How Much Does It Cost to Make an AI Drama?

Updated August 19, 2026 · Written by the AI Drama Reviews editorial team · 35 min read

DramaBoxEditor's pick — biggest short-drama library, lowest price to go unlimited 9.8Excellent

• Thousands of series, roughly 200 new titles every month
• Free episodes daily, plus ad-unlocks — no card needed to start
• Unlimited plans from about $5.99 a week, the lowest entry price we found
• Runs on Android, iOS and in a desktop browser

Go to DramaBoxAffiliate link · free to download · in-app purchases available

Affiliate link. We may earn a commission if you install through it, at no extra cost to you — it never changes our scores or placements. How we test · Disclaimer

The short answer, with the caveats attached

Budget $3,000 to $14,000 for a 60 to 80 episode AI microdrama if your pipeline is efficient and your labour is your own; budget closer to $28,000 if you are paying for the hours, as independent creator Jiang Lan did across twelve eight-minute episodes at about $280 per finished minute. Both figures are published, dated and real. Neither includes what it costs to make anyone watch.

Those two anchors sit on opposite sides of a gap that this page exists to explain. The NOW News figure of $3,000 to $14,000, published December 15, 2025, describes a full season — 60 to 80 episodes — for less than the cost of a single day of conventional vertical-drama shooting. Jiang Lan's accounting, reported by TVBS on August 7, 2026, covers about 96 minutes of finished runtime for roughly $28,000. Per finished minute those are wildly different operations, and the difference is not the tools. It is the acceptance rate, the labour rate and the country.

The honest framing is that AI has made the generation step cheap and left everything around it priced as before. Editing is still editing. Rights review is still rights review. And distribution, in a category where installs grew 238% year on year in the first quarter of 2026, has become more expensive precisely because production became cheaper. Costs did not disappear; they moved.

Who this page is for

This is the creator-side money page. If you want to know what an AI drama costs to watch, that is a different and much smaller number — coins run $0.20 to $0.50 an episode and unlimited weekly plans start around $5.99 — and it lives in how much AI dramas cost to watch. Everything below is about the side of the transaction that pays.

What we can and cannot tell you

No US benchmark exists for the cost of a completed, edited, legally cleared AI-drama episode. That is a real gap, not a modest one, and we would rather state it than invent a figure to fill it. Every dollar amount on this page is either a published figure with a source and a date, or arithmetic we performed on those figures and labelled as ours. Where a cost line has no published price — voice correction, music licensing, legal review — we say so and leave it open.

The headline numbers

Five published figures define the cost landscape for AI drama production in 2026, and they disagree with each other by two orders of magnitude. That disagreement is information, not noise: it tells you the spread between a first attempt and an industrialised pipeline is enormous, and that where you land inside it is mostly under your control.

One vendor figure sits alongside these and should be read as a vendor figure. FlexTV vice-president Tang Tang told MIT Technology Review on May 15, 2026 that moving North American production to AI cut costs by 80 to 90%, compressed schedules from three or four months to under one, and reduced teams to about ten people. That is a company describing its own results, unaudited, with no breakdown published. It is directionally consistent with the ranges above, which is the most that can be said for it.

Note what the chart above cannot show. None of these numbers includes user acquisition. Every one of them measures the cost of making the footage exist, not the cost of putting it in front of a paying viewer.

What does a full season of an AI drama cost?

Take a standard season — 70 episodes at 90 seconds each — and it comes to 6,300 seconds, or 105 finished minutes, or 1.75 screen hours. That is the unit every published rate can be converted into, and doing the conversion is the fastest way to see how far apart the published figures really are.

Why the same season spans $3,150 to $87,500

The spread is not explained by tool choice. Every producer in that table has access to broadly the same models. It is explained by three variables that compound: how many generations you throw away, how much you pay the humans who fix what survives, and whether you are amortising a workflow across many titles or building it once for one.

The first variable is the one this page keeps returning to, because it is the one you can change fastest. A pipeline that accepts one generation in two and a pipeline that accepts one in eight are separated by a factor of four on compute alone, before anyone opens an editor.

