How to Commission an AI Vertical Drama Series in a Language Your Team Does Not Speak
Episode 34 arrives. Your commissioning lead opens it, watches ninety seconds of Bahasa Indonesia dialogue, and has no way to tell whether the confession scene landed or collapsed. The retention curve will say something in three weeks. The production partner says it is fine. Nobody in the room can adjudicate. This is the moment most cross language commissions go wrong, and it happens long after the point where it could have been prevented.
The instinct is to solve it with translation. Subtitle the rushes, run the script through a translation pass, read the English and approve. That approach fails for a specific reason. A vertical drama script is not carrying information, it is carrying tone, status and escalation, and those are exactly the things that do not survive a literal rendering. A line that reads as flat in translation may be the sharpest insult in the source language. A line that reads as devastating may be ordinary. You cannot commission on a signal that degrades in the exact dimension you are trying to judge.
What follows is a commissioning framework that removes the need for your team to read the dialogue at all. It replaces fluency with structure. Every gate below is one a non speaker can operate, provided the work to set it up happens before the first script is written rather than after the first delivery lands.
Separate the Production Language From the Approval Language
The first decision is which language the series is actually produced in. There are two workable models and one that consistently fails. The first workable model is native origination, where the series is written, generated and edited in the target language from the first outline, with English used only for summaries and reporting. The second is master and variant, where a master series is produced in a language your team does read, then localised into the target market with the localisation planned from day one. The model that fails is translation origination, where an English script is written for an English sensibility and then converted, because the resulting series reads as imported in a market where local origination is the competitive standard.
Native origination gives the strongest result in a single target market. It also gives you the least direct visibility, which is why the rest of this framework exists. Master and variant gives you more visibility and better economics across several markets at once, at the cost of the source market always being the one the story was really built for. The distinction matters commercially, and the tradeoffs between running one master with variants versus separate originations are laid out in detail in the guidance on commissioning across three markets simultaneously.
Whichever model you pick, name the approval language separately and write it into the agreement. The approval language is the language your acceptance documents, quality reports, retake notes and delivery certificates are written in. It is almost always English even when nothing else is. Conflating the two is what produces a commission where your team believes it approved a script it never actually read.
Build the Brief So It Survives the Gap
A brief written for a team that shares your language can lean on shared assumptions. A cross language brief cannot. Everything implicit has to become explicit, because the person interpreting it is working from the document and not from the conversation that produced it. This is unglamorous work and it is where most of the risk is retired.
Specify the emotional beat rather than the line. Instead of naming a line of dialogue you want, name what the audience should feel at that second and what information they should now hold. Instead of describing a character as cold, describe what the character does when challenged and what the character never does. A generation operator or a writer in the target market can build a culturally correct version of a described behaviour. They cannot reverse engineer a described adjective, because adjectives carry different weight across languages and the drift is invisible to both sides.
Specify the status system. Vertical drama runs on status, and status is signalled differently in every market. Who defers to whom, what forms of address exist, which gestures read as disrespect, what a family hierarchy permits in public versus private. Write down the intended power relations in plain structural terms and let the writer translate them into the local grammar of status. The broader discipline of briefing an AI native production partner is covered in the document that gets you the series you want, and every principle in it tightens when a language gap is in play.
Specify what must not happen. Negative constraints travel across languages far more reliably than positive ones. No on screen depiction of a specific practice, no reference to a named institution, no resolution that depends on a legal mechanism that does not exist in the target market. A short list of hard prohibitions is worth more to a cross language production than three pages of aspirational tone notes.
Appoint the Native Reader Before the First Script Exists
The single highest leverage move in a cross language commission is hiring one person, on your side of the table, who reads the target language natively and is accountable to you rather than to the production partner. Not a translator. A reader with commissioning judgement who can watch an episode and tell you whether the confession landed.
This person needs three qualifications. Native or near native fluency in the target language including its register and slang. Enough familiarity with the format to know what a vertical drama beat is supposed to do. And independence from the production partner, because a reader employed by the party being reviewed is not a control, it is a formality. The cost of this role is small relative to a series budget and it is the cheapest insurance available on a cross language order.
