Frequently Asked Questions (FAQ)
Search common questions or submit a request to the TRONG support team.
What are TRONG's unique advantages compared with human translators?
TRONG processes large documents and complex file formats much faster and at a more reasonable cost than traditional human-centered workflows. It also uses AI models with a vast knowledge base, minimizing mistranslations caused by gaps in specialized domain knowledge. By avoiding distortions from an individual translator's subjective interpretation or style, TRONG delivers consistent, objective translations that remain faithful to the document's original purpose.
Why should I use TRONG instead of general large language models (LLMs) such as ChatGPT, Gemini, or Claude?
Large language models can produce excellent translations, but general-purpose LLMs have structural limitations in professional, commercial-grade translation workflows. TRONG is designed to address those limitations end to end. [object Object] [object Object] [object Object] [object Object] TRONG is more than a prompt service that simply asks an AI to translate. It combines document serialization and reconstruction with dozens of cross-checking inference calls to preserve context throughout the document. This brings together the nuance of human translation with the efficiency of AI.
What exactly is the difference between High-Quality Translation and Quick Translation?
TRONG provides two translation options optimized for different user goals. [object Object] [object Object]
How are translation costs calculated? Is a separate subscription required?
No subscription plan or recurring payment is required. TRONG uses a pay-as-you-go model where you pay only for what you use. Translation costs are calculated from token usage, based on the amount of text extracted from the uploaded file and the number of images. Images with no text or no need for translation are automatically detected and excluded from processing, preventing unnecessary charges.
Are there limits on manuscript length or file size?
You can upload a single file up to 100 MB. If measured only by plain text, that is roughly equivalent to about 200 full-length novels, so ordinary commercial document translation is effectively not constrained by volume. However, if the uploaded document contains many high-resolution images or complex media elements that significantly increase file size, the practical amount of processable text may vary.
What file formats are supported?
TRONG supports a wide range of document formats so you can work with your existing files and data. Currently supported formats include: General document files: .txt, .md, .docx (MS Word), .odt (LibreOffice), .hwp, .hwpx (Hangul Word Processor) Spreadsheets and data: .xlsx (MS Excel), .csv, .tsv, and Google Sheets via URL Future updates will expand support to specialized formats, including PDF documents, subtitle formats for film and animation, and image-only formats such as comics and webtoons.
Do you support translation between languages with different text directions (LTR ↔ RTL), such as English and Arabic?
Yes. TRONG supports text-direction conversion throughout the document, including table structure, paragraph alignment, and text layout within images, when translating between LTR (Left-to-Right) and RTL (Right-to-Left) languages. For the most reliable results, use .docx (MS Word) whenever possible. Some formats, including .hwp, have limited native RTL support, so conversion results may vary.
When translating a multi-volume series, can the context from previous volumes be maintained in later volumes?
Yes. After translating the first document with TRONG's High-Quality Translation mode, you can use Context Extraction to extract metadata unique to the series. The extracted data includes a glossary, character information, and core plot details that span the series. Attach this data as reference material for later translation work to maintain consistent settings and tone and produce high-quality translations across every volume.
Can images within a document also be translated?
Image translation is available only with High-Quality Translation; Quick Translation does not translate images. TRONG analyzes the uploaded document and automatically selects only images that contain text and need translation. This prevents unnecessary charges for simple graphics, background images, and other elements that do not need translation.
Which AI models does TRONG use for translation?
TRONG does not rely on a single AI model. It uses a hybrid approach, combining models based on each task's difficulty and requirements. TRONG uses top-performing frontier models for sections that require complex contextual reasoning and lightweight, high-speed models for tasks suited to rapid parallel processing. A single document can involve dozens of cross-model calls, maximizing both translation speed and quality. In addition to AI language models, TRONG's own document engineering technology is central to the entire process, disassembling complex source formatting and file specifications and then reconstructing them accurately.
What level of translation quality and context preservation can I expect?
TRONG delivers advanced translation quality and consistency comparable to a professional translation team. [object Object] [object Object] [object Object]
Can files downloaded after TRONG translation be used immediately for publishing or commercial use?
Data-oriented formats such as .txt, .md, .xlsx, and Google Sheets provide completed outputs that require no additional work. For documents with designed layouts, such as .docx files, the translation itself is complete, but we recommend a final layout review and minor adjustments because text length varies between languages. Please note the following design-related limitations when downloading document files. [object Object] [object Object] Note: TRONG continues to improve its automatic formatting and layout correction technology.
Does TRONG add any watermarks, logos, or service marks to the translated files?
TRONG never adds watermarks, logos, the service name, or AI-generated notices to your translated documents, allowing you to use them freely under your own brand. Please note that formatted documents such as DOCX may require a final layout review due to text length variations across languages.
Does TRONG simply call AI models in sequence through an API?
No. TRONG is not a simple prompt-by-prompt execution service. To manage millions of tokens of context without losing information, a single translation request can activate dozens of cross-checking AI inference calls and multiple control modules. TRONG's translation workflow uses a proprietary three-stage pipeline. [object Object] [object Object] [object Object] High-Quality Translation also uses an additional module to analyze relationships between the full text and translation tables, producing results that simple API calls cannot match.
Why does High-Quality Translation cost more than typical AI translation services?
TRONG's High-Quality Translation goes beyond first-pass machine translation and serves the roles of both a professional translator and an editor. It also uses multiple analysis tools to understand document context and the scenes, dialogue order, and flow inside translation tables. High-Quality Translation uses multi-pass cross-validation and deep contextual analysis to produce results suitable for immediate commercial use, so it costs more than a typical AI translation prompt call. Compared with the high costs and months of waiting often associated with traditional translation agencies, it can be a far more practical and efficient option.
Does Quick Translation use only cheaper, lower-performance models to run faster?
No. Its speed comes not from sacrificing model performance, but from an advanced parallel and batch processing architecture. Large volumes of text are divided into logical units and processed simultaneously, while the best-suited AI models are coordinated for each task's difficulty and requirements. This reduces wasted model capacity while maintaining strong translation quality and processing text quickly.
Are internal tags and system code used to format files such as DOCX included in billable translation tokens?
No. TRONG disassembles the uploaded document and extracts only the text that needs translation and the essential formatting controls. Token usage is calculated only from the core data the AI needs to understand the context and translate. A plain-text (.txt) file may use slightly fewer tokens than a document whose original layout or text positioning must be preserved. The difference is small, however, and text-focused documents without complex design cost almost the same to translate as plain-text files.
In game translation tables, dialogue branches are often scattered and rows are not in chronological order. Can TRONG still follow the context?
TRONG relies on the unique ID/Key hierarchy as the primary source for scene grouping. Detected non-empty Speaker, Scene Path, Scene Name, and Description columns provide additional context. Processing modes that include scene and order analysis utilize these hints to reconstruct dialogue context, whereas Quick mode does not add dialogue-order or scene-order AI analysis. Following translation, the output restores the original row order to maintain table structure.
