← all repositories
crisng95/flowkit

A YouTube video factory bolted onto Google Flow via a Chrome tab

FlowKit drives Google's Flow video generation end-to-end — story to upload — by routing its RPCs through a signed-in browser extension, because that's the only place they can be signed.

★1.1k stars Python AgentsImage · Video · Audio
flowkit
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does

FlowKit is a Python pipeline (FastAPI + SQLite) that turns a story concept into a finished, YouTube-ready video: reference images for characters, scene images, 8-second clips, TTS narration, concatenation, thumbnails, and SEO metadata. The catch is that Google Flow has no public API — so a Chrome MV3 extension acts as a bridge, running Flow’s batchexecute RPCs inside a signed-in flow.google.com tab and talking to the Python agent over WebSocket. It also ships 36 skills as plain Markdown recipes so AI coding agents (Claude Code, Codex, Muse, Antigravity) can drive the whole pipeline.

The interesting bit

The architecture is the story: since Flow’s old REST API stopped minting auth tokens, the only way to call it is from inside a real signed-in browser session — so the extension mints reCAPTCHA and executes the RPCs there, with a live dashboard logging every request. It’s automation by puppeteering the web app, not by API.

Key highlights

  • Full pipeline: story → entities → reference images → scene images → clips → narration → concat → thumbnails → YouTube upload
  • Reference-image system keeps characters visually consistent across dozens of scenes (shown off in a 50-scene project)
  • All Omni 1.1 Flash video modes live-verified: text-to-video, first frame, first+last frame, references
  • Agent-agnostic: skills are Markdown files any coding agent can read; review providers swap at runtime without a restart
  • Local React ops dashboard with KPIs, per-scene pipeline view, and seven languages

Caveats

  • Three Veo capabilities (video upscale, reference-to-video, start+end-frame chaining) are unported and fail with UNSUPPORTED_ON_BATCH_API; upscale has no fallback at all
  • Depends entirely on Google Flow’s internal transport — a UI change on Google’s side could break it overnight
  • Requires a signed-in Chrome session and ffmpeg; not a headless, server-friendly setup

Verdict

If you want to mass-produce AI-generated YouTube content and are comfortable with browser-bridge automation, this is a remarkably complete factory. If you need a stable, sanctioned API — this is the opposite of that, and proudly so.

Frequently asked

What is crisng95/flowkit?
FlowKit drives Google's Flow video generation end-to-end — story to upload — by routing its RPCs through a signed-in browser extension, because that's the only place they can be signed.
Is flowkit open source?
Yes — crisng95/flowkit is open source, released under the MIT license.
What language is flowkit written in?
crisng95/flowkit is primarily written in Python.
How popular is flowkit?
crisng95/flowkit has 1.1k stars on GitHub.
Where can I find flowkit?
crisng95/flowkit is on GitHub at https://github.com/crisng95/flowkit.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.