diff --git a/app/services/clip_video.py b/app/services/clip_video.py index 05dfcf8..81794e4 100644 --- a/app/services/clip_video.py +++ b/app/services/clip_video.py @@ -87,7 +87,7 @@ def check_hardware_acceleration() -> Optional[str]: def get_safe_encoder_config(hwaccel_type: Optional[str] = None) -> Dict[str, str]: """ - 获取安全的编码器配置，针对Windows平台优化 + 获取安全的编码器配置，基于ffmpeg_demo.py成功方案优化 Args: hwaccel_type: 硬件加速类型 @@ -95,28 +95,48 @@ def get_safe_encoder_config(hwaccel_type: Optional[str] = None) -> Dict[str, str Returns: Dict[str, str]: 编码器配置字典 """ + # 基础配置 - 参考ffmpeg_demo.py的成功方案 config = { "video_codec": "libx264", "audio_codec": "aac", "pixel_format": "yuv420p", - "preset": "fast", - "crf": "23" + "preset": "medium", + "quality_param": "crf", # 质量参数类型 + "quality_value": "23" # 质量值 } - # 根据硬件加速类型调整配置 - if hwaccel_type == "cuda": + # 根据硬件加速类型调整配置（简化版本） + if hwaccel_type in ["nvenc_pure", "nvenc_software", "cuda_careful", "nvenc", "cuda", "cuda_decode"]: + # NVIDIA硬件加速 - 使用ffmpeg_demo.py中验证有效的参数 config["video_codec"] = "h264_nvenc" - config["preset"] = "fast" + config["preset"] = "medium" + config["quality_param"] = "cq" # CQ质量控制，而不是CRF + config["quality_value"] = "23" config["pixel_format"] = "yuv420p" + elif hwaccel_type == "amf": + # AMD AMF编码器 + config["video_codec"] = "h264_amf" + config["preset"] = "balanced" + config["quality_param"] = "qp_i" + config["quality_value"] = "23" elif hwaccel_type == "qsv": + # Intel QSV编码器 config["video_codec"] = "h264_qsv" - config["preset"] = "fast" - elif hwaccel_type == "d3d11va" or hwaccel_type == "dxva2": - # Windows平台的硬件解码，但使用软件编码 - config["video_codec"] = "libx264" - config["preset"] = "fast" + config["preset"] = "medium" + config["quality_param"] = "global_quality" + config["quality_value"] = "23" elif hwaccel_type == "videotoolbox": + # macOS VideoToolbox编码器 config["video_codec"] = "h264_videotoolbox" + config["preset"] = "medium" + config["quality_param"] = "b:v" + config["quality_value"] = "5M" + else: + # 软件编码（默认） + config["video_codec"] = "libx264" + config["preset"] = "medium" + config["quality_param"] = "crf" + config["quality_value"] = "23" return config @@ -130,7 +150,10 @@ def build_ffmpeg_command( hwaccel_args: List[str] = None ) -> List[str]: """ - 构建优化的ffmpeg命令 + 构建优化的ffmpeg命令，基于测试结果使用正确的硬件加速方案 + + 重要发现：对于视频裁剪场景，CUDA硬件解码会导致滤镜链错误， + 应该使用纯NVENC编码器（无硬件解码）来获得最佳兼容性 Args: input_path: 输入视频路径 @@ -145,8 +168,14 @@ def build_ffmpeg_command( """ cmd = ["ffmpeg", "-y"] - # 添加硬件加速参数（如果有） - if hwaccel_args: + # 关键修正：对于视频裁剪，不使用CUDA硬件解码，只使用NVENC编码器 + # 这样能避免滤镜链格式转换错误，同时保持编码性能优势 + if encoder_config["video_codec"] == "h264_nvenc": + # 不添加硬件解码参数，让FFmpeg自动处理 + # 这避免了 "Impossible to convert between the formats" 错误 + pass + elif hwaccel_args: + # 对于其他编码器，可以使用硬件解码参数 cmd.extend(hwaccel_args) # 输入文件 @@ -159,25 +188,39 @@ def build_ffmpeg_command( cmd.extend(["-c:v", encoder_config["video_codec"]]) cmd.extend(["-c:a", encoder_config["audio_codec"]]) - # 像素格式（关键：避免滤镜链问题） + # 像素格式 cmd.extend(["-pix_fmt", encoder_config["pixel_format"]]) - # 编码质量设置 - if encoder_config["video_codec"] == "libx264": + # 质量和预设参数 - 针对NVENC优化 + if encoder_config["video_codec"] == "h264_nvenc": + # 纯NVENC编码器配置（无硬件解码，兼容性最佳） cmd.extend(["-preset", encoder_config["preset"]]) - cmd.extend(["-crf", encoder_config["crf"]]) - elif encoder_config["video_codec"] == "h264_nvenc": - cmd.extend(["-preset", encoder_config["preset"]]) - cmd.extend(["-rc", "vbr", "-cq", encoder_config["crf"]]) + cmd.extend(["-cq", encoder_config["quality_value"]]) + cmd.extend(["-profile:v", "main"]) # 提高兼容性 + logger.debug("使用纯NVENC编码器（无硬件解码，避免滤镜链问题）") + elif encoder_config["video_codec"] == "h264_amf": + # AMD AMF编码器 + cmd.extend(["-quality", encoder_config["preset"]]) + cmd.extend(["-qp_i", encoder_config["quality_value"]]) elif encoder_config["video_codec"] == "h264_qsv": + # Intel QSV编码器 