if (value && value.includes('eyJ')) { + const token = extractJWT(value); + if (token && isValidJWT(token)) return token; + } + } + // 2. sessionStorage + for (let i = 0; i < sessionStorage.length; i++) { + const key = sessionStorage.key(i); + const value = sessionStorage.getItem(key); + if (value && value.includes('eyJ')) { + const token = extractJWT(value); + if (token && isValidJWT(token)) return token; + } + } + // 3. window 对象 + const props = ['app', 'api', 'comfyAPI']; + for (const prop of props) { + if (window[prop]) { + const token = deepSearch(window[prop], 'auth_token_comfy_org', 5); + if (token) return token; + } + } + return null; + } + + function extractJWT(str) { + const match = str.match(/eyJ[A-Za-z0-9_-]+\.eyJ[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+/); + return match ? match[0] : null; + } + + function isValidJWT(token) { + try { + const parts = token.split('.'); + if (parts.length !== 3) return false; + let payload = parts[1]; + payload += '='.repeat((4 - payload.length % 4) % 4); + const info = JSON.parse(atob(payload)); + return info.exp !== undefined; + } catch (e) { return false; } + } + + function deepSearch(obj, key, depth) { + if (depth <= 0 || !obj || typeof obj !== 'object') return null; + try { + if (obj[key]) return obj[key]; + for (const k in obj) { + const result = deepSearch(obj[k], key, depth - 1); + if (result) return result; + } + } catch (e) { } + return null; + } + + function parseToken(token) { + try { + const parts = token.split('.'); + let payload = parts[1]; + payload += '='.repeat((4 - payload.length % 4) % 4); + const info = JSON.parse(atob(payload)); + const remaining = Math.floor((info.exp * 1000 - Date.now()) / 60000); + return { valid: remaining > 0, remaining }; + } catch (e) { return { valid: false, remaining: 0 }; } + } + + function updatePanel() { + const statusEl = document.getElementById('tg-status'); + const previewEl = document.getElementById('tg-preview'); + const expiryEl = document.getElementById('tg-expiry'); + const iconStatus = document.getElementById('tg-icon-status'); + + if (!foundToken) { + if (statusEl) { + statusEl.textContent = '❌ 未找到'; + statusEl.className = 'status expired'; + } + if (previewEl) previewEl.textContent = '请确保已登录'; + if (iconStatus) iconStatus.style.background = '#f66'; + return; + } + + const info = parseToken(foundToken); + if (info.valid) { + if (statusEl) { + statusEl.textContent = '✅ Token 已获取'; + statusEl.className = 'status captured'; + } + if (expiryEl) expiryEl.innerHTML = `⏱️ 剩余 ${info.remaining} 分钟`; + if (iconStatus) iconStatus.style.background = '#6f6'; + } else { + if (statusEl) { + statusEl.textContent = '❌ Token 已过期'; + statusEl.className = 'status expired'; + } + if (expiryEl) expiryEl.innerHTML = '请重新登录'; + if (iconStatus) iconStatus.style.background = '#f66'; + } + if (previewEl) previewEl.textContent = foundToken.substring(0, 50) + '...'; + } + + function copyToken() { + if (foundToken) { + GM_setClipboard(foundToken); + const btn = document.getElementById('tg-copy-btn'); + if (btn) { + btn.textContent = '✅ 已复制!'; + setTimeout(() => btn.textContent = '📋 复制 Token', 1500); + } + } else { + alert('未找到 Token'); + } + } + + function refreshToken() { + foundToken = searchToken() || ''; + if (foundToken) GM_setValue(STORAGE_KEY, foundToken); + updatePanel(); + } + + function toggleCollapse() { + isCollapsed = !isCollapsed; + GM_setValue(COLLAPSED_KEY, isCollapsed); + const panel = document.getElementById('token-grabber-panel'); + if (isCollapsed) { + panel.classList.add('collapsed'); + } else { + panel.classList.remove('collapsed'); + } + } + + // 同步到 Pandy Image 页面（使用 GM_xmlhttpRequest 绕过跨域） + async function syncToPandy() { + if (!foundToken) { + alert('未找到 Token'); + return; + } + + const btn = document.getElementById('tg-sync-btn'); + if (btn) btn.textContent = '⏳ 同步中...'; + + const PANDY_BASE = 'http://127.0.0.1:8180'; + + // 使用 GM_xmlhttpRequest 绕过跨域限制 + // 1. 