Episode length changes everything

Standard microdrama episodes run 60 to 120 seconds. Jiang Lan's episodes ran eight minutes — four to eight times the category norm — which is why twelve of them cost what a seventy-episode season of 90-second episodes might. Always convert to finished minutes before comparing two production budgets. Episode counts are marketing; minutes are the unit that costs money.

What is cost per accepted shot, and why does it matter?

Cost per accepted shot is your total spend divided by the number of shots that survive into the finished cut — and it is the metric every tool's pricing page is designed to hide. Vendors quote cost per second of output. You pay for every second you generate, including the ones with six fingers, the ones where the face changed, and the ones where the lips stopped matching the dialogue.

Why generated seconds and usable seconds are different quantities

Three structural facts make the gap large rather than marginal. Shots rarely exceed 10 to 12 seconds, which is a model constraint rather than a directorial choice, so a 90-second episode is assembled from a dozen or more separate generations. A scene is built from multiple short renders, not one long one, so continuity has to be re-established at every seam. And the known failure modes of generated video — identity drift, loss of object permanence, deforming hands, clothing and hairstyle changing between cuts, background text turning into pseudo-letters, lip movement drifting out of sync, screen direction reversing — are not rare events. They are the default output that direction and reference control reduce.

ACM's survey Spatiotemporal Consistency in Video Generation (May 18, 2026) treats exactly this as the open research problem. Work on OpenReview (March 26, 2026) and arXiv:2510.04999 (October 6, 2025) reaches the same conclusion about narrative and temporal coherence across long sequences. Translated into a budget line: consistency failures are a known, unsolved, recurring cost, not a beginner's mistake you will grow out of.

The arithmetic of acceptance rates

Here is the calculation, performed by us, for one 70-episode season of 90-second episodes — 6,300 finished seconds — with an average accepted shot length of six seconds. That is 1,050 accepted shots, or 15 per episode. The only variable is how many generations each accepted shot costs you.

Every number in that table is our arithmetic, not a reported result. The $0.10 per second input is OpenAI's published Sora 2 API price at 720p; the six-second average shot and the acceptance rates are assumptions we have labelled as assumptions because no source publishes acceptance rates. We include the table because the shape of it is the point: moving from one-in-twelve to one-in-four saves $5,000 of compute on a single season, and that improvement comes from a locked character bible, fixed seeds, first-and-last-frame controls and shorter shots — process discipline, not spending.

The mitigations that work are documented and limited. Character reference images, a locked character bible, fixed seeds, first-and-last-frame control, deliberately shorter shots, shot-level human review and continuity editing all reduce failure rates. None eliminates them. Budget on the assumption that a fixed percentage of your compute is waste, and measure that percentage from episode one.

How much do the AI video tools cost per month?

The video tools that produce AI dramas cost between nothing and $200 a month, and that range tells you almost nothing about what a season will cost. A subscription is a rate card for compute. What matters is what each dollar buys in seconds of output, which is the next section.

The caveat that travels with this table

Several prices above come from secondary sources rather than retrieved official pricing pages, and we have marked which. Tool pricing in this category changes on a scale of weeks. Treat the Kling and Pika figures as indicative, verify anything you are about to buy at checkout, and note that one of the most heavily marketed end-to-end products — Topview Drama Studio — does not publish a price at all. An unpublished price is a finding, and we score it as one.

A fuller walkthrough of what each model can and cannot do, including generation length limits and watermarking, is in the AI tools behind AI dramas.

The licensing trap in the cheap tiers

Luma's Lite plan at $9.99 a month is watermarked and explicitly non-commercial. Runway's free tier watermarks output. Google applies a SynthID watermark to Veo output regardless of tier. If you build a season on a plan that does not grant commercial rights, you have not saved money — you have produced footage you cannot monetise. Check the licence before the price.

How do AI video credits turn into dollars?