cmd.extend(["-preset", encoder_config["preset"]]) - cmd.extend(["-global_quality", encoder_config["crf"]]) + cmd.extend(["-global_quality", encoder_config["quality_value"]]) + elif encoder_config["video_codec"] == "h264_videotoolbox": + # macOS VideoToolbox编码器 + cmd.extend(["-profile:v", "high"]) + cmd.extend(["-b:v", encoder_config["quality_value"]]) + else: + # 软件编码器（libx264） + cmd.extend(["-preset", encoder_config["preset"]]) + cmd.extend(["-crf", encoder_config["quality_value"]]) # 音频设置 cmd.extend(["-ar", "44100", "-ac", "2"]) - # 避免滤镜链问题的关键参数 + # 优化参数 cmd.extend(["-avoid_negative_ts", "make_zero"]) + cmd.extend(["-movflags", "+faststart"]) # 输出文件 cmd.append(output_path) @@ -194,7 +237,7 @@ def execute_ffmpeg_with_fallback( end_time: str ) -> bool: """ - 执行ffmpeg命令，带有fallback机制 + 执行ffmpeg命令，带有智能fallback机制 Args: cmd: 主要的ffmpeg命令 @@ -222,11 +265,11 @@ def execute_ffmpeg_with_fallback( if is_windows: process_kwargs["encoding"] = 'utf-8' - subprocess.run(cmd, **process_kwargs) + result = subprocess.run(cmd, **process_kwargs) # 验证输出文件 if os.path.exists(output_path) and os.path.getsize(output_path) > 0: - logger.info(f"视频裁剪成功: {timestamp}") + logger.info(f"✓ 视频裁剪成功: {timestamp}") return True else: logger.warning(f"输出文件无效: {output_path}") @@ -236,14 +279,234 @@ def execute_ffmpeg_with_fallback( error_msg = e.stderr if e.stderr else str(e) logger.warning(f"主要命令失败: {error_msg}") - # 尝试fallback命令（纯软件编码） - logger.info(f"尝试fallback方案: {timestamp}") - return try_fallback_encoding(input_path, output_path, start_time, end_time, timestamp) + # 智能错误分析 + error_type = analyze_ffmpeg_error(error_msg) + logger.debug(f"错误类型分析: {error_type}") + + # 根据错误类型选择fallback策略 + if error_type == "filter_chain_error": + logger.info(f"检测到滤镜链错误，尝试兼容性模式: {timestamp}") + return try_compatibility_fallback(input_path, output_path, start_time, end_time, timestamp) + elif error_type == "hardware_error": + logger.info(f"检测到硬件加速错误，尝试软件编码: {timestamp}") + return try_software_fallback(input_path, output_path, start_time, end_time, timestamp) + elif error_type == "encoder_error": + logger.info(f"检测到编码器错误，尝试基本编码: {timestamp}") + return try_basic_fallback(input_path, output_path, start_time, end_time, timestamp) + else: + logger.info(f"尝试通用fallback方案: {timestamp}") + return try_fallback_encoding(input_path, output_path, start_time, end_time, timestamp) + except Exception as e: logger.error(f"执行ffmpeg命令时发生异常: {str(e)}") return False +def analyze_ffmpeg_error(error_msg: str) -> str: + """ + 分析ffmpeg错误信息，返回错误类型 + + Args: + error_msg: 错误信息 + + Returns: + str: 错误类型 + """ + error_msg_lower = error_msg.lower() + + # 滤镜链错误 + if any(keyword in error_msg_lower for keyword in [ + "impossible to convert", "filter", "format", "scale", "auto_scale", + "null", "parsed_null", "reinitializing filters" + ]): + return "filter_chain_error" + + # 硬件加速错误 + if any(keyword in error_msg_lower for keyword in [ + "cuda", "nvenc", "amf", "qsv", "d3d11va", "dxva2", "videotoolbox", + "hardware", "hwaccel", "gpu", "device" + ]): + return "hardware_error" + + # 编码器错误 + if any(keyword in error_msg_lower for keyword in [ + "encoder", "codec", "h264", "libx264", "bitrate", "preset" + ]): + return "encoder_error" + + # 文件访问错误 + if any(keyword in error_msg_lower for keyword in [ + "no such file", "permission denied", "access denied", "file not found" + ]): + return "file_error" + + return "unknown_error" + + +def try_compatibility_fallback( + input_path: str, + output_path: str, + start_time: str, + end_time: str, + timestamp: str +) -> bool: + """ + 尝试兼容性fallback方案（解决滤镜链问题） + + Args: + input_path: 输入路径 + output_path: 输出路径 + start_time: 开始时间 + end_time: 结束时间 + timestamp: 时间戳 + + Returns: + bool: 是否成功 + """ + # 兼容性模式：避免所有可能的滤镜链问题 + fallback_cmd = [ + "ffmpeg", "-y", "-hide_banner", "-loglevel", "error", + "-i", input_path, + "-ss", start_time, + "-to", end_time, + "-c:v", "libx264", + "-c:a", "aac", + "-pix_fmt", "yuv420p", # 明确指定像素格式 + "-preset", "fast", + "-crf", "23", + "-ar", "44100", "-ac", "2", # 