From a25fbf29c1cc5b7e9f10468b799b5cbd5b71eb47 Mon Sep 17 00:00:00 2001 From: pandy <1272175942@qq.com> Date: Wed, 22 Apr 2026 17:19:41 +0800 Subject: [PATCH] GPT2 213 --- __init__.py | 73 +++++------ py/IMAGE_GPT2_apikey.py | 211 +++++++++++++++++++++++++++++++ py/IMAGE_GPT2_authtoken.py | 209 +++++++++++++++++++++++++++++++ py/IMAGE_GPT_apikey.py | 249 ++++++++++--------------------------- py/IMAGE_GPT_authtoken.py | 242 +++++++++-------------------------- py/_image_gpt_common.py | 189 ++++++++++++++++++++++++++++ 6 files changed, 759 insertions(+), 414 deletions(-) create mode 100644 py/IMAGE_GPT2_apikey.py create mode 100644 py/IMAGE_GPT2_authtoken.py create mode 100644 py/_image_gpt_common.py diff --git a/__init__.py b/__init__.py index 9b9e0f8..bd13aab 100644 --- a/__init__.py +++ b/__init__.py @@ -1,74 +1,61 @@ -""" -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__) + +def get_ext_dir(subpath=None): + base_dir = os.path.dirname(__file__) if subpath is not None: - dir = os.path.join(dir, subpath) - return os.path.abspath(dir) + base_dir = os.path.join(base_dir, subpath) + return os.path.abspath(base_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'): + + 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'): + print(f"[PD_NODE] Loaded module: {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}") + print(f"[PD_NODE] Failed to load module {file_name}: {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("_"): + for file_name in os.listdir(py_dir): + if not file_name.endswith(".py") or file_name.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 + module_name = os.path.splitext(file_name)[0] + module_path = f"{__name__}.py.{module_name}" + safe_import_module(module_path, file_name) +else: + print(f"[PD_NODE] Missing py directory: {py_dir}") + + 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("[PD_NODE] Comfyui-PD_comfy-api-node loaded") 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(f"[PD_NODE] Total nodes: {len(NODE_CLASS_MAPPINGS)}") + for node_name, display_name in NODE_DISPLAY_NAME_MAPPINGS.items(): + print(f"[PD_NODE] {node_name} -> {display_name}") print("=" * 50) else: - print("⚠️ 警告: 没有找到任何可用的节点") + print("[PD_NODE] Warning: no nodes were loaded") + -# 导出变量 __all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS", "WEB_DIRECTORY"] diff --git a/py/IMAGE_GPT2_apikey.py b/py/IMAGE_GPT2_apikey.py new file mode 100644 index 0000000..7f0a948 --- /dev/null +++ b/py/IMAGE_GPT2_apikey.py @@ -0,0 +1,211 @@ +""" +PD: OpenAI GPT Image 2 Node (ComfyUI API Key) +""" + +import time + +import requests +import torch + +from ._image_gpt_common import ( + GPT_IMAGE_2_PRESET_SIZES, + calculate_token_price_2, + decode_response_images, + empty_image, + estimate_price_range_2, + format_rmb_range, + format_rmb_value, + mask_to_png_bytes, + resolve_gpt_image_2_size, + tensor_to_bytes, + validate_gpt_image_2_size, +) + + +class PDOpenAIGPTImage2APIKey: + @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 2" + }), + "model": (["gpt-image-2"], { + "default": "gpt-image-2" + }), + "seed": ("INT", { + "default": 0, + "min": 0, + "max": 2**31 - 1, + "step": 1, + "tooltip": "not implemented yet in backend" + }), + "quality": (["auto", "low", "medium", "high"], { + "default": "auto", + "tooltip": "GPT Image 2 quality" + }), + "background": (["auto", "opaque"], { + "default": "auto", + "tooltip": "GPT Image 2 does not support transparent background" + }), + "size": (GPT_IMAGE_2_PRESET_SIZES, { + "default": "1024x1024", + "tooltip": "Preset output size" + }), + "custom_width": ("INT", { + "default": 0, + "min": 0, + "max": 3840, + "step": 16, + "tooltip": "Override width when both