Directing an AI Film Crew with Gemini 4 and Omni 1.2
An author describes a seven-stage workflow for producing a narrative short film with Gemini 4, Omni 1.2, local speech tools, Python and ffmpeg. The stages are idea development, scripting, storyboarding, video-prompt writing, browser-based generation, redubbing with local models, and coded post-production with a final review. Gemini 4 is used as the coordinator and programmer, while Omni 1.2 generates short video clips in Google Flow, VoxCPM2 and Whisper handle dubbing and speech checks, and Python and ffmpeg support editing and visual overlays. The author also provides reusable prompts, role assignments and examples of common problems, but the reported workflow and results reflect one author’s experience.
The report offers a concrete, end-to-end example of coordinating AI video generation, dubbing, subtitles, quality control and coded post-production, together with reusable prompts and explicit production roles.
一个人指挥一支 AI 剧组:Gemini 4 + Omni 1.2 剧情短片制作实录
脑子里只有一个模糊的点子,怎么用 AI 把它拍成一支画面稳又好笑的完整成片?这篇文章就聊透这件事。
现在 AI 视频很火,特别是 Omni 1.2 出来之后,大家都在聊怎么写提示词、怎么抽卡。
但我做完几支片子下来,感触很深。会用视频生成工具,反而是整件事里最简单的部分。
真正让人头疼的是后面的全套制作:怎么把一句话点子扩成有反转的剧本,怎么让五六个镜头里的人物长相和衣服不串戏,怎么解决视频模型说中文容易吞尾音、读错多音字的老毛病,以及怎么用代码把后期字幕、界面和混音做干净。
前几天我用 Gemini 4 配合 Omni 1.2,做了一系列古装监控视角的西游记喜剧(比如《真假美猴王》和《西梁国留婚》)。
#Gemini 4#Omni 1.2#VoxCPM2#Whisper#Python#ffmpeg#Google Flow#Antigravity