Opinion, with arithmetic

The credits trap

Per-video pricing doesn't just cost money. It quietly changes what you make — and for a faceless channel, that is the expensive part.

We sell a flat-priced tool, so we have a stake in this argument. The math below is checkable; the section on when credits win is not a courtesy.

The mismatch in one line

The arithmetic

A serious faceless channel publishes about one video a day — call it 30 a month — and makes several more that never ship, because killing bad ideas is the job. So the real monthly need is closer to 35–45 renders.

Now price that. Vendors change plans constantly, so run your own numbers, but the shape is always the same:

The per-video figures come from our cost breakdown. The ratio doesn't improve with scale — that's the point. Credits are linear. Owning the compute is flat. The two lines diverge forever.

The part that actually costs you: the iteration tax

The money is the small problem. Here's the big one.

Faceless YouTube is a search-and-selection process. You don't know which title, hook, or thumbnail works — you find out by shipping and measuring. The channels that win are the ones that ran the most experiments.

Now put a meter on it. Every A/B test costs a credit. Every "let me try that hook a different way" costs a credit. Every regenerate-because-the-voice-was-off costs a credit. So you stop. You publish the first draft because the second one costs money. You skip the test.

This is well-worn ground in pricing design: metered pricing makes users ration. That's usually a feature — it stops waste. For a business whose success depends on volume of attempts, it's backwards.

Where the vendor's incentive points

Not a conspiracy — just arithmetic. A credit vendor rents GPUs by the second. Your render is their cost. So their margin improves when you render less, and their revenue improves when you render more. They resolve that by selling you a plan slightly too small for what you actually do, and charging for the overage.

A flat-priced tool that runs on your hardware has the opposite shape: your renders cost the vendor nothing, so nobody is trying to slow you down. (Our own bias, stated plainly: this is the model we sell. It also means we need a fair-use ceiling somewhere — anyone claiming literally infinite anything is lying.)

When credits genuinely are the right choice

This is not a bad product, it's a mismatched one. Credits are the better call when:

• You publish occasionally. Under ~10 videos a month, a credit plan is cheaper than a GPU that sits idle — see the amortisation table in our cost breakdown, where a $450 card costs more per video than credits at low volume.
• You have no capable GPU and don't want one. 8 GB is the realistic floor for local AI video; renting compute is a perfectly sane alternative to buying it.
• You need a specific hosted model that isn't available to run locally. Some are genuinely better. You cannot self-host what nobody ships.
• You want zero setup. Local AI means drivers, models, disk space and a bad evening at some point. Credits mean a browser tab. That convenience is worth real money to plenty of people.
• You're testing whether you even like this. Never buy hardware for a hobby you haven't started.

What to actually check before you pick

1. Cost at your real volume, not the volume on the pricing page. Multiply by 40, not 4.
2. What a retry costs. If regenerating burns a credit, your iteration rate is now a line item.
3. Whether credits roll over, and what happens when you run out mid-month.
4. Who owns the output — and whether there's a watermark on the tier you'd actually buy.
5. Whether the price survives you succeeding. Plans that work at 5 videos and break at 50 are a growth ceiling you're paying for.

FAQ

Related