width and height are non-zero" + }), + "custom_height": ("INT", { + "default": 0, + "min": 0, + "max": 3840, + "step": 16, + "tooltip": "Override height when both width and height are non-zero" + }), + "num_images": ("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)" + }), + }, + "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, seed, quality, background, size, custom_width, custom_height, num_images, unique_id, image=None, mask=None): + if not api_key or not api_key.strip(): + error_msg = "Error: ComfyUI API Key is required" + print(f"[IMAGE_GPT2_APIKEY] {error_msg}") + return (empty_image(), error_msg) + + try: + resolved_size = validate_gpt_image_2_size(size, custom_width, custom_height) + except Exception as e: + return (empty_image(), f"Error: {str(e)}") + + api_key = api_key.strip() + is_edit_mode = image is not None + base_url = "https://api.comfy.org/proxy/openai/images/edits" if is_edit_mode else "https://api.comfy.org/proxy/openai/images/generations" + start_time = time.time() + + try: + if is_edit_mode: + files = { + "image": ("image.png", tensor_to_bytes(image), "image/png") + } + if mask is not None: + files["mask"] = ("mask.png", mask_to_png_bytes(mask), "image/png") + + data = { + "model": model, + "prompt": prompt, + "n": str(num_images), + "size": resolved_size, + "quality": quality, + "background": background, + "moderation": "low" + } + headers = { + "X-API-KEY": api_key, + "Accept": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.1" + } + response = requests.post(base_url, headers=headers, files=files, data=data, timeout=300) + else: + payload = { + "model": model, + "prompt": prompt, + "n": num_images, + "size": resolved_size, + "quality": quality, + "background": background, + "moderation": "low" + } + headers = { + "X-API-KEY": api_key, + "Accept": "application/json", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.1" + } + response = requests.post(base_url, headers=headers, json=payload, timeout=300) + + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[IMAGE_GPT2_APIKEY] {error_msg}") + return (empty_image(), error_msg) + + result = response.json() + image_tensors = decode_response_images(result) + if not image_tensors: + error_msg = "Error: Failed to process output images" + print(f"[IMAGE_GPT2_APIKEY] {error_msg}") + return (empty_image(), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + duration = time.time() - start_time + min_cost, max_cost = estimate_price_range_2(quality, num_images) + token_cost = calculate_token_price_2(num_images) + size_note = resolve_gpt_image_2_size(size, custom_width, custom_height) + + info_str = ( + f"模型: {model}\n" + f"显示名称: OpenAI GPT Image 2\n" + f"模式: {'编辑' if is_edit_mode else '生成'}\n" + f"状态: 成功\n" + f"耗时: {duration:.2f} 秒\n" + f"图片数量: {len(image_tensors)}\n" + f"尺寸预设: {size}\n" + f"实际尺寸: {size_note}\n" + f"质量: {quality}\n" + f"背景: {background}\n" + f"预估价格区间(USD): ~${min_cost:.3f}-${max_cost:.3f}\n" + f"预估价格区间(RMB): {format_rmb_range(min_cost, max_cost)}\n" + f"Token 公式估算(USD): ${token_cost:.4f}\n" + f"Token 公式估算(RMB): {format_rmb_value(token_cost)}" + ) + return (final_image, info_str) + + except requests.exceptions.Timeout: + return (empty_image(), "Error: Request timeout (300s)") + except requests.exceptions.ConnectionError as e: + return (empty_image(), f"Error: Connection failed - {str(e)}") + except Exception as e: + import traceback + traceback.print_exc() + return (empty_image(), f"Error: {str(e)}") + + +NODE_CLASS_MAPPINGS = { + "PDOpenAIGPTImage2APIKey": PDOpenAIGPTImage2APIKey +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDOpenAIGPTImage2APIKey": "PD: GPT Image 2 (apikey)" +} diff --git