标准化音频 + "-avoid_negative_ts", "make_zero", + "-movflags", "+faststart", + "-max_muxing_queue_size", "1024", # 增加缓冲区大小 + output_path + ] + + return execute_simple_command(fallback_cmd, timestamp, "兼容性模式") + + +def try_software_fallback( + input_path: str, + output_path: str, + start_time: str, + end_time: str, + timestamp: str +) -> bool: + """ + 尝试软件编码fallback方案 + + Args: + input_path: 输入路径 + output_path: 输出路径 + start_time: 开始时间 + end_time: 结束时间 + timestamp: 时间戳 + + Returns: + bool: 是否成功 + """ + # 纯软件编码 + fallback_cmd = [ + "ffmpeg", "-y", "-hide_banner", "-loglevel", "error", + "-i", input_path, + "-ss", start_time, + "-to", end_time, + "-c:v", "libx264", + "-c:a", "aac", + "-pix_fmt", "yuv420p", + "-preset", "fast", + "-crf", "23", + "-ar", "44100", "-ac", "2", + "-avoid_negative_ts", "make_zero", + "-movflags", "+faststart", + output_path + ] + + return execute_simple_command(fallback_cmd, timestamp, "软件编码") + + +def try_basic_fallback( + input_path: str, + output_path: str, + start_time: str, + end_time: str, + timestamp: str +) -> bool: + """ + 尝试基本编码fallback方案 + + Args: + input_path: 输入路径 + output_path: 输出路径 + start_time: 开始时间 + end_time: 结束时间 + timestamp: 时间戳 + + Returns: + bool: 是否成功 + """ + # 最基本的编码参数 + fallback_cmd = [ + "ffmpeg", "-y", "-hide_banner", "-loglevel", "error", + "-i", input_path, + "-ss", start_time, + "-to", end_time, + "-c:v", "libx264", + "-c:a", "aac", + "-pix_fmt", "yuv420p", + "-preset", "ultrafast", # 最快速度 + "-crf", "28", # 稍微降低质量 + "-avoid_negative_ts", "make_zero", + output_path + ] + + return execute_simple_command(fallback_cmd, timestamp, "基本编码") + + +def execute_simple_command(cmd: List[str], timestamp: str, method_name: str) -> bool: + """ + 执行简单的ffmpeg命令 + + Args: + cmd: 命令列表 + timestamp: 时间戳 + method_name: 方法名称 + + Returns: + bool: 是否成功 + """ + try: + logger.debug(f"执行{method_name}命令: {' '.join(cmd)}") + + is_windows = os.name == 'nt' + process_kwargs = { + "stdout": subprocess.PIPE, + "stderr": subprocess.PIPE, + "text": True, + "check": True + } + + if is_windows: + process_kwargs["encoding"] = 'utf-8' + + subprocess.run(cmd, **process_kwargs) + + output_path = cmd[-1] # 输出路径总是最后一个参数 + if os.path.exists(output_path) and os.path.getsize(output_path) > 0: + logger.info(f"✓ {method_name}成功: {timestamp}") + return True + else: + logger.error(f"{method_name}失败，输出文件无效: {output_path}") + return False + + except subprocess.CalledProcessError as e: + error_msg = e.stderr if e.stderr else str(e) + logger.error(f"{method_name}失败: {error_msg}") + return False + except Exception as e: + logger.error(f"{method_name}异常: {str(e)}") + return False + + def try_fallback_encoding( input_path: str, output_path: str, @@ -252,7 +515,7 @@ def try_fallback_encoding( timestamp: str ) -> bool: """ - 尝试fallback编码方案（纯软件编码） + 尝试fallback编码方案（通用方案） Args: input_path: 输入路径 @@ -280,36 +543,7 @@ def try_fallback_encoding( output_path ] - try: - logger.debug(f"执行fallback命令: {' '.join(fallback_cmd)}") - - is_windows = os.name == 'nt' - process_kwargs = { - "stdout": subprocess.PIPE, - "stderr": subprocess.PIPE, - "text": True, - "check": True - } - - if is_windows: - process_kwargs["encoding"] = 'utf-8' - - subprocess.run(fallback_cmd, **process_kwargs) - - if os.path.exists(output_path) and os.path.getsize(output_path) > 0: - logger.info(f"Fallback编码成功: {timestamp}") - return True - else: - logger.error(f"Fallback编码失败，输出文件无效: {output_path}") - return False - - except subprocess.CalledProcessError as e: - error_msg = e.stderr if e.stderr else str(e) - logger.error(f"Fallback编码也失败: {error_msg}") - return False - except Exception as e: - logger.error(f"Fallback编码异常: {str(e)}") - return False + return execute_simple_command(fallback_cmd, timestamp, "通用Fallback") def clip_video( @@ -355,19 +589,24 @@ def clip_video( if hwaccel_type: hwaccel_args = ffmpeg_utils.get_ffmpeg_hwaccel_args() - logger.info(f"使用硬件加速: {hwaccel_type}") + hwaccel_info = ffmpeg_utils.get_ffmpeg_hwaccel_info() + logger.info(f"🚀 使用硬件加速: {hwaccel_type} ({hwaccel_info.get('message', '')})") else: - logger.info("使用软件编码") + logger.info("🔧 使用软件编码") # 获取编码器配置 encoder_config = get_safe_encoder_config(hwaccel_type) logger.debug(f"编码器配置: {encoder_config}") - # 存储裁剪结果 + # 统计信息 + total_clips = len(tts_result) result = {} failed_clips = [] + success_count = 0 - for item in tts_result: + logger.info(f"📹 开始裁剪视频，总共{total_clips}个片段") + + for i, item in enumerate(tts_result, 1): _id = item.get("_id", item.get("timestamp", "unknown")) timestamp = item["timestamp"] start_time, _ = parse_timestamp(timestamp) @@ -397,7 +636,7 @@ def clip_video( ) # 执行FFmpeg命令 - logger.info(f"裁剪视频片段: {timestamp} -> {ffmpeg_start_time}到{ffmpeg_end_time}") + logger.info(f"📹 [{i}/{total_clips}] 裁剪视频片段: {timestamp} -> {ffmpeg_start_time}到{ffmpeg_end_time}") success = execute_ffmpeg_with_fallback( ffmpeg_cmd, @@ -410,17 +649,26 @@ def clip_video( if success: result[_id] = output_path + success_count += 1 + logger.info(f"✅ [{i}/{total_clips}] 片段裁剪成功: {timestamp}") else: failed_clips.append(timestamp) - logger.error(f"裁剪视频片段失败: {timestamp}") + logger.error(f"❌ [{i}/{total_clips}] 片段裁剪失败: {timestamp}") + # 最终统计 + logger.info(f"📊 视频裁剪完成: 成功 {success_count}/{total_clips}, 失败 {len(failed_clips)}") + # 检查是否有失败的片段 if failed_clips: - logger.warning(f"以下片段裁剪失败: {failed_clips}") - if len(failed_clips) == len(tts_result): + logger.warning(f"⚠️ 以下片段裁剪失败: {failed_clips}") + if len(failed_clips) == total_clips: raise RuntimeError("所有视频片段裁剪都失败了，请检查视频文件和ffmpeg配置") + elif len(failed_clips) > total_clips / 2: + logger.warning(f"⚠️ 超过一半的片段裁剪失败 ({len(failed_clips)}/{total_clips})，请检查硬件加速配置") - logger.info(f"视频裁剪完成，成功: {len(result)}, 失败: {len(failed_clips)}") + if success_count > 0: + logger.info(f"🎉 视频裁剪任务完成! 输出目录: {output_dir}") + return result diff --git a/app/services/merger_video.py b/app/services/merger_video.py index 026a47c..75c686b 100644 --- a/app/services/merger_video.py +++ b/app/services/merger_video.py @@ -137,6 +137,9 @@ def process_single_video( ) -> str: """ 处理单个视频：调整分辨率、帧率等 + + 重要修复：避免在视频滤镜处理时使用CUDA硬件解码， + 因为这会导致滤镜链格式转换错误。使用纯NVENC编码器获得最佳兼容性。 Args: input_path: 输入视频路径 @@ -178,24 +181,11 @@ def process_single_video( logger.warning(f"视频探测出错，禁用硬件加速: {str(e)}") hwaccel = None - # 添加硬件加速参数（使用新的智能检测机制） - if hwaccel: - try: - # 使用新的硬件加速检测API - hwaccel_args = ffmpeg_utils.get_ffmpeg_hwaccel_args() - if hwaccel_args: - command.extend(hwaccel_args) - logger.debug(f"应用硬件加速参数: {hwaccel_args}") - else: - logger.info("硬件加速不可用，将使用软件编码") - hwaccel = False # 标记为不使用硬件加速 - except Exception as e: - logger.warning(f"应用硬件加速参数时出错: {str(e)}，将使用软件编码") - hwaccel = False # 标记为不使用硬件加速 - # 重置命令，移除可能添加了一半的硬件加速参数 - command = ['ffmpeg', '-y'] - - # 输入文件 + # 关键修复：对于涉及滤镜处理的场景，不使用CUDA硬件解码 + # 这避免了 "Impossible to convert between the formats" 错误 + # 我们将只使用纯NVENC编码器来获得硬件加速优势 + + # 输入文件（不添加硬件解码参数） command.extend(['-i', input_path]) # 处理音频 @@ -218,27 +208,36 @@ def process_single_video( '-r', '30', # 设置帧率为30fps ]) - # 选择编码器 - 使用新的智能编码器选择 - encoder = ffmpeg_utils.get_optimal_ffmpeg_encoder() - - if hwaccel and encoder != "libx264": - logger.info(f"使用硬件编码器: {encoder}") - command.extend(['-c:v', encoder]) - - # 根据编码器类型添加特定参数 - if "nvenc" in encoder: - command.extend(['-preset', 'p4', '-profile:v', 'high']) - elif "videotoolbox" in encoder: - command.extend(['-profile:v', 'high']) - elif "qsv" in encoder: - command.extend(['-preset', 'medium']) - elif "vaapi" in encoder: - command.extend(['-profile', '100']) - elif "amf" in encoder: - command.extend(['-quality', 'balanced']) - else: - command.extend(['-preset', 'medium', '-profile:v', 'high']) - else: + # 关键修复：选择编码器时优先使用纯NVENC（无硬件解码） + if hwaccel: + try: + # 检查是否为NVIDIA硬件加速 + hwaccel_info = ffmpeg_utils.detect_hardware_acceleration() + if hwaccel_info.get("type") in ["cuda", "nvenc"] and