Runway Gen-4.5 consumes 12 credits per second, so a five-second generation costs 60 credits and a ten-second generation costs 120 — which means a Standard plan's 625 monthly credits buys roughly 52 seconds of generation before a single one is wasted. That is the entire argument of this page in one calculation. Fifty-two seconds is less than one episode.

Everything in this section is arithmetic. The credit prices and credit costs are published by the vendors; the divisions are ours, and we have labelled every derived figure so you can check it.

Runway, plan by plan

Read the fourth column against the format. A season is 105 finished minutes. The Max plan, the most expensive Runway offers, generates about 13 minutes of raw footage a month — and raw is not finished. At a one-in-four acceptance rate, thirteen minutes of raw output yields roughly three minutes and fifteen seconds of usable material. A 70-episode season on a single Max seat would take about 32 months of subscription. That is the calculation nobody runs before they start.

Two mechanics make it worse than the table suggests. Monthly credits expire on several Runway plans, so unused capacity does not bank for a heavy month; purchased credits do not expire. And the web and API credit pools are separate, so credits bought for one do not serve the other.

Google Flow, tier by tier

A Plus plan buys sixteen seconds of Veo 3.1 Quality a month. Not sixteen seconds of finished drama — sixteen seconds of raw generation, from which you might keep four. That is the gap between a subscription price and a production budget, stated as plainly as it can be stated.

Three further mechanics matter. Google's monthly credits do not roll over, so a light month is money burned. The 1080p upscale is free on Plus, Pro and Ultra, but 4K is unavailable to non-Ultra tiers and costs 50 credits for Ultra. And Google's own help pages are not internally consistent about Ultra pricing — we have printed both listed configurations above rather than pick one, because we cannot verify which is current.

Sora 2, priced by the second

The Sora 2 API is the cleanest pricing in the category because it removes credits entirely: $0.10 per second at 720p, portrait or landscape, with synchronised audio, under model id sora-2. A ten-second shot costs $1.00. A 90-second episode, if every generated second were accepted, costs $9. At a one-in-four acceptance rate it costs $36. At one in eight, $72 — and a 70-episode season at that rate is $5,040 in compute alone.

Sora 2 Pro has been reported at $0.30 to $0.70 per second depending on resolution, which multiplies every figure in this section by three to seven. The original Sora consumer product was reported as no longer available on April 26, 2026, while the API remained active — worth knowing if a workflow you read about last year assumed the consumer app.

What the arithmetic tells you to do

Three conclusions fall out of these tables. Buy annual if you are certain — Runway's annual pricing cuts the per-second cost by about 20%. Buy at the top tier rather than stacking cheap seats, because per-second cost falls by more than half from Standard to Max. And treat per-second API pricing as the honest comparator: convert every credit plan to dollars per generated second before you compare it to anything.

The costs nobody quotes

The headline price of an AI drama hides at least a dozen cost lines, and in most first-time budgets they add up to more than the generation itself. Below is the full list, each with what actually drives it. Where no US price is published, we say so rather than estimate.

The legal line is not optional

The US Copyright Office's Copyright and Artificial Intelligence, Part 2: Copyrightability (January 29, 2025) holds that purely AI-generated material is not protected, while human selection, arrangement and editing may be — and that applicants must disclose more-than-de-minimis AI material and exclude it from the human-authored claim. Part 3, on training data, remains pre-publication and is not a final rule. SAG-AFTRA's digital-replica framework requires consent, disclosure, compensation and control where a real performer's voice, image or performance is used. Tennessee's ELVIS Act has been in force since July 1, 2024; California's AB 1836 and AB 2602 cover digital replicas and contracts.

None of that is a budget line you can skip by deciding not to think about it. Our full breakdown, including what is and is not settled, is in are AI dramas legal in the US?

The lines that scale with episode count

Storage, QC and translation scale linearly with runtime, which means a 100-episode season is not "a bit more work" than a 40-episode one. Editing scales worse than linearly, because continuity errors accumulate across a series: a character bible that holds for ten episodes starts drifting by forty. Plan the season length against your editing capacity, not your generation capacity.

