One-person content agency: 30 short videos a week, the real cost sheet

Seven avatar reels cost me $14.71 last week. That is the whole render bill, measured against the credit balance before and after the batch, not an estimate.

Most content about running a one-person content operation skips the part where you find out what it costs. This guide is only that part. Real per-unit numbers from a HeyGen and Claude pipeline that runs every week, the arithmetic extended to 30 videos a week, and then the public rates the category charges, so you can see the gap the business actually lives in.

One thing up front about the money. The cost numbers here are measured from my own pipeline. The revenue numbers are public rates for this category taken from freelance marketplaces and agency price pages, quoted so you have an anchor before you invent one. They are not a forecast of what you will earn.

What you'll have when you're done

• A per-unit cost for every stage of a short-form video, in cents
• The monthly bill for 30 videos a week, with the arithmetic shown
• A cost per finished video you can put on a quote without guessing
• The public rate bands for short-form retainers, to price against
• The three constraints that decide whether this is a business or a hobby
• [SCREENSHOT: the render log showing 7 reels, 294.16 seconds, $14.71]

Before you start

• A HeyGen account with an avatar you own. Photo avatar is fine and is what the numbers below are from. Whoever's face it is has to have consented on HeyGen's own consent flow.
• An ElevenLabs account. The numbers below assume the Creator plan, which is 100,000 credits a month at roughly $22.
• ffmpeg and faster-whisper installed locally. Both free, both run on the machine you already have.
• A scheduler. One Metricool seat covers the publishing side across networks.
• Claude, on whatever plan you already pay for. Pro is $20 a month at time of writing; check the current pricing page. It writes scripts and drives the pipeline, it does not render anything.

What each stage actually costs

These are the per-unit numbers. Everything else in the guide is arithmetic on top of them.

The line that surprises people is the editing row. Captions from faster-whisper word timestamps, cuts and layout from ffmpeg, both driven by a script: video 130 costs exactly what video 1 cost, which is nothing. That is the entire reason 30 a week is a batch job rather than 30 sessions.

The render bill, per reel, from a real week

A 40-second video costs about $2.00. That is the number to hold in your head, and the reason to never render a vertical feed video at 4K: the bill scales with finished seconds either way, but the resolution multiplies the rate for a format nobody watches on a large screen.

The script length arithmetic nobody does

You cannot cost a batch without knowing how many characters a 40-second script is, and the naive answer is wrong in both directions.

Measured across a run of real scripts, the read rate is not one number:

Size a money script at 15 characters a second and it overruns; size a quick-hit at 14 and it undershoots. Both mistakes cost you a re-render, and a re-render is the only genuinely wasteful line item in this whole stack.

At 14.0 characters a second, a 40-second script is about 560 characters.

The arithmetic for 30 videos a week

Thirty a week is 130 a month, using 4.33 weeks. Forty seconds each.

Render. 130 videos x 40 seconds = 5,200 finished seconds. At $0.05: $260 a month.

Voice. 40 seconds at 14.0 characters/second = 560 characters per video. 130 x 560 = 72,800 characters a month, which is 72,800 credits. The Creator plan's 100,000 credits covers it with about 27,000 to spare, which is your re-record headroom. $22 a month.

Editing. 130 videos x $0 = $0. The cost here is machine time, not money: budget a few minutes per video of local render, which is a scheduling problem rather than a billing one.

Scheduling. One Metricool seat. About $22 a month at the entry paid tier for one brand; check the current page.

Claude. Flat, whatever plan you are on. $20 to $200 a month depending on how hard you drive it.

That is the whole cost sheet. Two dollars fifty to three dollars ninety per finished, captioned, published short-form video, at 130 a month, run by one person.

What the category charges

Now the other side. These are public rates, so you have a floor and a ceiling before your first conversation.

Put that next to the cost sheet and the shape of the business is obvious: production is not where the money goes, and it is therefore not where your competitive advantage is either. A retainer at the bottom of that band, 8 videos a month, costs you about $30 in units. The other $1,470 is being paid for scripts, judgment, consistency and the fact that the client does not have to think about it.

The three constraints that actually decide this

The cost sheet is the easy part. Here is what is hard, in the order it bites.

1. Thirty scripts a week is the bottleneck, not thirty renders

The render queue does not care how many videos you send it. Thirty scripts a week that a stranger will watch past the third second is a writing job, and it is the entire job. If you take one thing from the cost sheet, take this: production is $2.50 a video, so stop optimising it and go spend that time on the openers.

2. No single client wants 30 a week

Thirty a week is a capacity number, not a client number. The realistic shape is four to six clients at five to eight videos each, which means four to six approval loops, four to six brand voices, and four to six people who want a change on a Friday. That coordination load, not the render bill, is what caps a one-person operation, and it is why the retainers in the band above are priced where they are.

3. Disclosure and consent are not paperwork, they are the account

Every platform this pipeline publishes to has an AI-content declaration and a policy requiring synthetic media to be labelled. Set the flag on every post:

And the likeness question, settled before the first invoice, in writing: whose face is on the avatar, who consented on the platform's own consent flow, and what happens to that avatar when the engagement ends. Your own likeness is the clean answer. A client's likeness needs their explicit written consent, personally, not their manager volunteering them.

What I would do differently at the start

Three things, from running this.

Measure the unit cost in week one, not month three. One 30-second test render and a credit balance before and after gives you the number your entire quoting model rests on. Estimating it is how people end up quoting a retainer below their own floor.

Do not buy the volume plan on anything until the volume exists. Every tool here has a tier that is cheaper per unit at high volume, and every one of them is more expensive per unit at the volume you actually have in month one. The Creator plan covers 30 a week; do not pre-buy for 60.

Build the re-render guard before you need it. The single largest avoidable cost is rendering the same video twice because the script was mistimed or the audio was wrong. Time the voice track first, check it against your characters-per-second constant, and only then send it to render. That check is worth more than every other optimisation in this guide combined.