hwaccel_info.get("encoder") == "h264_nvenc": + # 使用纯NVENC编码器（最佳兼容性） + logger.info("使用纯NVENC编码器（避免滤镜链问题）") + command.extend(['-c:v', 'h264_nvenc']) + command.extend(['-preset', 'medium', '-cq', '23', '-profile:v', 'main']) + else: + # 其他硬件编码器 + encoder = ffmpeg_utils.get_optimal_ffmpeg_encoder() + logger.info(f"使用硬件编码器: {encoder}") + command.extend(['-c:v', encoder]) + + # 根据编码器类型添加特定参数 + if "amf" in encoder: + command.extend(['-quality', 'balanced']) + elif "qsv" in encoder: + command.extend(['-preset', 'medium']) + elif "videotoolbox" in encoder: + command.extend(['-profile:v', 'high']) + else: + command.extend(['-preset', 'medium', '-profile:v', 'high']) + except Exception as e: + logger.warning(f"硬件编码器检测失败: {str(e)}，将使用软件编码") + hwaccel = None + + if not hwaccel: logger.info("使用软件编码器(libx264)") command.extend(['-c:v', 'libx264', '-preset', 'medium', '-profile:v', 'high']) diff --git a/app/utils/ffmpeg_utils.py b/app/utils/ffmpeg_utils.py index 538af7a..15f1077 100644 --- a/app/utils/ffmpeg_utils.py +++ b/app/utils/ffmpeg_utils.py @@ -469,167 +469,171 @@ def _detect_macos_acceleration(supported_hwaccels: str) -> None: def _detect_windows_acceleration(supported_hwaccels: str) -> None: """ - 检测Windows系统的硬件加速 - + 检测Windows系统的硬件加速 - 基于实际测试结果优化 + + 重要发现：CUDA硬件解码在视频裁剪场景下会导致滤镜链错误， + 因此优先使用纯NVENC编码器方案，既保证性能又确保兼容性。 + Args: supported_hwaccels: FFmpeg支持的硬件加速器列表 """ global _FFMPEG_HW_ACCEL_INFO - + # 在Windows上，首先检查显卡信息 gpu_info = _get_windows_gpu_info() - - # 检查是否为AMD显卡 - if 'amd' in gpu_info.lower() or 'radeon' in gpu_info.lower(): - logger.info("检测到AMD显卡，为避免兼容性问题，将使用软件编码") - _FFMPEG_HW_ACCEL_INFO["message"] = "检测到AMD显卡，为避免兼容性问题，将使用软件编码" - return - + logger.debug(f"Windows GPU信息: {gpu_info}") + # 检查是否为Intel集成显卡 is_intel_integrated = False if 'intel' in gpu_info.lower() and ('hd graphics' in gpu_info.lower() or 'uhd graphics' in gpu_info.lower()): logger.info("检测到Intel集成显卡") is_intel_integrated = True - - # 检测NVIDIA CUDA支持 - if 'cuda' in supported_hwaccels and 'nvidia' in gpu_info.lower(): - # 添加调试日志 - logger.debug(f"Windows检测到NVIDIA显卡，尝试CUDA加速") + + # 1. 优先检测NVIDIA硬件加速 - 基于实际测试的最佳方案 + if 'nvidia' in gpu_info.lower() or 'geforce' in gpu_info.lower() or 'quadro' in gpu_info.lower(): + logger.info("检测到NVIDIA显卡，开始测试硬件加速") + + # 检查NVENC编码器是否可用 try: - # 先检查NVENC编码器是否可用，使用UTF-8编码 encoders_cmd = subprocess.run( ["ffmpeg", "-hide_banner", "-encoders"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, encoding='utf-8', text=True, check=False + stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False ) has_nvenc = "h264_nvenc" in encoders_cmd.stdout.lower() logger.debug(f"NVENC编码器检测结果: {'可用' if has_nvenc else '不可用'}") - - # 测试CUDA硬件加速，使用UTF-8编码 - test_cmd = subprocess.run( - ["ffmpeg", "-hwaccel", "cuda", "-i", "NUL", "-f", "null", "-t", "0.1", "-"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, encoding='utf-8', text=True, check=False - ) - - # 记录详细的返回信息以便调试 - logger.debug(f"CUDA测试返回码: {test_cmd.returncode}") - logger.debug(f"CUDA测试错误输出: {test_cmd.stderr[:200]}..." if len(test_cmd.stderr) > 200 else f"CUDA测试错误输出: {test_cmd.stderr}") - - if test_cmd.returncode == 0 or has_nvenc: - _FFMPEG_HW_ACCEL_INFO["available"] = True - _FFMPEG_HW_ACCEL_INFO["type"] = "cuda" - _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_nvenc" - _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = ["-hwaccel", "cuda"] - _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = True - return - - # 如果上面的测试失败，尝试另一种方式，使用UTF-8编码 - test_cmd2 = subprocess.run( - ["ffmpeg", "-hide_banner", "-loglevel", "error", "-hwaccel", "cuda", "-hwaccel_output_format", "cuda", "-i", "NUL", "-f", "null", "-t", "0.1", "-"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, encoding='utf-8', text=True, check=False - ) - - if test_cmd2.returncode == 0: - _FFMPEG_HW_ACCEL_INFO["available"] = True - _FFMPEG_HW_ACCEL_INFO["type"] = "cuda" - _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_nvenc" - _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = ["-hwaccel", "cuda", "-hwaccel_output_format", "cuda"] - _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = True - return - except Exception as e: - logger.debug(f"测试CUDA失败: {str(e)}") - - # 检测Intel QSV支持（如果是Intel显卡） - if 'qsv' in supported_hwaccels and 'intel' in gpu_info.lower(): - try: - test_cmd = subprocess.run( - ["ffmpeg", "-hwaccel", "qsv", "-i", "/dev/null", "-f", "null", "-"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, text=True, check=False - ) - if test_cmd.returncode == 0: - _FFMPEG_HW_ACCEL_INFO["available"] = True - _FFMPEG_HW_ACCEL_INFO["type"] = "qsv" - _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_qsv" - _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = ["-hwaccel", "qsv"] - _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = not is_intel_integrated - return - except Exception as e: - logger.debug(f"测试QSV失败: {str(e)}") - - # 检测D3D11VA支持 - if 'd3d11va' in supported_hwaccels: - logger.debug("Windows尝试D3D11VA加速") - try: - test_cmd = subprocess.run( - ["ffmpeg", "-hwaccel", "d3d11va", "-i", "NUL", "-f", "null", "-t", "0.1", "-"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, encoding='utf-8', text=True, check=False - ) - - # 记录详细的返回信息以便调试 - logger.debug(f"D3D11VA测试返回码: {test_cmd.returncode}") - - if test_cmd.returncode == 0: - _FFMPEG_HW_ACCEL_INFO["available"] = True - _FFMPEG_HW_ACCEL_INFO["type"] = "d3d11va" - _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264" # D3D11VA只用于解码，编码仍使用软件编码器 - _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = ["-hwaccel", "d3d11va"] - _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = not is_intel_integrated - return - except Exception as e: - logger.debug(f"测试D3D11VA失败: {str(e)}") - - # 检测DXVA2支持 - if 'dxva2' in supported_hwaccels: - logger.debug("Windows尝试DXVA2加速") - try: - test_cmd = subprocess.run( - ["ffmpeg", "-hwaccel", "dxva2", "-i", "NUL", "-f", "null", "-t", "0.1", "-"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, encoding='utf-8', text=True, check=False - ) - - # 记录详细的返回信息以便调试 - logger.debug(f"DXVA2测试返回码: {test_cmd.returncode}") - - if test_cmd.returncode == 0: - _FFMPEG_HW_ACCEL_INFO["available"] = True - _FFMPEG_HW_ACCEL_INFO["type"] = "dxva2" - _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264" # DXVA2只用于解码，编码仍使用软件编码器 - _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = ["-hwaccel", "dxva2"] - _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = not is_intel_integrated - return - except Exception as e: - logger.debug(f"测试DXVA2失败: {str(e)}") - - # 如果检测到NVIDIA显卡但前面的测试都失败，尝试直接使用NVENC编码器 - if 'nvidia' in gpu_info.lower(): - logger.debug("Windows检测到NVIDIA显卡，尝试直接使用NVENC编码器") - try: - # 检查NVENC编码器是否可用，使用UTF-8编码 - encoders_cmd = subprocess.run( - ["ffmpeg", "-hide_banner", "-encoders"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, encoding='utf-8', text=True, check=False - ) - - if "h264_nvenc" in encoders_cmd.stdout.lower(): - logger.debug("NVENC编码器可用，尝试直接使用") - # 测试NVENC编码器，使用UTF-8编码 - test_cmd = subprocess.run( - ["ffmpeg", "-f", "lavfi", "-i", "color=c=black:s=640x360:r=30", "-c:v", "h264_nvenc", "-t", "0.1", "-f", "null", "-"], - stderr=subprocess.PIPE, stdout=subprocess.PIPE, encoding='utf-8', text=True, check=False - ) - - logger.debug(f"NVENC编码器测试返回码: {test_cmd.returncode}") - + + if has_nvenc: + # 优先方案：纯NVENC编码器（测试证明最兼容） + logger.debug("测试纯NVENC编码器（推荐方案，避免滤镜链问题）") + test_cmd = subprocess.run([ + "ffmpeg", "-hide_banner", "-loglevel", "error", + "-f", "lavfi", "-i", "testsrc=duration=0.1:size=640x480:rate=30", + "-c:v", "h264_nvenc", "-preset", "medium", "-cq", "23", + "-pix_fmt", "yuv420p", "-f", "null", "-" + ], stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False) + if test_cmd.returncode == 0: _FFMPEG_HW_ACCEL_INFO["available"] = True - _FFMPEG_HW_ACCEL_INFO["type"] = "nvenc" + _FFMPEG_HW_ACCEL_INFO["type"] = "nvenc" # 