Bring them in at outline stage, not at delivery. A native reader who sees the outline, the character bible and the first three scripts can correct a drift that costs almost nothing to fix at that point. The same reader seeing the same drift at episode 40 is delivering bad news about work that is already generated and edited. Write their review points into the schedule as gates with a defined turnaround.
Set Quality Gates That Do Not Require Dialogue Comprehension
A surprising amount of vertical drama quality is language independent, and a non speaking commissioning team can assess all of it directly. Build your gate structure so that these checks belong to you and only the language dependent checks are delegated.
The checks you can run yourself include character consistency across episodes, wardrobe and continuity discipline, framing and legibility in the vertical frame, cut rhythm and pacing, audio mix behaviour on phone speakers, hook construction in the first seconds, and the placement of cliffhangers relative to the monetisation points. You can watch an episode with the sound off and learn most of what you need about whether the visual craft is holding. You can watch it with sound and no subtitles and learn whether the performance energy is present, because delivery energy reads across languages even when meaning does not.
The checks you must delegate are dialogue quality, naturalness of register, humour, cultural specificity, name and address correctness, and whether a scene reads as sincere or as parody. Give these to the native reader with a fixed scoring rubric so that the output arriving back to you is comparable across episodes. A score with a written justification is auditable. A verbal assurance that it is fine is not.
Write the split explicitly into the quality plan so that there is never ambiguity about who owns a given failure. When a series underperforms, the first question is always which gate should have caught it, and a commission that has not documented the split will spend a fortnight arguing about that instead of fixing anything.
Run the Episode One Check in Both Directions
Before the full block goes into generation, produce episode one and run it through two separate reviews that are deliberately kept apart.
The first is the native reader review, unsubtitled, judged as a piece of drama in the target market against local comparables. The question is simple. Would this hold against what is currently performing on the platforms in this market, and where specifically does it fall short. The second is your own review, watched with subtitles you commission independently of the production partner, judged on structure. Does the hook work, does the escalation build, is the paywall moment placed where the story actually turns.
Then compare. Agreement across both reviews is the green light. Disagreement is diagnostic rather than fatal. If your structural review is positive and the native review is negative, the problem is usually in the dialogue layer and is fixable through rewriting without touching the arc. If the native review is positive and your structural review is negative, the problem is usually arc or pacing and the fix is more expensive, which is precisely why you are running this check at episode one rather than at episode thirty.
Keep the independent subtitle pass independent. Subtitles produced by the party whose work is under review will smooth exactly the roughness you are looking for, not through bad faith but because the person writing them knows what the line was intended to mean. The distinction between translating a text and adapting a product for a market is well established in the localisation field, and the W3C explanation of localisation and internationalisation is a useful short reference for anyone on the team who has not worked across languages before.
Write Acceptance Criteria a Non Speaker Can Apply
Acceptance is where cross language commissions quietly break down, because the moment of acceptance is the moment where somebody has to sign. If that person cannot read the deliverable, the signature is a formality and everybody knows it.
Fix this by making acceptance conditional on documented outputs rather than on personal judgement. Acceptance requires a completed native reader scorecard above a stated threshold for every episode. It requires a continuity report showing character and wardrobe consistency across the block. It requires a technical certificate covering resolution, aspect ratio, audio levels, stem delivery and subtitle file integrity. It requires the retake log showing what was flagged, what was fixed and what was accepted as is with a reason. Your signature then attests to the completeness of that evidence, which is something you can genuinely verify.
Set the thresholds before delivery, not during it. A threshold negotiated while a delivery is sitting in the queue is a threshold that will move. Define the remedy path for a failed episode in the same document too. Most of this belongs in the production agreement, and the components that a production agreement should carry are set out in the piece on what an AI vertical drama production SLA should actually contain, which applies without modification to cross language work.
Plan the Localisation Assets on the Way In, Not the Way Out
Even a natively originated series usually travels. The cheapest moment to prepare for that is during production, and the most expensive moment is eighteen months later when the source files have been archived by somebody who has since left.
