Claude Code, a version of the AI model designed for local computer execution, can now perform video editing tasks. The system operates by processing a transcript of the video rather than the visual stream itself. It identifies filler words, pauses, and repeated takes based on text, then generates an edit decision list (EDL) for the FFmpeg tool to assemble the final file.
The functionality is enabled through an open-source tool called video-use, released under the MIT license in spring 2026. While the project is compatible with various agents like OpenAI's Codex, it is primarily optimized for Claude Code. Users can also utilize freecut, a fork of the project that allows for local transcription using Whisper, ensuring that video data remains on the user's machine.
The AI can handle color grading, subtitle generation, and programmatic animation using tools like Remotion and HyperFrames. However, the system struggles with visual-based editing, such as timing cuts to specific gestures or facial expressions, as it only analyzes individual frames. Users are advised to provide detailed text prompts to achieve specific results.
Operating Claude Code requires either a Claude Pro subscription or an API key with pay-per-token billing. For users in Russia, access to models like Claude Opus 5.5 and Claude Sonnet 5 via API is available through SpeShu.AI, which supports ruble payments. The cost of processing scales with the length of the video, as longer recordings consume more tokens.
