一个简单的视频合成任务,没有想到deepseek,豆包,qwen3.0等国内所谓顶级的模型居然搞了5个小时没有搞出来。实在没办法上梯子到GPT最普通的模型里输入同样的问题。30秒搞定,计算了一下国产众多模型浪费了我多少时间。国产仍需努力,别再天天吹牛B了。
我的问题:
合成的效果:
GPT代码:
import subprocess
import osfile_name = 'name'
# 配置参数
SUBTITLE_FILE = f"{file_name}.srt"
VIDEO_INPUT = "bmg.mp4"
AUDIO_INPUT = f"{file_name}.mp3"
OUTPUT_VIDEO = f"{file_name}.mp4"
TITLE_IMAGE = f"{file_name}.png"
ffmpeg_path = r"D:\03必备软件\ffmpeg-2025-06-08\bin\ffmpeg.exe"# 字幕样式参数
SUBTITLE_Y_OFFSET = 40 # 字幕距离底部
SUBTITLE_FONT_SIZE = 14 # 字体大小# ASS 样式参数字符串
subtitle_style = (f"force_style='Fontsize={SUBTITLE_FONT_SIZE},MarginV={SUBTITLE_Y_OFFSET}'"
)# 如果文件已存在,跳过
if os.path.exists(OUTPUT_VIDEO):print(f"{OUTPUT_VIDEO} 已存在,跳过合成。")
else:# 构建 FFmpeg 命令command = [ffmpeg_path,"-y","-i", VIDEO_INPUT,"-i", AUDIO_INPUT,"-i", TITLE_IMAGE,"-filter_complex",(f"[0:v]scale=500:900[bg];"f"[2:v]scale=500:-1[img];"f"[bg][img]overlay=0:0[temp];"f"[temp]subtitles='{SUBTITLE_FILE}':{subtitle_style}[v]"),"-map", "[v]","-map", "1:a",# ✅ 编码参数位置正确"-c:v", "libx264","-preset", "veryfast","-threads", "0","-c:a", "aac","-shortest",OUTPUT_VIDEO]subprocess.run(command)print(f"{OUTPUT_VIDEO} 生成完成。")