Hiring Video Production in China 08/10

When an overseas brand hires a production team in China, communication quality is often reduced to one question: how good is the supplier's English? Language ability matters, but projects more often lose control because one comment contains three different decisions.

“This does not feel premium enough. Make the port clearer, and emphasize battery life more.” That sentence combines an aesthetic judgment, a shot task and a product claim. If the production team treats it as one ordinary revision note, the director may change the light, the editor may add a caption, and the producer may schedule a pickup. Yet nobody has established whether the battery statement is approved, which model it applies to, or what qualification it needs. The sentence has been translated; the decision has drifted.

The practical conclusion is that a cross-border video project needs a bilingual decision record, not merely more translation. Every item should first be classified as product fact, creative choice or execution action, then attached to an object, version, owner and closure condition. Translation makes the same approved decision legible in two languages. It cannot invent product facts, decide brand taste or determine whether a change is inside the production scope.

This guide addresses a specific situation: an overseas brand remotely commissions a China team to film a consumer-electronics or smart-hardware product video. The project includes Chinese and English scripts, storyboards, remote review, edit feedback, captions or voiceover, and landscape and portrait deliverables. It is a production-management method, not a language-proficiency test or legal and advertising-compliance advice.

Classify communication errors before translating them more completely

Before a note enters production, mark which layer it belongs to.

Layer 1: product facts and claims

This layer answers what the product is, what it does and under which conditions a statement holds. It includes:

  • brand, product, model, SKU, colour and release version;
  • feature names, operating steps, compatible devices and system requirements;
  • figures, units, test conditions, runtime, wattage, capacity, dimensions and materials;
  • certification, award, comparison, health, safety or environmental wording that requires a basis;
  • the names that actually appear in the app, firmware, interface and packaging;
  • required qualifications, warnings and prohibited expressions.

Neither a translator nor a production supplier should complete this layer through linguistic instinct. For advertising in the United States, the current FTC advertising guidance says ads must be truthful and non-deceptive and that advertisers need a reasonable basis for claims. That does not make one phrase suitable for every market or product category. The client should name the product, market or compliance owner who approves the fact source and its scope.

Layer 2: creative judgment

This layer answers how the audience should understand and feel. It covers information order, camera distance, performance intensity, light, colour, pace, music direction, caption tone and brand character.

“Stronger,” “more restrained” and “more like real use” are not translation errors, but they are not executable instructions either. The director or creative owner should turn the judgment into an observable change: make the action more decisive, reveal the product earlier, reduce camera drift, leave room for functional sound, or replace inflated adjectives with a specific outcome. The client approves the expression direction; the production team translates it into picture and sound.

Layer 3: execution action

This layer answers what changes, who changes it and when the issue closes. Examples include filming a side close-up of the port, replacing the caption at 00:18 with approved English, retaining the rhythm of V03 while removing an unconfirmed figure, or rebuilding the safe area for the 9:16 version.

An execution action must point to a shot, timecode, version or file. Adobe's current documentation shows that Frame.io review comments can be attached to a specific frame and located on the timeline. That solves location, not interpretation. The tool does not decide whether a comment is a factual correction, creative preference or scope change.

Once the three layers are separated, many disputes become easier to diagnose:

  • natural English with the wrong model number is a fact error;
  • two accurate translations with different tones present a creative choice;
  • an approved caption placed in an old master is an execution error;
  • adding a new performance promise is not “polishing the translation” but creating a new claim;
  • asking for a portrait version is not “just another export” if framing, captions and the delivery scope must change.

Build a product-fact master before sending a Chinese document to a translator

The project does not need every chat message translated. It needs a controlled set of fields that affect shots, captions and buyer understanding. Put them in a product-fact master, with one traceable entry for each concept:

  1. Concept ID: for example, BATTERY-RUNTIME-01, so the same concept is not renamed across files.
  2. Chinese source term and definition: explain what it means in this product, not merely the word.
  3. Approved English: include capitalization, hyphenation, singular or plural form and protected brand terms.
  4. Applicable model and market: prevent a specification from one SKU reaching another.
  5. Figures, units and conditions: retain the source value, display value, conversion rule and operating or test conditions.
  6. Permitted and prohibited wording: state what can be said and which stronger expression lacks approval.
  7. Interface and packaging match: identify the button, menu and label that will appear in the frame.
  8. Fact owner and approval date: record who approved it and against which source version.

This need not become a vast terminology database. Microsoft's current language resources provide reusable terminology, UI strings and localization style guides. The European Commission's translation service likewise describes quality as accurate, clear and fit for purpose, supported by uniform principles and terminology resources. For a product film, the minimum useful glossary may contain only 20 or 30 concepts. Each must still answer, “What exactly does this term mean here?”

