It Makes Videos While I Sleep

The real cost of a thirty clip day

Batching gets sold as speed, and that is the wrong promise. A clip that takes four minutes at 2pm takes four minutes at 3am too. The queue moves the generation minutes into hours you were never going to use, and your attention collapses from a scattered day into one calm evening block.

Babysitting looks harmless on paper. Each clip needs you for only a moment, type the prompt, wait, check it, save it, start the next. But those moments land every few minutes across eight hours, and the gaps between them are too short to do anything real in. So the day dissolves, and your actual typing time was maybe an hour while the whole eight hours are gone.

There is a second cost the clock cannot show. Attention wears down, so by clip twenty two you are tired of looking at renders and you keep things you would have cut at clip three. A queue spends none of your attention while it generates, so all of it stays for the one pass that matters, the morning review.

Write every prompt before you generate anything

The queue forced one habit on me that improved the videos more than any model setting, which is writing all thirty prompts in one calm sitting before generating anything. When prompts get written scattered across a week, the style drifts without you noticing. Monday says soft light and slow camera, Thursday says punchy and vivid, and a month later the account looks like three different people run it. Prompts written in one sitting carry one mood, the same camera language, the same lighting words, because your head stayed in one mode while you wrote them.

Batch writing also exposes weak prompts before they cost you a render. A lazy prompt hiding alone in a chat box looks fine. The same prompt sitting in a plain text file under fourteen sharper ones is embarrassing, and you fix it in thirty seconds. So write the whole set in one file, one prompt per line, and read it top to bottom once before any of it touches the queue. That one habit costs you nothing and you can start it today.

How the queue runs while I sleep

The tool itself is simple in concept. The queue holds your prompts, takes the first one, submits it to VEO or Google Flow, waits out the generation the way I used to, saves the finished clip with a sensible name, marks the row done, and moves to the next. It repeats that all night with nobody present. In the morning there is a folder of clips and a log of what happened while I slept.

Simple in concept still needs to survive the night, and this is where most homemade versions die. Three failures are guaranteed if you run long enough. A generation fails or gets refused. A session expires mid run. And one night Windows decides that 3am is the perfect time for an update and a restart. If your queue holds its state only in memory, that restart erases the whole night and you wake to an empty folder.

Here is the one rule that saves you, and you can use it in anything you build tonight. Save the queue's state to a small file on every single change, every prompt added, every clip finished, every retry counted. Then a crash costs you nothing, because the tool reads the file on launch and resumes from clip nineteen instead of starting again from clip one. I use the exact same rule in my X posting tool, and I wrote about it in how I schedule X posts weeks ahead.

That rule is only the floor. The full night shift also needs retry logic that gives a failed clip two more tries then moves on, so one stubborn prompt cannot hold the other twenty nine hostage. It needs session handling that pauses politely instead of hammering a dead login. And it needs naming and logging you can actually read at breakfast. Wiring all of that yourself is a real weekend, so I spent the weekend once and packaged it.

What batching multiplies and what it cannot fix

One warning saves you a bad month. A queue is a multiplier, and a multiplier does not care about the sign of what it multiplies. Feed it a sharp prompt and you wake to thirty clips that look like a brand. Feed it a lazy one and you wake to thirty copies of the same mistake, produced efficiently, at volume, while you slept. Batching took the babysitting off my plate and left the responsibility for the input exactly where it always sat.

So review the morning folder in one pass and cut without mercy. Thirty clips judged in one sitting is faster per clip than judging them through the day, because your eye calibrates, clip seven sets the bar and clip eight gets judged against it in three seconds. Some clips will miss, wrong motion, a face that melts, a camera move nobody asked for. Cut them and do not look back, because the queue will make more tonight.

This is also why the method points at paying work. The money sits in content pipelines where volume with consistency is the game, accounts and clients who want a steady stream that never drifts off style. One person with a day job cannot hand build that. One person with a prompt file, a queue, and a merciless morning pass can, and the wider system I use to run the freelance side is its own story in how I actually make money with Claude.