By Andrew Armstrong 
Short-form video is where the traffic is, and nobody who writes guides for a living has an evening spare to cut one. The tools that promise to do it for you want a subscription, stamp a watermark on the free tier, and fill the frame with AI-generated pictures of things that do not exist.
ShortGeek is the other way round. Point it at one of your own guides, an RSS feed, any article URL, or a one line idea, and it drafts a script, narrates it, burns in word by word captions and renders a 1080×1920 MP4 you can upload as it is. It is out today, version 1.0.0, free at home and at work, and every frame is rendered on your own machine.
Nothing in it is AI generated imagery
This is the part worth being blunt about. The default background is the guide’s own real screenshots, panned slowly full bleed behind each card, one image per beat. That is real content specific to that article, not a stock loop somebody else is also using.
The alternatives are drawn in code, frame by frame: gradients, a terminal scroll, a typing loop, a clean light look, Code Rain, Bounce Orbit, and a Sort Visualizer running an actual bubble sort on a bar chart. Or drop your own MP4 clips in and use those. Nothing is AI art and nothing is somebody else’s footage passed off as yours.
The script comes from the source, and you can edit it
A heuristic writer condenses the article into a hook, two to four beats and a call to action, preferring steps that carry a real command or a real image over plain prose. Nothing is invented: every claim comes from the text it was drawn from, and the whole script is editable before you render.
If you want the lines to sound more spoken, add your own Anthropic or OpenAI key in Settings and an optional polish pass rewords them. Same facts, same structure, only the wording. Leave the provider set to None and nothing external is ever called.
A real voice, without signing up for anything
The default is Microsoft Edge’s free neural text to speech, which needs no account. If that endpoint is ever unreachable, ShortGeek falls back to the offline eSpeak NG voice rather than failing the render, and says in the queue that it did. ElevenLabs is there too if you already have a key and a voice you like.
Caption timing comes from the voice engine itself rather than being estimated from word lengths, which is why the highlight lands on the word actually being said. Three styles: Bold Highlight, Minimal and Classic Subtitle.
What it will not do
The same rules as the rest of the range, and they are stated inside the app as well as here.
- No AI imagery. Real screenshots, backgrounds drawn in code, or clips you supplied. That is the whole list.
- No invented facts. Every line traces back to the source text, and you can edit all of it before rendering.
- No uploading on your behalf. Those platform APIs are locked down, quota limited and change constantly. You get a finished file and post it yourself.
- No subscription, no credits, no watermark. Nothing it renders is marked, and nothing is metered.
- No account. The only outbound calls are to the free voice service, and to an AI provider if you chose to add your own key.
Get it
One installer, about 71 MB, with Python and ffmpeg bundled. There is nothing to install first: no Python, no PATH ticking, no winget commands.
The full feature list, screenshots of every screen and the FAQ are on the ShortGeek product page.
What is next
Square and landscape output alongside the default 9:16, more backgrounds drawn in code, and a thumbnail picker for the library. Direct uploading is deliberately not on the list.
If a render fails or a script comes out wrong, open an issue on GitHub and say which source you used and what came out. A person reads them.
Discover more from TechyGeeksHome
Subscribe to get the latest posts sent to your email.