How much does user acquisition cost?

Paid user acquisition is now often the largest single cost of putting out an AI drama, and it is the line that almost no production guide includes. Adjust reported on June 23, 2026 that 67% of marketers were advertising in short-drama apps or interested in doing so, and that installs grew 238% year on year in the first quarter of 2026. Both numbers mean the same thing for your budget: you are bidding against everyone.

Why cheap production made distribution expensive

The mechanism is not complicated. When production cost falls 80 to 90%, as FlexTV says its did, the barrier to publishing a title collapses and title volume explodes. DataEye counted roughly 221,900 AI dramas published on Douyin in the first half of 2026 — about 1,200 a day — against 44,200 AI titles versus 3,248 live-action ones on the platform in April 2026 alone. Attention did not multiply to match. The only thing that rations a finite audience across an exploding supply of titles is price, and the price is paid in the ad auction.

This is why the savings do not land where new producers expect. AI moved money out of the production line and into the media line. A producer who models an 80% cost reduction and no change in distribution cost has modelled half of the transaction.

What we cannot tell you about acquisition costs

No published US benchmark exists for cost per install or cost per paying user in the short-drama category, and we will not estimate one. What is published is the demand side: 67% marketer participation or interest, 238% install growth year on year in Q1 2026, 2.3 billion global downloads in 2025, and Q4 2025 download growth of 186% year on year. APAC revenue per monthly active user was $1.45 in Q1 2026 (Adjust). That last figure is the useful one to hold next to whatever a media buyer quotes you: if acquiring a user costs more than the revenue a user generates, volume makes the problem larger, not smaller.

The structural alternative

There is one way to avoid this line entirely, and it is why most independent AI drama production goes through a platform rather than around it. Publishing through FlexTV, StoReel or a comparable operator hands the acquisition cost to a company that already runs the auction at scale, in exchange for the revenue split. Fifty percent of monetised views usually beats one hundred percent of a title nobody found. StoReel raised $34 million and targets 100 AI dramas a month, which is a spending capacity no individual producer competes with.

A worked 70-episode budget

Here is a complete season budget built line by line from the sourced figures on this page, with every derived number labelled as calculated by us and every unpublished line left open rather than guessed. The season is 70 episodes at 90 seconds — 6,300 finished seconds, 105 finished minutes, 1.75 screen hours. The acceptance rate assumed is one in four.

What this budget proves

Compute alone consumes $2,420 to $2,520 of a season that NOW News says can be made for $3,000. That leaves $480 to $580 for scripting, voice, music, editing, translation, legal review and marketing — which is not a budget, it is a description of unpaid labour. The $3,000 end of the published range is achievable only if the producer's own time is free and the acceptance rate is far better than one in four.

Conversely, the $14,000 end of the same range leaves roughly $11,500 for everything after generation, which is a coherent small-studio budget for 105 minutes of content — and still contains no acquisition spend at all.

How to use this template

Change three inputs and the whole thing re-derives: episode count, average shot length and acceptance rate. Measure your own acceptance rate over the first two episodes rather than assuming ours, then multiply. Producers who track cost per accepted shot from episode one are the ones who find out in week two, rather than month three, that their pipeline is not viable.

Cheap pipeline vs optimised pipeline

The distance between $30 per finished minute and $833 per finished minute is process, not technology. Both ends of that range use models available to anyone with a credit card. What separates them is how many decisions have been made before generation starts.

The genre lever is the cheapest one

Fantasy and xianxia titles are cheaper to make well than contemporary drama, because generation artefacts read as style rather than error and because effects that would consume a live-action microdrama budget are nearly free. Reviewer Jenny Cooper, writing on June 24, 2026 after watching 25 AI titles and finishing three, concluded that fantasy was the strongest category and that story remained decisive. That is one reviewer's qualitative judgement, not data — but it points the same way as the economics.