先获取现有配置 + GM_xmlhttpRequest({ + method: 'GET', + url: `${PANDY_BASE}/config?name=settings`, + onload: function (response) { + let config = {}; + try { + if (response.status === 200) { + config = JSON.parse(response.responseText); + } + } catch (e) { } + + // 2. 合并 token + config.token = foundToken; + + // 3. POST 回去 + GM_xmlhttpRequest({ + method: 'POST', + url: `${PANDY_BASE}/config?name=settings`, + headers: { 'Content-Type': 'application/json' }, + data: JSON.stringify(config), + onload: function (resp) { + if (resp.status === 200) { + if (btn) { + btn.textContent = '✅ 已同步!'; + setTimeout(() => btn.textContent = '🚀 同步到 Pandy', 2000); + } + console.log('🍌 Token 已同步到 Pandy'); + } else { + fallbackCopy(btn); + } + }, + onerror: function () { + fallbackCopy(btn); + } + }); + }, + onerror: function () { + fallbackCopy(btn); + } + }); + } + + function fallbackCopy(btn) { + GM_setClipboard(foundToken); + if (btn) { + btn.textContent = '❌ 服务未启动，已复制'; + setTimeout(() => btn.textContent = '🚀 同步到 Pandy', 3000); + } + console.log('🍌 同步失败，请确保 Pandy 服务器已启动 (python server.py)'); + } + + unsafeWindow.copyToken = copyToken; + unsafeWindow.refreshToken = refreshToken; + unsafeWindow.toggleCollapse = toggleCollapse; + unsafeWindow.syncToPandy = syncToPandy; + + function createPanel() { + const panel = document.createElement('div'); + panel.id = 'token-grabber-panel'; + if (isCollapsed) panel.classList.add('collapsed'); + + // 设置位置 + if (savedPos.x !== null) { + panel.style.left = savedPos.x + 'px'; + panel.style.right = 'auto'; + } + panel.style.top = savedPos.y + 'px'; + + panel.innerHTML = ` + +
+ 🍌 + Token Grabber + + ▼ +
+
+
🔍 搜索中...
+
正在搜索...
+
+ + + +
+ `; + document.body.appendChild(panel); + + // 拖拽功能 + const header = document.getElementById('tg-header'); + let isDragging = false; + let startX, startY, startLeft, startTop; + + header.addEventListener('mousedown', (e) => { + if (e.target.classList.contains('collapse-btn')) return; + isDragging = true; + const rect = panel.getBoundingClientRect(); + startX = e.clientX; + startY = e.clientY; + startLeft = rect.left; + startTop = rect.top; + panel.style.right = 'auto'; + panel.style.left = startLeft + 'px'; + e.preventDefault(); + }); + + document.addEventListener('mousemove', (e) => { + if (!isDragging) return; + const dx = e.clientX - startX; + const dy = e.clientY - startY; + let newLeft = startLeft + dx; + let newTop = startTop + dy; + + // 边界限制 + newLeft = Math.max(0, Math.min(newLeft, window.innerWidth - panel.offsetWidth)); + newTop = Math.max(0, Math.min(newTop, window.innerHeight - panel.offsetHeight)); + + panel.style.left = newLeft + 'px'; + panel.style.top = newTop + 'px'; + }); + + document.addEventListener('mouseup', () => { + if (isDragging) { + isDragging = false; + // 保存位置 + const rect = panel.getBoundingClientRect(); + savedPos = { x: rect.left, y: rect.top }; + GM_setValue(POS_KEY, savedPos); + } + }); + } + + function init() { + console.log('🍌 Token Grabber: init() 开始执行'); + console.log('🍌 document.body:', document.body); + + if (!document.body) { + console.log('🍌 body 不存在，延迟重试...'); + setTimeout(init, 500); + return; + } + + createPanel(); + console.log('🍌 面板已创建'); + + // 确保面板在可视范围内 + setTimeout(clampPosition, 100); + + // 监听窗口大小变化 + window.addEventListener('resize', clampPosition); + + setTimeout(() => { + foundToken = searchToken() || GM_getValue(STORAGE_KEY, ''); + if (foundToken) GM_setValue(STORAGE_KEY, foundToken); + updatePanel(); + console.log('🍌 Token 搜索完成:', foundToken ? '找到' : '未找到'); + }, 1500); + + setInterval(() => { + const newToken = searchToken(); + if (newToken && newToken !== foundToken) { + foundToken = newToken; + GM_setValue(STORAGE_KEY, foundToken); + updatePanel(); + } + }, 10000); + } + + // 多种方式确保执行 + console.log('🍌 Token Grabber: 脚本已加载, readyState:', document.readyState); + + if (document.readyState === 'complete') { + init(); + } else if (document.readyState === 'interactive') { + setTimeout(init, 100); + } else { + window.addEventListener('load', init); + document.addEventListener('DOMContentLoaded', () => setTimeout(init, 100)); + } +})();