darkzOGx/youtube-automation-agent

Adopt this if you want a repeatable video pipeline on your own machine and are willing to babysit provider keys, quota and review gates. It is not a push-button channel; the approval step is the product.

The problem

Running a faceless or semi-automated channel means juggling research, scripting, voiceover, stock or generated visuals, editing, metadata and upload scheduling across several tabs. The agent turns that sequence into one pipeline: you describe the niche and cadence, it proposes topics and drafts, then assembles an MP4 and queues it in a Pipeline view. The step that changes is the handoff between script and publish, which becomes a checkpoint instead of a manual re-upload. You get finished drafts waiting for fact review, media-rights confirmation and approval before anything reaches YouTube.

What's different

The design is built around gates rather than full autonomy. Simulated or placeholder media is marked and cannot be approved or uploaded, narration must work for a video to be approvable, and analytics-derived lessons only alter future output after you sign off. Production is resumable, so a crash or timeout at the render stage picks up from its checkpoint rather than regenerating everything. Provider support is broad, with Gemini, OpenAI and OpenRouter among the text options and several paid video providers alongside a free local slideshow default, so you can choose between cost and visual quality per channel.

The trade-off is that the interesting parts depend on external accounts. Image generation needs OpenAI or Gemini with billing enabled; without either, slides fall back to gradients and thumbnails to a built-in designer. AI video is opt-in with a paid-seconds cap, and every provider is a separate key and a separate failure mode. The system also does not remove the craft: hooks, factual accuracy and rights checks remain your job, and the agent formats and stages them rather than judging whether a script is good.

Who should try it

A good fit is a solo operator already comfortable with API keys and Google Cloud who wants to run, say, one or two explainer videos a week and keep editorial control. The local slideshow default is attractive there: you can validate the pipeline end to end before paying for generated clips. A poor fit is someone who wants a hands-off channel or has no Google Cloud project, since YouTube Data API credentials are mandatory and quota limits posting frequency. It is also a poor fit for anyone expecting a hosted service; the dashboard runs from your checkout at localhost.

Requirements and limitations

Node.js 18.17+ and a Google account with YouTube Data API credentials are hard requirements, and the walkthrough is the documented way to connect a channel. Chromium must be installed through Playwright for slideshow rendering, and FFmpeg is fetched by npm install, with an environment override if you already have it. At least one text provider key is needed; the startup capability check tells you what is missing, and a verified readiness run exercises text, narration, channel access and a test MP4 without uploading.

The ongoing costs are the ones to watch: Google Cloud quota can block publishing, paid image or video probes are separate opt-ins, and narration failures such as an ElevenLabs bad request or wrong voice ID are handled by regenerating narration only. Android/Termux users need a special install route for the sharp module. Review each provider's billing and quota page rather than assuming a free tier covers your cadence; the troubleshooting table and setup walkthrough are the places to check before committing.

HotRepo's take

HotRepo only observed this repository on a single day: on 2026-10-26 it sat at 4,644 stars with +35 that day on the daily board. One dated snapshot cannot show whether interest is rising or fading, so persistence can't be judged. The useful check before adopting is whether your Google Cloud project and chosen narration provider pass the documented readiness check in a test video, not the star count.