Why the $30 floor is not available to you

The roughly $30 per finished minute figure describes a heavily optimised Chinese pipeline, and the source is explicit that it is not a typical US result. It is reached at a volume — more than 95% of new Chinese microdramas AI-generated in Q1 2026, around 1,200 new AI titles a day on Douyin — where the workflow itself is the asset and each individual title is a marginal cost. A US producer making their first title is buying the workflow, not amortising it.

How does an AI drama make its money back?

The common revenue split between platform and creator is 50/50, and a title reaching a billion views has been reported to return $100,000 to $500,000 to the creator. That is the top of the distribution, reported as a range because it depends on where the views came from and how they were monetised.

What a viewer is worth

Work it from the viewer side. Coins across this category cost $0.20 to $0.50 an episode, so a viewer who finishes a 70 to 80 episode season one unlock at a time pays $14 to $40. At a 50/50 split the creator sees $7.50 to $25 from that viewer — our arithmetic, applied to the published coin rates and the published split. A viewer on a weekly unlimited plan, priced across the category between $5.99 and $29.99, contributes considerably less per season finished.

The demand side is genuinely large. US microdrama revenue was $1.3 billion in 2025 (Fieldfisher, June 1, 2026), with Omdia forecasting about $1.5 billion for 2026 and $2 billion for 2027. US monthly active users went from 26 million in 2024 to 66 million in 2025. Business Insider reported in January 2026 that roughly 30% of surveyed ReelShort users pay $25 to $100 a month while about 27% pay nothing. Every number in that paragraph describes the microdrama category as a whole, not AI titles specifically — no dataset separates them.

The three revenue routes

The third route deserves a specific warning. Publishing a high-volume AI slate to an ad-supported platform is exactly the pattern that monetisation policies on mass-produced content are written to catch. Original value is the test, and it is applied after you have produced the season.

What are the odds an AI drama becomes a hit?

DataEye counted roughly 221,900 AI dramas published on Douyin in the first half of 2026 and found about 1,055 of them — 0.47% — passed 100 million views. That is the clearest number in this entire category, and it is the one every cost calculation has to survive.

Read the threshold carefully. The 0.47% figure is for 100 million views. The $100,000 to $500,000 creator return is reported for a title reaching one billion views — ten times further out. The proportion of titles clearing that higher bar is not published, but it is necessarily smaller than 0.47%, which means the expected-value arithmetic in the next section is generous rather than pessimistic.

This is Chinese platform data, because China is the only market where AI-drama output is measured at scale. The US catalogue is smaller and less transparent — no authoritative dataset shows what percentage of any app's library is AI-generated — but the structural logic transfers. Cheap production has not made hits common. It has made misses cheap, and it has made the pool of competitors enormous.

What volume does to the odds

Roughly 1,200 new AI titles a day, and 44,200 AI titles against 3,248 live-action ones on Douyin in April 2026 alone, describe a market where the constraint has shifted entirely from production capacity to audience attention. If you are producing one title, you are one of 1,200 that day. The strategies that respond to that fact are portfolio strategies — many cheap titles, measured hard — not craft strategies.

Can an AI drama break even?

Multiply the 0.47% hit rate by the $100,000 to $500,000 reported for a billion-view title and the expected return per title is roughly $470 to $2,350 — against a production cost of $3,000 to $14,000. That calculation is ours, it uses a hit threshold ten times lower than the payout threshold, and it still comes out negative. State it plainly: on published figures, the average AI drama loses money.

The paying-viewer route to break-even

Expected value is not how most titles actually earn, so here is the other calculation. Take a $10,000 production budget with no acquisition spend. At a 50/50 split you need $20,000 of gross revenue. A viewer who finishes a season on coins pays $14 to $40, giving the creator $7.50 to $25. Break-even therefore needs somewhere between 800 and 2,667 viewers who finish and pay for the whole season — not viewers who watch the free episodes, of which there are typically three to ten.

The last row is the one to sit with. Adding a modest acquisition budget doubles the audience you need, which is why the acquisition line decides more outcomes than the production line does. And none of these rows accounts for the free episodes every viewer consumes before the paywall, or for the drop-off between episode one and episode seventy.

