Self-hosted clip production
An 8-hour stream is 40 clips. Finding them is the work.
clipfarmer ingests a long stream VOD, finds the moments worth cutting, renders them as vertical clips with subtitles, writes a caption for each one, and posts the clips you approve to your own TikTok account. It runs on your computer, not ours.
Every stage, every VOD, on one screen
A long VOD is a long chain with a human decision in the middle of it. clipfarmer shows the whole chain and what each stage is doing right now — so "what happens next" is never a guess.
What it does
-
Ingests the VOD
Point it at a Twitch VOD or a YouTube video you have the right to use. It pulls the video, extracts audio, caches keyframes, and fetches chat replay when there is one.
-
Transcribes with word-level timing
Whisper, per clip, with word timestamps — which is what makes subtitles land on the syllable instead of drifting behind the speech.
-
Ranks candidate moments
Short sliding windows scored on audio energy, hook phrases, laughter, punctuation density and speech pace, then re-ranked by a local language model. Both rankings are scored against the clips real viewers made from the same VOD, so the ranker's quality is measured rather than assumed.
-
Hands you the shortlist
You watch trims and approve or reject. Nothing auto-selects. Generation is a model's strength; choosing is taste, and taste is the part that stays yours.
-
Renders the vertical clip
Subtitles, per-speaker rows, the right layout for the scene, a punch-in when the gameplay stops mattering, and a burned-in header on the clips that need one.
-
Writes a caption, then posts what you approve
A caption and hashtags per clip, which you edit. Approved clips go to your own connected TikTok account — as a draft in your inbox, or posted directly.
The review step is the point
Six LLM ranking variants lost to a z-score on this dataset. That result shaped the whole tool: the machine does the work a machine is good at — watching eight hours, cutting frames, measuring loudness — and every judgment that decides what a stranger sees stays a human one.
Posting to TikTok
clipfarmer connects to one TikTok account: yours. You authorise it on TikTok's own consent screen, and you can revoke it from the TikTok app at any time. Approved clips can land in your TikTok inbox as a draft you finish in the app, or post directly with the caption you wrote.
It runs on your machine
clipfarmer is software you run, not a service you sign into. There is no clipfarmer server, no account, and no upload of your video anywhere except to the platform you chose to post to. The VODs, transcripts, renders and access tokens all sit in a working directory on your own disk.
No account
Nothing to register for. You install it, and it runs as a local web app in your browser.
Your files stay yours
Source video, audio, transcripts and finished clips live in jobs/ on your
disk. Finished jobs can be archived to reclaim the space.
One account, yours
The TikTok connection is a single account you authorised yourself, and it is revocable from TikTok at any time.
No tracking
No analytics, no telemetry, no advertising. This website sets no cookies either.
Built to be run by one person
clipfarmer is a single-operator tool, and that is a design constraint rather than a stage it is trying to grow out of. It assumes one person's taste, one channel's material, and one account's voice, and it optimises for the case where that person is the only reviewer the clips will ever get. The documentation covers installing and running it; the FAQ answers what it does with your data and your account.