From a source feedto a published video.
TubeFactory is a production line, not a prompt box. Six stages run in order on the schedule you set, each one recording what it did, so a finished video is something you can inspect rather than something that just appears.
Fifteen minutes,once.
Link your YouTube channel
One Google OAuth flow. TubeFactory gets upload permission on the channel you choose and nothing else; you keep full ownership and can unlink at any moment.
Pick a niche and a format
Choose a ready-made niche — it arrives with its own scraper sources, script voice and target length — then set landscape, Shorts or square. The format changes how the script is written, not just the crop.
Set the schedule
Add as many generation slots as you want (times and weekdays, in UTC) plus one publish time. The scheduler creates a real cron entry per slot, editable whenever you like.
Turn on WhatsApp approval
Add your number and choose a lead time. From then on, every video sends you a preview link that far ahead of its publish moment — approve, reject with a reason, or ignore it and let it publish.
What happenson every run.
Each stage writes to the video's stage log, so when something fails you get the provider, the stage and the reason — not a stack trace.
Scrape
Fresh source material, pulled on a schedule.
Every niche ships with a set of RSS, news and API sources. A scraper runs every few hours, deduplicates what it finds and files each story against the niche it belongs to, so there is always fresh raw material waiting when a channel's generation slot fires.
Script
A hook-first script written to a length target.
The scraped stories go to the language model with your niche's voice and a hard length mandate. It returns a title, a scroll-stopping hook, a scene-by-scene narration, a thumbnail concept, a description and tags — then expands the draft until it hits the word count a video of that duration actually needs.
Voice
A lifelike read, tuned against robotic delivery.
The narration is synthesised with your chosen ElevenLabs voice at settings picked for a natural human read — balanced stability, no style exaggeration, speaker boost on for clarity. One voice per channel keeps the show consistent.
Caption
Word-timed captions, burned in.
The finished audio is transcribed back with Whisper so the captions match what was actually spoken, down to the timing. Retention on faceless content lives and dies on captions, so this is never skipped.
Visuals
Stock B-roll or generated shots, per scene.
Each scene carries its own visual query. Point the channel at stock footage (Pexels, Pixabay) for a documentary feel, or at AI generation (DALL·E, Runway, Luma, Pika) when the shot you need does not exist. The prompt is rewritten to suit whichever source you picked.
Render & publish
Cut, scored, thumbnailed, uploaded.
Scenes are assembled to your channel's aspect ratio, scored with mood-matched music, captioned and rendered. A custom thumbnail is generated from the script's thumbnail concept, and the finished file is uploaded to YouTube — public, unlisted or private, now or on a schedule.
Nothing publishesbehind your back.
Rendering is the end of the machine's job, not the end of the process. With approval on, the finished video waits behind a preview link until its publish moment arrives.
You approve
The video uploads at the scheduled time with your privacy setting, custom thumbnail and — for episodes — its series playlist.
You reject
The upload is cancelled and your reason is stored on the channel as standing guidance for every future script.
You're busy
No reply by the publish moment counts as approval. The video goes out on schedule and you get a WhatsApp note saying so.