The odds, stated

We cannot tell you what share of AI dramas break even, because nobody publishes it and inventing the figure would be worse than admitting the gap. What we can tell you is the shape: a 0.47% chance of clearing a threshold ten times below the payout threshold, a market publishing 1,200 competing titles a day, an acquisition auction with 67% marketer participation, and a production cost that has fallen far faster than the cost of being found. Anyone selling you a course that describes this as a reliable income is describing the 0.47%.

Where do first-time producers lose money?

Every failure mode below is a budgeting error rather than a creative one, and each is visible in the arithmetic on this page before a single frame is generated.

Mistaking the subscription for the budget

A $15 Runway Standard plan buys about 52 seconds of Gen-4.5 output a month. A $19.99 Google One Pro plan buys 80 seconds of Veo 3.1 Quality. Neither is a season. The producers who fail earliest are the ones who budgeted twelve months of a consumer subscription and expected a hundred minutes of finished drama out of it.

Never measuring the acceptance rate

If you do not know how many generations your last accepted shot cost, you do not know your unit economics. Track it from the first episode. The difference between one-in-four and one-in-twelve is $5,000 of compute across a season and it is measurable in an afternoon.

Choosing photoreal contemporary drama

The genre with the highest failure rate is the one most first-time producers pick, because it looks like normal television. Micro-expression and sustained performance are the exact weaknesses of current models. Fantasy hides what contemporary drama exposes.

Budgeting zero for acquisition

This is the error that survives all the way to launch. A finished season with no distribution plan is a sunk cost, and the auction you are about to enter has 67% of surveyed marketers in it and 238% year-on-year install growth behind it.

Building on a non-commercial licence

Luma's Lite tier is explicitly non-commercial. Free tiers watermark. Discovering this after 70 episodes are generated is a total loss, and it is entirely avoidable by reading the licence before the price.

Ignoring the copyright disclosure requirement

The US Copyright Office requires disclosure of more-than-de-minimis AI material and excludes it from the human-authored claim. Producers who assume a detailed prompt confers ownership have built a business on an asset they may not own. Legal review is cheaper before production than after.

Season length set by ambition rather than editing capacity

Continuity errors accumulate. A hundred-episode season is not two fifty-episode seasons; it is a fifty-episode season plus a continuity problem that grows with every episode. Set the length against the hours you have for shot-level review.

Should you make one?

Yes if you are producing a portfolio of cheap titles through a platform that carries the acquisition cost, and you can hold generation under $3,000 a season. No if this is one title, self-distributed, in contemporary photoreal drama, funded by money you need back. That is the straight answer; here is who sits on each side of it.

Who it suits

• Producers with an existing audience. The acquisition line is the budget-killer, and anyone who already has distribution has removed it.
• Studios amortising a workflow across many titles. The $30-per-minute floor is only reachable this way, and the hit-rate arithmetic only works across a portfolio.
• Producers working with platforms. FlexTV and StoReel already run the auction at a scale you cannot match; a 50/50 split on found viewers beats all of nothing.
• Fantasy, xianxia and animated-manju specialists. The genres where the technology's weaknesses read as style.
• Anyone treating it as a craft investment. The workflow itself is durable even when an individual title is not.

Who it does not suit

• Anyone who needs the money back on one title. Expected return per title on published figures is $470 to $2,350 against a $3,000 to $14,000 cost.
• Self-distributors without a measured cost per install. Entering an auction with 67% marketer participation and no benchmark is not a plan.
• Contemporary photoreal drama producers. Highest reject rate, highest editing burden, weakest fit to the models.
• Anyone who cannot afford legal review. Copyright, likeness, voice and music are four separate surfaces and state law differs.
• Anyone assuming the tool subscription is the cost. If the credit arithmetic on this page was new to you, run it on your own plan before committing.

