Flow Kit

The problem

AI video generation is powerful but fiddly. A single good clip takes dozens of manual steps across separate tools for references, images, video, narration, and publishing — and characters drift from shot to shot because nothing enforces consistency. We wanted to compress that whole pipeline into one command an AI CLI could run, so producing a finished video felt like invoking a single skill instead of babysitting ten tools.

What we built

Flow Kit is an open-source toolkit that turns your AI coding agent into a cinematic video studio. It packages the full production pipeline as 36 fk-* skills — plain Markdown recipes that Claude Code, Codex, Antigravity, and Muse can all run — taking a story idea to a published YouTube video in one session, with a reference system that keeps characters consistent across scenes. What used to be an afternoon of wrangling separate tools becomes a single session inside the agent you already use.

• 36 fk-* skills for references, images, video, TTS, and publishing
• Reference system keeps characters consistent across scenes
• Veo 3.1 video and OmniVoice TTS in 600+ languages
• Runs in Claude Code, Codex, Antigravity, or Muse

How it is built

Flow Kit drives Google Flow end to end from inside the AI coding agent the user already has. Each /fk-* command is a composable skill; chained together they take a brief through reference creation, image generation, scene-chained video on Veo 3.1, voice-cloned narration via OmniVoice TTS, and upload — the exact pipeline ISEMI runs for its own media work.

• Veo 3.1 for scene-chained video generation
• OmniVoice TTS for narration in 600+ languages
• Reference-based character consistency across shots
• Driven from Claude Code, Codex, Antigravity, or Muse through Google Flow

The result

Flow Kit is open source on GitHub, where it has passed 960 stars, and it is the production engine behind our AI video and image service — proof that we do not just call a model, we engineer the repeatable pipeline around it. Teams can read the source, run it themselves, and see exactly how a pipeline-first approach beats one-off generation.

• Open source on GitHub, with 960+ stars
• One command takes a brief all the way to a published video
• Characters stay consistent across every scene
• The production engine behind our AI video service