Claude Opus 5.5 and GPT-6 Astra push into video, testing video models' moats
An industry analysis describes two ways general-purpose models are entering AI video from upstream: Anthropic's Claude Opus 5.5 (said to have launched Sept 22) writes front-end code — SVG, Canvas, Three.js, Remotion — renders frames and composites them with FFmpeg, spawning a creator-dubbed "Vibe Video" style; OpenAI's GPT-6 Astra (said to have launched Sept 3) acts as director-producer inside Blender, Unreal Engine and DaVinci Resolve, handling storyboards, previsualisation and rough cuts before handing rendering to video models such as Seedance. The piece says ByteDance's Jianying Hub, MiniMax Design and Kuaishou's Kling AI 4.0 and ikli are meanwhile pushing deeper into end-to-end creation workflows to defend their positions. View counts, API costs and an "80%-90% of AGI" claim in the article come from creators and bloggers and are unverified.
It suggests the AI-video fight may shift from raw image quality to who owns the creative workflow's entry point and orchestration.
编者按：本文转自AI价值官，作者星野，编者美圻，创业邦经授权转载。
过去一个月，AI视频领域最激动人心的讨论热潮，并非围绕Seedance、Minmax、可灵等专业视频模型，而是“越界”闯入的通用大模型。
Claude Opus 5.5 靠代码渲染视频，短短数日内催生出大量科普动画、创意短片和音乐MV，在X等平台涌现出多条爆款，甚至带火了一种新的创作范式：Vibe Video。
GPT-6 Astra则扮演起了“AI导演”，从分镜规划做到白模预演，通过深入 Blender、Unreal Engine、DaVinci Resolve 等专业软件，承担场景搭建、镜头预演、剪辑执行等工作，逐渐开始占据视频生成的上游。
两者走的是截然不同的两条路：一条是代码渲染的程序化创意，一条是空间规划的影视工业化。它们都没有在画质上正面硬刚，却从上游入口和细分场景切入，悄悄改写着AI视频的能力边界。
#Anthropic#Claude Opus 5.5#OpenAI#GPT-6 Astra#Seedance#MiniMax#Kling AI#ByteDance#Kuaishou#DaVinci Resolve#Blender#Unreal Engine