Frequently asked questions

The bottom line

A 60 to 80 episode AI drama costs $3,000 to $14,000 on the most-quoted published figure, about $29,400 at the one independent creator rate anyone has published, and $35,000 to $87,500 at industry per-screen-hour rates — for the same 105 minutes of finished drama. The spread is process, not technology, and the single biggest lever inside it is how many generations you throw away.

The subscription price is not the budget. Runway Standard's 625 monthly credits buy about 52 seconds of Gen-4.5 at 12 credits per second. Google Flow Plus buys 16 seconds of Veo 3.1 Quality a month, and those credits do not roll over. Sora 2's $0.10 per second is the honest comparator, and at a realistic acceptance rate it puts $2,520 of compute into a single 70-episode season before anyone edits anything.

The largest line is the one nobody quotes. Paid user acquisition is now often bigger than production, in an auction where 67% of surveyed marketers are participating or interested and installs grew 238% year on year in Q1 2026. Production cost fell 80 to 90%; the cost of being found did not.

And the odds are published. Of roughly 221,900 AI dramas on Douyin in the first half of 2026, about 0.47% passed 100 million views — a threshold ten times below the one at which $100,000 to $500,000 returns have been reported. Make one title and you are buying a lottery ticket. Make forty, measure cost per accepted shot on every one of them, and you are running a business.

DramaBoxEditor's pick — biggest short-drama library, lowest price to go unlimited 9.8Excellent

• Thousands of series, roughly 200 new titles every month
• Free episodes daily, plus ad-unlocks — no card needed to start
• Unlimited plans from about $5.99 a week, the lowest entry price we found
• Runs on Android, iOS and in a desktop browser

Go to DramaBoxAffiliate link · free to download · in-app purchases available

Affiliate link. We may earn a commission if you install through it, at no extra cost to you — it never changes our scores or placements. How we test · Disclaimer