使用nvenc类型标识纯编码器 _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_nvenc" - _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = [] # 不使用hwaccel参数，直接使用编码器 + _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = [] # 不使用硬件解码参数 _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = True + _FFMPEG_HW_ACCEL_INFO["message"] = "纯NVENC编码器（最佳兼容性）" + logger.info("✓ 纯NVENC编码器测试成功") return + + # 备用方案：如果需要的话，可以测试CUDA硬件解码（但不推荐用于视频裁剪） + if 'cuda' in supported_hwaccels: + logger.debug("测试CUDA硬件解码（仅用于非裁剪场景）") + test_cmd = subprocess.run([ + "ffmpeg", "-hide_banner", "-loglevel", "error", + "-hwaccel", "cuda", "-hwaccel_output_format", "cuda", + "-f", "lavfi", "-i", "testsrc=duration=0.1:size=640x480:rate=30", + "-c:v", "h264_nvenc", "-preset", "medium", "-cq", "23", + "-pix_fmt", "yuv420p", "-f", "null", "-" + ], stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False) + + if test_cmd.returncode == 0: + _FFMPEG_HW_ACCEL_INFO["available"] = True + _FFMPEG_HW_ACCEL_INFO["type"] = "cuda" # 保留cuda类型用于特殊场景 + _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_nvenc" + _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = ["-hwaccel", "cuda", "-hwaccel_output_format", "cuda"] + _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = True + _FFMPEG_HW_ACCEL_INFO["message"] = "CUDA+NVENC（限特殊场景使用）" + _FFMPEG_HW_ACCEL_INFO["fallback_available"] = True + _FFMPEG_HW_ACCEL_INFO["fallback_encoder"] = "h264_nvenc" + logger.info("✓ CUDA+NVENC硬件加速测试成功（备用方案）") + return + except Exception as e: - logger.debug(f"测试NVENC编码器失败: {str(e)}") - - _FFMPEG_HW_ACCEL_INFO["message"] = f"Windows系统未检测到可用的硬件加速，显卡信息: {gpu_info}" + logger.debug(f"NVIDIA硬件加速测试失败: {str(e)}") + + # 2. 检测AMD硬件加速 + if 'amd' in gpu_info.lower() or 'radeon' in gpu_info.lower(): + logger.info("检测到AMD显卡，开始测试硬件加速") + + # 检查AMF编码器是否可用 + try: + encoders_cmd = subprocess.run( + ["ffmpeg", "-hide_banner", "-encoders"], + stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False + ) + has_amf = "h264_amf" in encoders_cmd.stdout.lower() + logger.debug(f"AMF编码器检测结果: {'可用' if has_amf else '不可用'}") + + if has_amf: + # 测试AMF编码器 + logger.debug("测试AMF编码器") + test_cmd = subprocess.run([ + "ffmpeg", "-hide_banner", "-loglevel", "error", + "-f", "lavfi", "-i", "testsrc=duration=0.1:size=640x480:rate=30", + "-c:v", "h264_amf", "-quality", "balanced", "-qp_i", "23", + "-pix_fmt", "yuv420p", "-f", "null", "-" + ], stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False) + + if test_cmd.returncode == 0: + _FFMPEG_HW_ACCEL_INFO["available"] = True + _FFMPEG_HW_ACCEL_INFO["type"] = "amf" + _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_amf" + _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = [] + _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = True + _FFMPEG_HW_ACCEL_INFO["message"] = "AMD AMF编码器" + logger.info("✓ AMD AMF编码器测试成功") + return + + except Exception as e: + logger.debug(f"AMD硬件加速测试失败: {str(e)}") + + # 3. 检测Intel硬件加速 + if 'intel' in gpu_info.lower() and 'qsv' in supported_hwaccels: + logger.info("检测到Intel显卡，开始测试硬件加速") + + try: + encoders_cmd = subprocess.run( + ["ffmpeg", "-hide_banner", "-encoders"], + stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False + ) + has_qsv = "h264_qsv" in encoders_cmd.stdout.lower() + logger.debug(f"QSV编码器检测结果: {'可用' if has_qsv else '不可用'}") + + if has_qsv: + # 测试QSV编码器 + logger.debug("测试QSV编码器") + test_cmd = subprocess.run([ + "ffmpeg", "-hide_banner", "-loglevel", "error", + "-f", "lavfi", "-i", "testsrc=duration=0.1:size=640x480:rate=30", + "-c:v", "h264_qsv", "-preset", "medium", "-global_quality", "23", + "-pix_fmt", "yuv420p", "-f", "null", "-" + ], stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False) + + if test_cmd.returncode == 0: + _FFMPEG_HW_ACCEL_INFO["available"] = True + _FFMPEG_HW_ACCEL_INFO["type"] = "qsv" + _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_qsv" + _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = [] + _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = not is_intel_integrated + _FFMPEG_HW_ACCEL_INFO["message"] = "Intel QSV编码器" + logger.info("✓ Intel QSV编码器测试成功") + return + + except Exception as e: + logger.debug(f"Intel硬件加速测试失败: {str(e)}") + + # 4. 