What an AI Short-Drama Episode Actually Costs

The honest headline is that an episode costs less than lunch and more than nothing, and that the spread between a careless production and a disciplined one is roughly five to one. That spread is the whole subject of this page: not what the platform charges, which you can read on its pricing page, but what your habits charge on top.

Data Notice: Plan names, prices and allowances in AI media platforms change frequently. Figures below reflect the vendor’s published pricing at time of writing. Treat the vendor’s pricing page as the authoritative source; this guide is about the method, which does not go stale.

How generation is metered

Two models are in play, and which one applies depends on your plan.

Pay-as-you-go, in credits. Credits are what generating costs. Anything that makes new media spends them — an image, a video, an erase, an upscale, each item in a batch counted separately. The rule that matters most: you are only charged for work that produced something.

When you start a generation the cost is held, and only actually spent once a result is saved. If the work fails, is cancelled or is rejected, the hold is released. This is why a balance can briefly read lower than expected while something is running — that amount is reserved, not gone.

Subscription, with generation included. TJP’s paid tiers are sold on volume rather than per-generation metering: Pro at $50/month for people posting weekly, Max at $200/month for agencies and heavy users. What separates them is queue priority, batch size, access to advanced avatar and video features, and — on Max — the ability to train your own avatar.

Either way, the cost of what you are about to run is printed on the button that starts it. You never find out afterwards.

What is free

Worth knowing, because it changes how you work:

Browsing, searching and looking at anything
Your whole library: organising, folders, favourites
Downloading your own work, as often as you like
Uploading files
Saving a video into your library from a link

Only generating costs. Reviewing, filing and exporting are free, which means there is no reason to be economical about looking at your own work carefully before deciding what to run next.

The shape of an episode bill

A ninety-second episode from the shot list method is roughly twenty-five shots. The cost is not twenty-five video generations — or should not be.

Stage	Volume	Relative cost
Prompt drafting on stills	40–80 images	Low — this is the cheap part
Approved key frames	25 images	Low
Video generations	25–40 clips	This is the bill
Re-rolls on failed shots	5–15 clips	The variable that decides everything
Upscales on keepers	3–8	Low
Cutting, sound, titles	—	Free (outside the platform)

Video is where essentially all the money goes. A still costs a fraction of a clip and takes seconds rather than minutes. Everything below follows from that one fact.

The five habits that cut the bill

1. Settle every idea on stills first

A still answers nearly every question you have about a shot — framing, wardrobe, light, mood — for a fraction of the price of finding the same answer by generating video. Then animate the frame you approved.

This single habit is worth more than the other four combined. It also improves the result, because you are directing motion on a composition you already chose rather than gambling on both at once.

2. Draft on something cheap, finish on something good

Models differ in speed, cost and look, and the cost of each is shown where you pick it. Get the prompt right on a fast, inexpensive model; then run the final version on a better one.

Most of the work is deciding what you want, and that part does not need an expensive model. Spend at the end, on the version you are keeping.

3. Generate small, upscale the keeper

Larger resolutions cost more and take longer. Generating everything at 4K and discarding most of it is the most expensive possible way to work.

Generate at a normal size, choose, then upscale only what survives. Upscaling reconstructs detail as it enlarges rather than stretching, so the keeper still ends up sharp — see covers and thumbnails.

4. Use “how many” instead of repeated single attempts

Where the panel offers a quantity control, four at once costs the same as four in a row — but four at once gives you a choice, and four in a row gives you four verdicts, each one prompting another attempt.

Two to four is a reasonable habit while a prompt is still being refined.

5. Never batch an unproven prompt

A batch multiplies whatever you give it, including the mistake. A hundred items built on a prompt that was not ready costs a hundred times as much to discover it.

Run one. Look at it properly. Then run the batch. See producing a season.

What failures cost

Less than people fear, and this is worth knowing before you start experimenting.

Items that fail do not consume credits — the amount held is released back. For a batch this applies item by item: you pay for the items that produced something and not for the ones that did not. A batch that half-fails charges you for half, and partial success is normal on large batches rather than a sign that something is wrong.

What does cost you is a generation that succeeds and is unusable. The model produced something; you simply do not want it. That is the number the five habits above are aimed at.

Which plan, and when

The advice that survives every pricing change: use the free tier until you run out.

Running out tells you far more about which plan you need than any comparison table can, and it costs nothing to find out. Specifically, it tells you your actual production rate — which is the only input that matters.

Rough guidance on the decision:

Free — you are learning the pipeline and have not published a series yet. Correct for almost everyone reading this for the first time.
Pro — you are posting on a schedule and generation volume, not ideas, is the constraint.
Max — you are producing at agency volume, or you specifically need to train your own avatar, which is gated on the top tier and an active subscription. See same face every episode.

Do not buy the training tier before you know your format works. A trained avatar is the right investment for a series that has an audience; it is an expensive way to procrastinate on a series that does not exist yet.

The comparison worth keeping in mind

The reason any of this reads as cheap is the thing being replaced. A single day of conventional production for the same ninety seconds — crew, camera, location, talent, edit — does not have a three-figure budget. It does not usually have a four-figure one.

That gap is the actual product here, and it is why the discipline matters: the compression is so large that it is easy to waste half of it and still feel like you got a bargain. You did. You could have had twice as much.