Ask for split audio stems on every episode so that dialogue can be replaced without rebuilding the mix. Ask for the script in a timed format rather than a document, so that a future dub has timing to work against. Ask for character reference assets and the prompt or generation records that produced them, so that additional footage can be matched rather than approximated. Ask for a clean pass without burned in text so that on screen elements can be re rendered for another market. None of this costs much when it is specified at the outset and all of it is close to impossible to recover afterwards.
The same asset discipline is what makes a second season or a spin off viable in the original language, so it is not localisation spending, it is franchise spending. Treat the language your team does not speak as the first market rather than the only one, and the asset requirements follow naturally.
Know What to Do When the Reader and the Numbers Disagree
Eventually the native reader will say an episode is weak and the retention data will say it performed. Or the reverse. Both are real and neither should automatically win.
When the data is positive and the reader is negative, look at what the reader flagged. If it was dialogue naturalness or register, the series may be performing on plot while accumulating a quality reputation problem that will show up in season two conversion rather than season one retention. If it was cultural correctness, treat it as urgent regardless of the numbers, because the cost of getting that wrong is not measured in retention. When the reader is positive and the data is negative, the problem is usually upstream of language entirely, in hook construction, thumbnail, placement or pricing, and rewriting dialogue will not touch it.
The resolution is to hold both inputs and to decide explicitly which one is driving the next commissioning decision, in writing, so that the reasoning survives into the next order. A cross language slate that cannot explain why it renewed one series and dropped another is a slate that will make the same mistake twice.
Axis AI Studios Perspective
Axis AI Studios is an AI native vertical drama production studio based in the Netherlands, working with platforms, media companies and IP holders on scripted vertical series. A significant share of that work is commissioned by teams who do not read the language the series is delivered in, and the structures above are the ones we operate to rather than theoretical best practice.
Our position is that a commissioning team should never be asked to take language quality on trust. That means the brief carries structural and behavioural specification rather than tone adjectives, the native review sits on the client side of the table, the language independent quality checks stay with the client in full, and acceptance runs on documented evidence that a non speaker can verify line by line. It also means the localisation assets are specified at the start of production, because the value of a series that travels is decided long before anyone decides to travel it.
We are direct about the boundary. We control the production process, the quality standard, the review structure and the delivery. We do not control how a market receives a story, and we do not present cultural judgement as something a production partner can settle unilaterally. That is why the native reader belongs to the client.
If you are planning a commission in a market where your team does not read the language, or you are reviewing a delivery you cannot assess directly, we are glad to talk through the gate structure before anything is committed. Reach us at business@axisaistudios.com.
FAQ
Do I need a native speaker on my own team, or can the production partner supply one? You need one accountable to you. A reader supplied by the production partner is reviewing their own work, which makes the review a quality step inside the vendor rather than a control on the client side. The role does not have to be full time and it does not have to be a permanent hire, but it does have to be independent, and the reader should see the outline and early scripts rather than only the finished episodes.
Is it cheaper to produce one master series and localise it, or to originate separately in each language? One master with planned localisation is almost always the lower cost route across three or more markets, and it also concentrates your quality control effort in a language you can read directly. Separate origination produces a stronger result in a single priority market because the story is built for that audience from the first outline. The decision usually comes down to whether one market is clearly leading or whether several are being entered at once.
How do I judge dialogue quality if I cannot read the script at all? You judge it indirectly, through a scored native review with written justifications, and directly on everything that is language independent, which is more of the craft than most teams expect. Watch episodes without subtitles to assess pacing, framing, performance energy and cut rhythm. Then require the native reader scorecard as a condition of acceptance rather than as a comment, so that dialogue quality is evidenced rather than asserted.
Further Reading
For the economics of running one production across several language markets rather than funding each separately, the guide to commissioning vertical drama across three markets simultaneously covers how the master and variant structure is costed and where it breaks down.
For the production side of the asset discipline described in section seven, what localisation built into production from day one actually looks like sets out the script formatting, stem separation and generation decisions that have to be made before the first episode is built.
For teams deciding how far a catalogue can travel once the assets are in place, using AI to localize your vertical drama catalog covers market prioritisation and what a multi language rollout actually requires.
Let's set
the new standard together.
If you're working on something, we'd like to hear about it.