Read next

Sources

• NOW News (December 15, 2025) — a 60 to 80 episode AI microdrama produced for $3,000 to $14,000. The most widely quoted season-level figure in the category; the report does not itemise what is included.
• TVBS (August 7, 2026) — independent creator Jiang Lan's own accounting: roughly $280 per finished minute and about $28,000 for twelve eight-minute episodes. A single creator's books, not an industry average.
• Industry reporting, 2026 — AI production at $20,000 to $50,000 per screen hour; a heavily optimised pipeline at roughly $30 per finished minute, stated as a Chinese result and explicitly not typical for the US; live-action vertical drama at $150,000 to $200,000 per screen hour, about $2,500 to $3,300 per finished minute.
• MIT Technology Review (May 15, 2026) — FlexTV vice-president Tang Tang on an 80 to 90% cost reduction, schedules falling from three or four months to under one, and teams of about ten people. A vendor describing its own results, unaudited.
• Runway — official pricing: Free $0 with 125 one-time watermarked credits, Standard $15/month ($12 annual) with 625 credits, Pro $35 ($28) with 2,250, Max $95 ($76) with 9,500. Gen-4.5 released December 1, 2025 at 12 credits per second. Monthly credits expire on several plans; purchased credits do not; web and API credit pools are separate. All second-counts and per-second costs derived from these figures are our arithmetic.
• Google Flow / Veo 3.1 help pages — free tier 50 credits a day in supported regions, Plus 200 credits a month, Pro 1,000 credits a month ($19.99 Google One Pro), Ultra listed at $100 for 10,000 credits and $200 for 25,000. Veo 3.1 Lite 10 credits (5 for Ultra), Fast 20 (10), Quality 100; Lite and Fast at 4, 6 or 8 seconds, Quality at 8 seconds. 1080p upscale free on paid tiers, 4K Ultra-only at 50 credits. Monthly credits do not roll over. Google's own pages are not internally consistent on Ultra pricing — we print both listed configurations rather than choose.
• OpenAI — Sora 2 API at $0.10 per second at 720p, portrait or landscape, synchronised audio, model id sora-2. Sora 2 Pro reported at $0.30 to $0.70 per second by resolution. The consumer Sora product was reported no longer available as of April 26, 2026; the API remains active.
• Luma — pricing page updated January 25, 2026: Lite $9.99/month web ($12.99 iOS), 3,200 credits, watermarked and non-commercial; Plus $29.99 ($37.99 iOS), 10,000 credits, no watermark, full commercial rights; Unlimited $94.99 ($119.99 iOS). Ray3.2 announced June 9, 2026.
• Adobe — Firefly Standard at $9.99/month; Firefly Video Model announced in limited public beta October 14, 2024, 100+ prompt languages. Adobe's claim that its training data is commercially safe is a vendor claim.
• Secondary sources, 2026 — Kling at approximately $10 / $37 / $92 / $180 across Standard, Pro, Premier and Ultra; Pika at Free / ~$8 / ~$28 / ~$76. These come from independent reports rather than retrieved official pricing pages and should be verified at checkout. Topview Drama Studio, publicly marketed July 2026, displays no price on its public page.
• Adjust (June 23, 2026) — 67% of marketers advertising in or interested in short-drama apps in 2026; Q1 2026 install growth of 238% year on year; 2.3 billion global downloads in 2025 and Q4 2025 download growth of 186% year on year via Sensor Tower; APAC revenue per monthly active user of $1.45 in Q1 2026.
• DataEye (2026) — approximately 221,900 AI dramas published on Douyin in the first half of 2026, roughly 1,200 a day, of which about 1,055 (0.47%) passed 100 million views; approximately 44,200 AI titles against 3,248 live-action on Douyin in April 2026; around 470 AI titles a day in January 2026 via MIT Technology Review. China Netcasting Services Association via Global Times — more than 95% of new Chinese microdramas AI-generated in Q1 2026.
• Omdia (2026, via Reuters) — US microdrama revenue forecast of about $1.5 billion for 2026 and $2 billion for 2027, against approximately $0.82 billion in 2024; US monthly active users of 66 million in 2025, up from 26 million in 2024. Fieldfisher (June 1, 2026) — US microdrama revenue of $1.3 billion in 2025. Forecasts are forecasts, and Omdia and Deloitte figures use different scopes and must not be combined.
• Business Insider (January 28, 2026) — roughly 30% of surveyed ReelShort users pay $25 to $100 a month while about 27% pay nothing. 36Kr (March 26, 2026) — StoReel's $34 million raise and target of 100 AI dramas a month.
• US Copyright Office — Copyright and Artificial Intelligence, Part 2: Copyrightability (January 29, 2025): purely AI-generated material is not protected, human selection and arrangement may be, and more-than-de-minimis AI material must be disclosed. Part 3 on training data remains pre-publication. SAG-AFTRA digital-replica materials (November 17, 2025). Tennessee ELVIS Act in force July 1, 2024. California AB 1836 and AB 2602 (2024). YouTube monetisation guidance on mass-produced, repetitive, low-value content.
• ACM, A Survey: Spatiotemporal Consistency in Video Generation (May 18, 2026); OpenReview on narrative and temporal consistency in long video (March 26, 2026); arXiv:2510.04999, Bridging Text and Video Generation: A Survey (October 6, 2025). Vertical Drama Love (June 24, 2026) — reviewer Jenny Cooper on 25 AI titles watched and three finished, cited as a stated opinion rather than data.

Affiliate disclosure. Some outbound links on this page, including the DramaBox links, are affiliate links. If you install through one we may receive a commission at no additional cost to you, and it never changes what we write — this page tells prospective producers that the average AI drama loses money, which is not a claim that helps anyone sell anything. Accuracy notice. Every dollar figure here is either a published figure with a source and a date, or arithmetic we performed on those figures and labelled as ours. Acceptance rates, shot lengths and season structures used in the worked examples are stated assumptions, not measurements — no source publishes them. Tool pricing changes on a scale of weeks and several prices come from secondary sources; verify at checkout. No US benchmark exists for the cost of a completed, edited, legally cleared AI-drama episode, and we have left that gap open rather than filled it. Verified August 19, 2026.