Numbers need protection too. 5000 mAh, 5,000 mAh and “about two days of use” are not interchangeable facts. Converting Celsius to Fahrenheit or millimetres to inches is not merely linguistic. Preserve the original value, conversion method, rounding decision and market-facing display separately. Translation may change syntax; it must not silently change the product.

A bilingual script needs four answers, not two columns

A conventional bilingual script has a Chinese-source column and an English-translation column. A remote production needs two more answers: how the image supports the statement, and who approved the fact.

Define each script unit with:

  • source intent: what the audience should understand, instead of treating a Chinese paragraph as an untouchable literal template;
  • approved expression: the current Chinese and target-language wording, with locked terms and qualifications marked;
  • visual task: the action, detail, interface or reaction that carries the evidence;
  • responsibility status: fact owner, language reviewer, creative approver and current state.

“Install it with one hand” should not move into production simply because it has a fluent English translation. The script also needs to define which hand, the product state, whether a tool is required, which installation positions are covered and which steps must remain visible. If the actual operation requires the other hand to hold a component, the conflict is between claim and picture, not between languages.

ONCE's existing guidance on English voiceover and subtitle scripts explains how copy, timing and picture relate. The additional procurement step here is to lock source intent, fact ownership and the visual task before voice or caption production begins. Otherwise, post-production merely typesets an unresolved product decision.

Decide what must be bilingual and what can be approved by looking

Not every decision needs a long bilingual explanation. Effective remote production reduces unnecessary translation and reserves text for matters that the image alone cannot establish.

Lock these items bilingually

  • product name, model, feature, figure, unit, compatibility and qualification;
  • voiceover, captions, on-screen copy, end cards, CTA and disclaimers;
  • the claim assigned to each shot and what must not be implied;
  • whether a client comment is mandatory, an option to test or reference only;
  • approval, rejection, pending and out-of-scope status;
  • file purpose, target market, language, aspect ratio and delivery state.

Approve these visually, while still naming the shot

  • composition, product angle, action, eyeline and performance intensity;
  • light softness, reflection control, material rendering and background depth;
  • continuity, prop position, interface brightness and hand occlusion;
  • edit rhythm, music entry and the relationship between picture and production sound;
  • whether product, performer and caption safe areas survive a change of aspect ratio.

Visual approval should still say more than “this works.” A useful record would be: “Approve the composition of S12 in Storyboard V05. Keep the action, raise display brightness, and retain model and UI copy from FACT-UI-03.” The visual judgment and the product fact remain connected without becoming the same decision.

If the overseas team will not attend the China shoot, first use episode 03's six remote approval gates. That guide decides what must freeze before filming, what belongs to the live set and what can wait for asynchronous review. This article takes the next step: after a gate closes, both languages must still point to the same fact and version.

Give storyboards and edit feedback the same executable grammar

From storyboard through rough cut, use one comment structure:

Object: V02 / 00:14–00:17 / Shot C06
Type: FACT / CREATIVE / EXECUTION
Observation: The English caption says “works with all devices,” but approved material lists only two system versions.
Required outcome: Replace it with the wording in TERM-COMPAT-02; retain the picture.
Owners: Client product owner confirms the fact, English reviewer confirms expression, editor replaces it.
Deadline and closure: By 18:00 China Standard Time on 8 October; close when caption, voiceover and end card match in V03.

This is an illustrative method, not a real client record. It is longer than “translation inaccurate, please improve,” but it prevents three new interpretations in the next cut.

Timecode needs a common reference. Is the note attached to a review file with a slate, or a clean export? What is the frame rate? Has a new shot shifted the sequence? If a scene at 00:18 in V02 moves to 00:23 in V03, copying the old timecode points to the wrong object. A robust note includes version, time range, shot ID and a short visual description.

Caption files need a time identity as well. The current W3C WebVTT specification defines captions, subtitles, chapters and other text tracks as resources aligned with audio or video time. A project need not use WebVTT, but the production principle holds: content, in point, out point, language and corresponding master belong together. A detached word-processing file is not a complete timed-text deliverable.

A broader remote post-production workflow still needs asset and version control. With bilingual fields added, the China producer should run an executability check before notes reach the editor: Does the comment point to the right object? Does it conflict with another instruction? Does it change an approved fact? Does the client need to make a new decision? The producer can organize and escalate; the producer should not secretly choose the winner among conflicting client instructions.