a/py/IMAGE_GPT2_authtoken.py b/py/IMAGE_GPT2_authtoken.py new file mode 100644 index 0000000..ae17510 --- /dev/null +++ b/py/IMAGE_GPT2_authtoken.py @@ -0,0 +1,209 @@ +""" +PD: OpenAI GPT Image 2 Node (ComfyUI Auth Token) +""" + +import time + +import requests +import torch + +from ._image_gpt_common import ( + GPT_IMAGE_2_PRESET_SIZES, + calculate_token_price_2, + decode_response_images, + empty_image, + estimate_price_range_2, + format_rmb_range, + format_rmb_value, + mask_to_png_bytes, + resolve_gpt_image_2_size, + tensor_to_bytes, + validate_gpt_image_2_size, +) + + +class PDOpenAIGPTImage2AuthToken: + @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-2"], { + "default": "gpt-image-2", + "tooltip": "Locked GPT Image 2 model" + }), + "quality": (["auto", "low", "medium", "high"], { + "default": "auto", + "tooltip": "GPT Image 2 quality" + }), + "background": (["auto", "opaque"], { + "default": "auto", + "tooltip": "GPT Image 2 does not support transparent background" + }), + "size": (GPT_IMAGE_2_PRESET_SIZES, { + "default": "1024x1024", + "tooltip": "Preset output size" + }), + "custom_width": ("INT", { + "default": 0, + "min": 0, + "max": 3840, + "step": 16, + "tooltip": "Override width when both width and height are non-zero" + }), + "custom_height": ("INT", { + "default": 0, + "min": 0, + "max": 3840, + "step": 16, + "tooltip": "Override height when both width and height are non-zero" + }), + "num_images": ("INT", { + "default": 1, + "min": 1, + "max": 8, + "step": 1, + "tooltip": "How many 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, custom_width, custom_height, num_images, seed, unique_id, image=None, mask=None): + if not auth_token or not auth_token.strip(): + error_msg = "Error: Auth Token is required" + print(f"[IMAGE_GPT2_AUTHTOKEN] {error_msg}") + return (empty_image(), error_msg) + + try: + resolved_size = validate_gpt_image_2_size(size, custom_width, custom_height) + except Exception as e: + return (empty_image(), f"Error: {str(e)}") + + auth_token = auth_token.strip() + if auth_token.lower().startswith("bearer "): + auth_token = auth_token[7:].strip() + + base_url = "https://api.comfy.org" + is_edit_mode = image is not None + start_time = time.time() + + try: + if is_edit_mode: + url = f"{base_url}/proxy/openai/images/edits" + files = { + "image": ("image.png", tensor_to_bytes(image), "image/png") + } + if mask is not None: + files["mask"] = ("mask.png", mask_to_png_bytes(mask), "image/png") + + data = { + "model": model, + "prompt": prompt, + "n": str(num_images), + "size": resolved_size, + "quality": quality, + "background": background, + "moderation": "low" + } + headers = { + "Authorization": f"Bearer {auth_token}", + "User-Agent": "ComfyUI-PD-Node/2.1" + } + response = requests.post(url, headers=headers, files=files, data=data, timeout=300) + else: + url = f"{base_url}/proxy/openai/images/generations" + payload = { + "model": model, + "prompt": prompt, + "n": num_images, + "size": resolved_size, + "quality": quality, + "background": background, + "moderation": "low" + } + headers = { + "Authorization": f"Bearer {auth_token}", + "Content-Type": "application/json", + "User-Agent": "ComfyUI-PD-Node/2.1" + } + response = requests.post(url, headers=headers, json=payload, timeout=300) + + if response.status_code != 200: + error_msg = f"API Error {response.status_code}: {response.text[:200]}" + print(f"[IMAGE_GPT2_AUTHTOKEN] {error_msg}") + return (empty_image(), error_msg) + + result = response.json() + image_tensors = decode_response_images(result) + if not image_tensors: + error_msg = "Error: Failed to process output images" + print(f"[IMAGE_GPT2_AUTHTOKEN] {error_msg}") + return (empty_image(), error_msg) + + final_image = torch.cat(image_tensors, dim=0) + duration = time.time() - start_time + min_cost, max_cost = estimate_price_range_2(quality, num_images) + token_cost = calculate_token_price_2(num_images) + size_note = resolve_gpt_image_2_size(size, custom_width, custom_height) + + info_str = ( + f"模型: {model}\n" + f"显示名称: OpenAI GPT Image 2\n" + f"模式: {'编辑' if is_edit_mode else '生成'}\n" + f"状态: 成功\n" + f"耗时: {duration:.2f} 秒\n" + f"图片数量: {len(image_tensors)}\n" + f"尺寸预设: {size}\n" + f"实际尺寸: {size_note}\n" + f"质量: {quality}\n" + f"背景: {background}\n" + f"预估价格区间(USD): ~${min_cost:.3f}-${max_cost:.3f}\n" + f"预估价格区间(RMB): {format_rmb_range(min_cost, max_cost)}\n" + f"Token 公式估算(USD): ${token_cost:.4f}\n" + f"Token 公式估算(RMB): {format_rmb_value(token_cost)}" + ) + return (final_image, info_str) + + except requests.exceptions.Timeout: + return (empty_image(), "Error: Request timeout (300s)") + except requests.exceptions.ConnectionError as e: + return (empty_image(), f"Error: Connection failed - {str(e)}") + except Exception as e: + import traceback + traceback.print_exc() + return (empty_image(), f"Error: {str(e)}") + + +NODE_CLASS_MAPPINGS = { + "PDOpenAIGPTImage2AuthToken": PDOpenAIGPTImage2AuthToken +} + +NODE_DISPLAY_NAME_MAPPINGS = { + "PDOpenAIGPTImage2AuthToken": "PD: GPT Image 2 (comfyui_AuthToken)" +} diff --git a/py/IMAGE_GPT_apikey.py b/py/IMAGE_GPT_apikey.py index 317f203..1d65571 100644 --- a/py/IMAGE_GPT_apikey.py +++ b/py/IMAGE_GPT_apikey.py @@ -1,86 +1,47 @@ """ -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 +PD: OpenAI GPT Image 1 & 1.5 Node (ComfyUI API Key) """ +import time + 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)) +from ._image_gpt_common import ( + calculate_token_price_1_or_1_5, + decode_response_images, + empty_image, + estimate_price_range_1_or_1_5, + format_rmb_range, + format_rmb_value, + load_image_to_tensor, + mask_to_png_bytes, + tensor_to_bytes, +) 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, + "default": "", + "multiline": False, "placeholder": "Your ComfyUI API Key" }), "prompt": ("STRING", { - "multiline": True, - "default": "", - "tooltip": "Text prompt for GPT Image" + "multiline": True, + "default": "", + "tooltip": "Text prompt for GPT Image 1 or 1.5" + }), + "model": (["gpt-image-1", "gpt-image-1.5"], { + "default": "gpt-image-1.5", + "tooltip": "GPT Image model version" }), "seed": ("INT", { - "default": 0, - "min": 0, + "default": 0, + "min": 0, "max": 2**31 - 1, "step": 1, "tooltip": "not implemented yet in backend" @@ -98,9 +59,9 @@ class PDOpenAIGPTImageAPIKey: "tooltip": "Image size" }), "n": ("INT", { - "default": 1, - "min": 1, - "max": 8, + "default": 1, + "min": 1, + "max": 8, "step": 1, "tooltip": "How many images to generate" }), @@ -112,9 +73,6 @@ class PDOpenAIGPTImageAPIKey: "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", @@ -126,57 +84,25 @@ class PDOpenAIGPTImageAPIKey: 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 + def generate_image(self, api_key, prompt, model, seed, quality, background, size, n, unique_id, image=None, mask=None): 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() + return (empty_image(), error_msg) - # 判断模式：编辑 vs 生成 + api_key = api_key.strip() 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" - + base_url = "https://api.comfy.org/proxy/openai/images/edits" if is_edit_mode else "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 + files = { + "image": ("image.png", tensor_to_bytes(image), "image/png") + } 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') + files["mask"] = ("mask.png", mask_to_png_bytes(mask), "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, @@ -186,22 +112,13 @@ class PDOpenAIGPTImageAPIKey: "background": background, "moderation": "low" } - - # 请求头 headers = { "X-API-KEY": api_key, "Accept": "application/json", - "User-Agent": "ComfyUI-PD-Node/2.0" + "User-Agent": "ComfyUI-PD-Node/2.1" } - - 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, @@ -209,104 +126,64 @@ class PDOpenAIGPTImageAPIKey: "size": size, "quality": quality, "background": background, - "response_format": "url" + "moderation": "low" } - headers = { "X-API-KEY": api_key, "Accept": "application/json", "Content-Type": "application/json", - "User-Agent": "ComfyUI-PD-Node/2.0" + "User-Agent": "ComfyUI-PD-Node/2.1" } - - 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) + return (empty_image(), 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) - + image_tensors = decode_response_images(result) 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) + return (empty_image(), 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 + duration = time.time() - start_time + min_cost, max_cost = estimate_price_range_1_or_1_5(model, quality, n) + token_cost = calculate_token_price_1_or_1_5(model, 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}" + f"模型: {model}\n" + f"显示名称: OpenAI GPT Image 1 & 1.5\n" + f"模式: {'编辑' if is_edit_mode else '生成'}\n" + f"状态: 成功\n" + f"耗时: {duration:.2f} 秒\n" + f"图片数量: {len(image_tensors)}\n" + f"尺寸: {size}\n" + f"质量: {quality}\n" + f"背景: {background}\n" + f"预估价格区间(USD): ${min_cost:.3f}-${max_cost:.3f}\n" + f"预估价格区间(RMB): {format_rmb_range(min_cost, max_cost)}\n" + f"Token 公式估算(USD): ${token_cost:.4f}\n" + f"Token 公式估算(RMB): {format_rmb_value(token_cost)}" ) - - 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) - + return (empty_image(), "Error: Request timeout (300s)") 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) - + return (empty_image(), f"Error: Connection failed - {str(e)}") 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) + return (empty_image(), f"Error: {str(e)}") -# 节点注册 NODE_CLASS_MAPPINGS = { "PDOpenAIGPTImageAPIKey": PDOpenAIGPTImageAPIKey } NODE_DISPLAY_NAME_MAPPINGS = { - "PDOpenAIGPTImageAPIKey": "PD: GPT Image (apikey)" + "PDOpenAIGPTImageAPIKey": "PD: GPT Image 1 & 1.5 (apikey)" } diff --git a/py/IMAGE_GPT_authtoken.py b/py/IMAGE_GPT_authtoken.py index d8d392e..c4b4a62 100644 --- a/py/IMAGE_GPT_authtoken.py +++ b/py/IMAGE_GPT_authtoken.py @@ -1,85 +1,41 @@ """ -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 +PD: OpenAI GPT Image 1 & 1.5 Node (ComfyUI Auth Token) """ -import torch -import requests import time -import io -import base64 -import numpy as np -from PIL import Image +import requests +import torch -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)) +from ._image_gpt_common import ( + calculate_token_price_1_or_1_5, + decode_response_images, + empty_image, + estimate_price_range_1_or_1_5, + format_rmb_range, + format_rmb_value, + mask_to_png_bytes, + tensor_to_bytes, +) 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, + "default": "", + "multiline": False, "placeholder": "Paste your Auth Token here" }), "prompt": ("STRING", { - "multiline": True, - "default": "", + "multiline": True, + "default": "", "tooltip": "Describe what you want to generate or edit" }), "model": (["gpt-image-1", "gpt-image-1.5"], { - "default": "gpt-image-1", + "default": "gpt-image-1.5", "tooltip": "GPT Image model version" }), "quality": (["low", "medium", "high"], { @@ -88,22 +44,22 @@ class PDOpenAIGPTImageAuthToken: }), "background": (["auto", "opaque", "transparent"], { "default": "auto", - "tooltip": "Background transparency (GPT Image