如果没有硬件编码器，使用软件编码 + logger.info("未检测到可用的硬件编码器，使用软件编码") + _FFMPEG_HW_ACCEL_INFO["available"] = False + _FFMPEG_HW_ACCEL_INFO["type"] = "software" + _FFMPEG_HW_ACCEL_INFO["encoder"] = "libx264" + _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = [] + _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = False + _FFMPEG_HW_ACCEL_INFO["message"] = "使用软件编码" def _detect_linux_acceleration(supported_hwaccels: str) -> None: @@ -997,9 +1001,15 @@ def force_software_encoding() -> None: def reset_hwaccel_detection() -> None: """ 重置硬件加速检测结果，强制重新检测 + + 这在以下情况下很有用： + 1. 驱动程序更新后 + 2. 系统配置改变后 + 3. 需要重新测试硬件加速时 """ global _FFMPEG_HW_ACCEL_INFO - + + logger.info("🔄 重置硬件加速检测，将重新检测...") _FFMPEG_HW_ACCEL_INFO = { "available": False, "type": None, @@ -1014,4 +1024,94 @@ def reset_hwaccel_detection() -> None: "tested_methods": [] } - logger.info("已重置硬件加速检测结果") + +def test_nvenc_directly() -> bool: + """ + 直接测试NVENC编码器是否可用（无硬件解码） + + Returns: + bool: NVENC是否可用 + """ + try: + logger.info("🧪 直接测试NVENC编码器...") + + # 测试纯NVENC编码器 + test_cmd = subprocess.run([ + "ffmpeg", "-hide_banner", "-loglevel", "error", + "-f", "lavfi", "-i", "testsrc=duration=1:size=640x480:rate=30", + "-c:v", "h264_nvenc", "-preset", "fast", "-profile:v", "main", + "-pix_fmt", "yuv420p", "-t", "1", "-f", "null", "-" + ], stderr=subprocess.PIPE, stdout=subprocess.PIPE, + encoding='utf-8', text=True, check=False) + + if test_cmd.returncode == 0: + logger.info("✅ NVENC编码器测试成功！") + return True + else: + logger.warning(f"❌ NVENC编码器测试失败: {test_cmd.stderr}") + return False + + except Exception as e: + logger.error(f"NVENC测试异常: {str(e)}") + return False + + +def force_use_nvenc_pure() -> None: + """ + 强制使用纯NVENC编码器模式 + + 当自动检测失败但你确定NVENC可用时使用 + """ + global _FFMPEG_HW_ACCEL_INFO + + logger.info("🎯 强制启用纯NVENC编码器模式...") + + # 先测试NVENC是否真的可用 + if test_nvenc_directly(): + _FFMPEG_HW_ACCEL_INFO["available"] = True + _FFMPEG_HW_ACCEL_INFO["type"] = "nvenc_pure" + _FFMPEG_HW_ACCEL_INFO["encoder"] = "h264_nvenc" + _FFMPEG_HW_ACCEL_INFO["hwaccel_args"] = [] + _FFMPEG_HW_ACCEL_INFO["is_dedicated_gpu"] = True + _FFMPEG_HW_ACCEL_INFO["message"] = "强制启用纯NVENC编码器" + logger.info("✅ 已强制启用纯NVENC编码器模式") + else: + logger.error("❌ NVENC编码器不可用，无法强制启用") + + +def get_hwaccel_status() -> Dict[str, any]: + """ + 获取当前硬件加速状态的详细信息 + + Returns: + Dict: 硬件加速状态信息 + """ + hwaccel_info = get_ffmpeg_hwaccel_info() + + status = { + "available": hwaccel_info.get("available", False), + "type": hwaccel_info.get("type", "software"), + "encoder": hwaccel_info.get("encoder", "libx264"), + "message": hwaccel_info.get("message", ""), + "is_dedicated_gpu": hwaccel_info.get("is_dedicated_gpu", False), + "platform": platform.system(), + "gpu_vendor": detect_gpu_vendor(), + "ffmpeg_available": check_ffmpeg_installation() + } + + return status + + +# 自动重置检测（在模块导入时执行） +def _auto_reset_on_import(): + """模块导入时自动重置硬件加速检测""" + try: + # 检查是否需要重置（比如检测到配置变化） + current_platform = platform.system() + if _FFMPEG_HW_ACCEL_INFO.get("platform") != current_platform: + reset_hwaccel_detection() + except Exception as e: + logger.debug(f"自动重置检测失败: {str(e)}") + +# 执行自动重置 +_auto_reset_on_import() diff --git a/config.example.toml b/config.example.toml index a05cb30..270ed0d 100644 --- a/config.example.toml +++ b/config.example.toml @@ -1,5 +1,5 @@ [app] - project_version="0.6.5" + project_version="0.6.6" # 支持视频理解的大模型提供商 # gemini (谷歌, 需要 VPN) # siliconflow (硅基流动) diff --git a/project_version b/project_version index e0ea44c..bf21f52 100644 --- a/project_version +++ b/project_version @@ -1 +1 @@ -0.6.5 \ No newline at end of file +0.6.6 \ No newline at end of file