A filename should identify the asset, not prove how many revisions happened

final_v2_new_USE_THIS.mp4 fails because it does not identify language, market, aspect ratio, stage or approval status. A readable structure could be:

PROJECT_ASSET_LANG-MARKET_RATIO_STAGE_vNN_YYYYMMDD

For example:

NOVA_HERO_en-US_16x9_CLIENT-REVIEW_v03_20261007.mp4

NOVA_FEATURE03_zh-CN_9x16_APPROVED_v05_20261008.mp4

Adapt the fields to the project. A filename does not need to carry the entire database, but the record should distinguish:

  • project and asset task;
  • language and, where needed, market;
  • aspect ratio or platform version;
  • internal review, client review, approved or delivered state;
  • ascending version number and date.

“English” is not always one market version. The US and UK may share a language while differing in units, spelling, price, required qualifications or CTA. One Arabic language version can also serve markets with different business conditions. Content language, target market and destination channel should remain separate fields rather than collapsing into _EN or _AR.

One approver does not mean one commentator

The overseas team may need comments from product, marketing, brand, legal, sales and local-market colleagues. The production supplier should still receive one authoritative result. The named approver's job is not to know everything personally. It is to:

  1. route fact questions to the fact owner;
  2. route natural-language questions to the target-language reviewer;
  3. route creative disputes to the brand or creative owner;
  4. consolidate conflicts into one executable instruction set;
  5. make authorized decisions on new scope, delay and version change.

When two stakeholders disagree, the supplier should not choose the easier note. Mark the item DECISION REQUIRED, show the alternatives, explain the effect on product meaning and schedule, and give the latest decision time. Silence is not approval, and an emoji in a group chat should not lock picture.

The China side also needs named interfaces: who maintains the fact master, who runs bilingual meetings, who issues the written decision summary, who controls file access, and who releases the final notes to the edit or set. One producer may cover several roles on a small project, but responsibility must remain visible.

AI translation can accelerate a draft; it cannot approve a fact

Machine translation and generative tools can help with a first pass, term candidates, caption-length warnings, version comparisons and meeting-note organization. Protect product terms, figures, models, units and prohibited wording as controlled fields. Any newly generated qualification, performance promise, target user or stronger tone should return to the fact owner rather than being accepted because the English reads better.

Back-translation can reveal a difference, but it cannot prove correctness. Two texts produced by the same model can agree with each other and still depart from the product source. A stronger sequence is to run a fact pass against the product-fact master, have a target-language reviewer assess naturalness and audience understanding, then review captions, voice and UI in the actual picture.

Do not merge these approvals into one “translation passed” state:

  • fact approved: model, figures, functions, qualifications and picture agree;
  • language approved: the target text is natural, clear and consistent with terminology and brand voice;
  • creative approved: information order, pace and expression fit the version's job;
  • master approved: the correct text entered the correct master with valid timing, aspect ratio, sound and filename.

Completion means anyone can explain why this sentence appears in this version

A bilingual remote review is not complete because everyone saw the link or the English has been polished. At minimum:

  • critical product facts, terms, models, figures and units have a bilingual master and owner;
  • each script unit maps source intent, approved expression, visual task and status;
  • storyboard and edit comments distinguish FACT, CREATIVE and EXECUTION;
  • feedback identifies a version, timecode or shot, owner, deadline and closure condition;
  • language, market, aspect ratio, stage and version can be identified in the file or register;
  • one approver consolidates comments and the supplier does not silently resolve client conflicts;
  • AI drafts, back-translations and automated checks are not treated as product evidence;
  • final captions, voiceover, UI, end cards and delivered files match the approval record.

ONCE's ecommerce product video production solution brings product, use case, target market and multi-version delivery into one production scope. When Chinese, English or additional languages are involved, the product-fact master, bilingual script and version record should enter the project before filming—not after picture lock as a request to “translate it.” The public page describes a production approach available for discussion; it does not mean that every project includes the same languages, reviewers, platform versions or revision rounds.

If you are preparing to commission a China production remotely, send ONCE the product and factory or location, target markets, channels, launch date, sample status, current product materials, approved terminology, script or storyboard, approvers and time zones, plus the required languages and deliverables. We can first translate those inputs into a production scope, bilingual decision record and approval gates, then discuss filming and post-production. ONCE provides commercial video production and project coordination. It does not replace the client's product, legal, tax, platform or local-language specialists.

Source note: checked on 7 October 2026 against Adobe's current documentation for frame-specific Frame.io comments and review versions; Microsoft terminology and localization language resources; the European Commission Directorate-General for Translation quality framework; Google's current guidance on writing for a global audience; the W3C WebVTT specification; and the US FTC advertising guidance. Each source is used only within its stated tool, language-management or market context.

← Filming Ads in China: Global Usage Is Not One Right Before Paying a China Video Supplier: Match Four Records →