only)" + "tooltip": "Background transparency" }), "size": (["auto", "1024x1024", "1024x1536", "1536x1024"], { "default": "auto", "tooltip": "Image size" }), "n": ("INT", { - "default": 1, - "min": 1, - "max": 8, + "default": 1, + "min": 1, + "max": 8, "step": 1, "tooltip": "Number of images to generate" }), "seed": ("INT", { - "default": 0, - "min": 0, + "default": 0, + "min": 0, "max": 0xFFFFFFFFFFFFFFFF }), }, @@ -122,57 +78,28 @@ class PDOpenAIGPTImageAuthToken: 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 + return (empty_image(), error_msg) + 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 + files = { + "image": ("image.png", tensor_to_bytes(image), "image/png") + } 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') + files["mask"] = ("mask.png", mask_to_png_bytes(mask), "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, @@ -182,23 +109,13 @@ class PDOpenAIGPTImageAuthToken: "background": background, "moderation": "low" } - - # 请求头 headers = { "Authorization": f"Bearer {auth_token}", - "User-Agent": "ComfyUI-PD-Node/2.0" + "User-Agent": "ComfyUI-PD-Node/2.1" } - - 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, @@ -206,108 +123,63 @@ class PDOpenAIGPTImageAuthToken: "size": size, "quality": quality, "background": background, - "moderation": "low", - "response_format": "url" + "moderation": "low" } - headers = { "Authorization": f"Bearer {auth_token}", "Content-Type": "application/json", - "User-Agent": "ComfyUI-PD-Node/2.0" + "User-Agent": "ComfyUI-PD-Node/2.1" } - - 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) + return (empty_image(), 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) - + image_tensors = decode_response_images(result) 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) + return (empty_image(), 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 + duration = time.time() - start_time + min_cost, max_cost = estimate_price_range_1_or_1_5(model, quality, n) + token_cost = calculate_token_price_1_or_1_5(model, 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}" + f"模型: {model}\n" + f"显示名称: OpenAI GPT Image 1 & 1.5\n" + f"模式: {'编辑' if is_edit_mode else '生成'}\n" + f"状态: 成功\n" + f"耗时: {duration:.2f} 秒\n" + f"图片数量: {len(image_tensors)}\n" + f"尺寸: {size}\n" + f"质量: {quality}\n" + f"背景: {background}\n" + f"预估价格区间(USD): ${min_cost:.3f}-${max_cost:.3f}\n" + f"预估价格区间(RMB): {format_rmb_range(min_cost, max_cost)}\n" + f"Token 公式估算(USD): ${token_cost:.4f}\n" + f"Token 公式估算(RMB): {format_rmb_value(token_cost)}" ) - - 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) - + return (empty_image(), "Error: Request timeout (300s)") 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) - + return (empty_image(), f"Error: Connection failed - {str(e)}") 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) + return (empty_image(), f"Error: {str(e)}") -# 节点注册 NODE_CLASS_MAPPINGS = { "PDOpenAIGPTImageAuthToken": PDOpenAIGPTImageAuthToken } NODE_DISPLAY_NAME_MAPPINGS = { - "PDOpenAIGPTImageAuthToken": "PD: GPT Image (comfyui_AuthToken)" + "PDOpenAIGPTImageAuthToken": "PD: GPT Image 1 & 1.5 (comfyui_AuthToken)" } diff --git a/py/_image_gpt_common.py b/py/_image_gpt_common.py new file mode 100644 index 0000000..a298cac --- /dev/null +++ b/py/_image_gpt_common.py @@ -0,0 +1,189 @@ +import base64 +import io + +import numpy as np +import requests +import torch +from PIL import Image + + +GPT_IMAGE_EDIT_MAX_PIXELS = 2048 * 2048 +GPT_IMAGE_2_PRESET_SIZES = [ + "1024x1024", + "1536x1024", + "1024x1536", + "2048x2048", + "2048x1152", + "3840x2160", + "2160x3840", + "auto", +] + +# USD/CNY reference rate used for RMB estimation in node info output. +# Based on the most