Google's gemini-api CLI lets your coding agent make video, music, voiceover and SRTs with one command each

Google's AI Studio blog post is dated Oct 19, and Dmitry Lyalin's launch post on Oct 22 passed 1,100 likes and 600 bookmarks. The repo (google-gemini/gemini-api-cli, Apache-2.0) was at v0.4.3 and about 116 stars when we checked. Google calls it an experimental preview. It installs with a curl script on macOS/Linux, a PowerShell line on Windows, or `go install github.com/google-gemini/gemini-api-cli/cmd/gemini-api@latest`, then reads your GEMINI_API_KEY.

The video-relevant commands map onto a full asset pipeline. `gemini-api video` uses Gemini Omni 1.1 Flash. `image` uses Nano Banana 2.1 for storyboards and thumbnails. `tts` uses Gemini 3.8 Flash TTS and supports multi-speaker dialogue via --multi-speaker "Alice=Kore,Bob=Puck". `music` uses Lyria 3.5 for background beds. `transcribe -i file.mp3 --format srt` returns timestamped subtitles, and `analyze` can read documents, images, audio or YouTube videos. Each media command writes the file and prints only the path, so your agent's context never fills with base64.

The agent-friendly details matter. Every command describes itself offline through --help, --usage (machine-readable), --schema and --dry-run. Output can be JSON or a compact format with inline --jq filtering. Google's suggestion is a single line in AGENTS.md telling your agent to use the gemini-api CLI for media. For a Remotion or HyperFrames project, that gives you voiceover, a music bed, b-roll and burned-in captions without leaving the agent loop.

Not tested by us yet. It is a v0.4 preview, and Google says the command surface is still changing. Every call bills your Gemini API key at normal model rates. Several replies under the launch post found the name confusing next to the existing Gemini CLI, which is a different tool. Pin a version in your scripts.