AI Short-Drama Director Says Manual Prompts Can Be More Efficient for Sensitive Scenes
In a first-person essay published by the WeChat public account Future浮标, an overseas AI short-drama director who says they have worked on more than ten productions describes a recurring problem: multimodal models often refuse to write storyboards for suggestive scenes, while models that accept them may produce weaker storyboards. The director says splitting the work between models is less efficient than manually writing prompts, and estimates that suggestive scenes consume roughly three times as many tokens as other scenes of comparable length. The essay also says Facebook, Google and TikTok tightened enforcement against non-compliant material from late September and that DramaBox was temporarily removed from iOS, but cites no official platform statements explaining either development.
The account highlights a practical limitation in AI short-drama production: models with stronger storyboarding ability may apply stricter moderation, while models with looser moderation may produce weaker storyboards. This trade-off can make manually written prompts more efficient for suggestive scenes.

本文作者是一位做出海AI短剧的一线导演,经手过十多部作品。我们约他本来是想写写最近的工作心得和难点卡点,结果他说,最难的是擦边戏。大模型写擦边戏不是这里不行就是那里不行,最高效的方法竟然是手搓。但手搓女频擦边戏对于一位年轻男导演来说并非易事,而就在他千辛万苦地学会了之后,平台又开始封禁了。连擦边这块人类自留地......
本文来自微信公众号:Future浮标,作者:敬修己,编辑:何润萱,题图来自:AI生成
本文作者是一位做出海AI短剧的一线导演,经手过十多部作品。我们约他本来是想写写最近的工作心得和难点卡点,结果他说,最难的是擦边戏。大模型写擦边戏不是这里不行就是那里不行,最高效的方法竟然是手搓。但手搓女频擦边戏对于一位年轻男导演来说并非易事,而就在他千辛万苦地学会了之后,平台又开始封禁了。连擦边这块人类自留地也岌岌可危了。
#GPT#Claude#DeepSeek#Grok#Seedance#豆包#DramaBox#Facebook#Google#TikTok