recent publicly indexed April 2026 daily spot pages we could verify: +# 2026-04-03: 1 USD ≈ 6.88154 CNY +USD_TO_CNY_RATE = 6.8815 + + +def empty_image(): + return torch.zeros((1, 512, 512, 3)) + + +def tensor_to_bytes(tensor, max_pixels=GPT_IMAGE_EDIT_MAX_PIXELS): + if tensor.ndim == 4: + tensor = tensor[0] + + height, width = tensor.shape[:2] + current_pixels = height * width + if current_pixels > max_pixels: + scale = (max_pixels / current_pixels) ** 0.5 + new_height = max(1, int(height * scale)) + new_width = max(1, int(width * scale)) + array = np.clip(255.0 * 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(255.0 * 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): + 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: + return empty_image() + + +def mask_to_png_bytes(mask): + mask_tensor = mask + if mask_tensor.ndim == 3: + mask_tensor = mask_tensor[0] + + mask_h, mask_w = mask_tensor.shape + rgba = np.zeros((mask_h, mask_w, 4), dtype=np.uint8) + mask_np = mask_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) + return mask_buffer + + +def decode_response_images(result): + data_list = result.get("data", []) + 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) + return image_tensors + + +def estimate_price_range_1_or_1_5(model, quality, n): + ranges = { + "gpt-image-1": { + "low": (0.011, 0.016), + "medium": (0.042, 0.063), + "high": (0.167, 0.250), + }, + "gpt-image-1.5": { + "low": (0.009, 0.013), + "medium": (0.034, 0.050), + "high": (0.133, 0.200), + }, + } + model_ranges = ranges.get(model, ranges["gpt-image-1"]) + min_cost, max_cost = model_ranges.get(quality, model_ranges["medium"]) + return (min_cost * n, max_cost * n) + + +def calculate_token_price_1_or_1_5(model, n): + if model == "gpt-image-1": + cost = (1000 * 10 + 2000 * 40) / 1_000_000 + else: + cost = (1000 * 8 + 2000 * 32) / 1_000_000 + return cost * n + + +def estimate_price_range_2(quality, n): + ranges = { + "auto": (0.005, 0.250), + "low": (0.005, 0.030), + "medium": (0.041, 0.120), + "high": (0.165, 0.250), + } + min_cost, max_cost = ranges.get(quality, ranges["auto"]) + return (min_cost * n, max_cost * n) + + +def calculate_token_price_2(n): + cost = (1000 * 8 + 2000 * 30) / 1_000_000 + return cost * n + + +def resolve_gpt_image_2_size(size, custom_width, custom_height): + if custom_width > 0 and custom_height > 0: + return f"{custom_width}x{custom_height}" + return size + + +def usd_to_cny(usd_amount): + return usd_amount * USD_TO_CNY_RATE + + +def format_rmb_range(min_usd, max_usd): + return f"约￥{usd_to_cny(min_usd):.2f}-￥{usd_to_cny(max_usd):.2f}" + + +def format_rmb_value(usd_amount): + return f"约￥{usd_to_cny(usd_amount):.2f}" + + +def validate_gpt_image_2_size(size, custom_width=0, custom_height=0): + resolved_size = resolve_gpt_image_2_size(size, custom_width, custom_height) + if resolved_size == "auto": + return resolved_size + + try: + width_str, height_str = resolved_size.lower().split("x") + width = int(width_str) + height = int(height_str) + except Exception as exc: + raise ValueError(f"Invalid size: {resolved_size}") from exc + + if width > 3840 or height > 3840: + raise ValueError("Width and height must not exceed 3840") + if width % 16 != 0 or height % 16 != 0: + raise ValueError("Width and height must be multiples of 16") + + shorter_edge = min(width, height) + longer_edge = max(width, height) + if shorter_edge == 0 or (longer_edge / shorter_edge) > 3: + raise ValueError("Aspect ratio must not exceed 3:1") + + total_pixels = width * height + if total_pixels < 655360 or total_pixels > 8294400: + raise ValueError("Total pixels must be between 655360 and 8294400") + + return resolved_size