Como traduzir texto no ecrã em vídeos de formação
Training videos travel faster than ever. Video is expected to account for roughly 82% of all internet traffic by 2025 (Mindstamp). But training only scales globally if learners can actually understand what they see.
That is where many teams stumble: they translate the voiceover, maybe add subtitles, but leave on-screen text (labels, diagrams, UI callouts, safety warnings, charts) in the original language. The result is cognitive dissonance for learners who are trying to match spoken explanations to visual cues (Translated.com), and it can be a real risk in technical or compliance training where inconsistency causes mistakes.
Neste guia, mostrar-lhe-ei como translate on-screen text in training videos step by step, including tool choices, workflow estimates, formatting rules, and the most common pitfalls.
What is on-screen text localization in training videos?
Localização de texto no ecrã is the process of translating any text that appears visually inside the video frame, not just what is spoken.
Typical examples include:
- Slide titles and bullet points in a lecture recording
- UI labels in software walk-throughs
- Lower thirds with names and roles
- Callouts and annotations
- Charts, diagrams, and safety signage
- Open captions that are burned into the video
This differs from audio translation (dubbing) or subtitle translation because visual text often needs graphic replacement, dynamic overlays, or detailed editing if it is hard-baked (burned into frames).
Introduction: Why you must localize on-screen text
Why on-screen text translation is non-negotiable for global training
Training videos are an indispensable tool for education, onboarding, and skill development, with proven gains in retention, learning, and engagement (interproinc.com). But learners cannot benefit from those improvements if the text that carries key meaning stays in the source language.
Here is why training video text translation for on-screen elements is essential:
- Reduced cognitive load and better retention: When on-screen text matches a learner’s language, they spend less mental energy translating and more energy understanding (Translated.com).
- Higher comprehension for complex concepts: Diagrams, charts, and bullet lists often contain the real training content. Translating only the audio leaves critical information inaccessible (Think Branded Media).
- Consistency and risk reduction: In technical and compliance training, mismatched terminology between audio, subtitles, and visuals can lead to confusion, assessment failures, or safety risks (Translated.com).
- Acessibilidade: Proper text localization helps non-native speakers and improves access for deaf and hard-of-hearing learners, especially when paired with captions.
- Sound-off reality: 85% of videos on some platforms are watched with sound off (Mindstamp, Think Branded Media). If training is consumed in quiet offices or on mobile, visual text carries even more weight.
- Engagement benefits of text: Text overlays can be powerful. Companies have reported up to a 12x increase in conversion rates from video ads with text overlays (Mindstamp). The training parallel is clearer comprehension and stronger completion behavior.
Also, if your organization uses training to support products, the business case is hard to ignore: 72.4% of consumers are more likely to buy when information is available in their own language, and 42% will never purchase in a language they do not understand (interproinc.com).
The unique challenges of visual text localization
Translating on-screen text is harder than translating a script. These are the issues that most commonly break global training rollouts:
- Hard-baked text: If text is embedded directly into the video frames, it requires masking, re-creation, and re-rendering (Compass Languages).
- Text expansion: Many languages take more space than English. Spanish and German often expand by 20 to 30%, which can break layouts (verbalate.ai, idearocketanimation.com).
- Font and aesthetic integrity: Professional training needs consistent fonts, colors, and motion design across languages (ajsp.net, Storykit).
- Temporização e sincronização: On-screen text often appears with precise animations and must align with narration. That timing has to be preserved (Compass Languages).
- Cultural nuances: Even short phrases can be culturally off. Visual examples, symbols, and tone need to be appropriate for the target locale.
- Non-Latin and RTL scripts: Arabic and other right-to-left languages require layout changes and careful font support. East Asian scripts can require different spacing and typography approaches.
Prerequisites and essential tools for on-screen text translation
Foundational requirements and preparation
Before you start translating, gather these inputs. They determine speed, quality, and cost.
- Source video and project files: Ideally a high-resolution MP4 or MOV. Best case, original editable project files (for example, layered motion graphics).
- Video transcription: Accurate transcript of spoken content. Common formats include SRT ou VTT.
- Source text list: A complete inventory of on-screen text with timestamps, exact wording, context notes (what it labels or explains), and styling notes (font, color, size, position).
- Terminology glossary and style guide: Critical for technical training and brand consistency (Translated.com). Include approved translations for product features, UI terms, safety language, and role titles.
- Target language specifications: Character set and font coverage, reading direction (RTL for Arabic), cultural sensitivities, and formality expectations (for example, training tone in different locales).
- LMS compatibility requirements: Know what your Learning Management System expects, including video codec constraints, subtitle formats, and whether you need SCORM packaging for portability across LMS platforms.
Key software and platforms
You can translate visuals with many combinations of tools. The key is matching the toolset to the kind of text you have, meaning editable layers versus hard-baked.
AI-powered video translation and dubbing platforms
A strong starting point for most teams is an AI localization platform that handles transcription, translation, dubbing, and subtitle generation together.
- Vozo AI’s Video Translator: https://www.vozo.ai/video-translate
Editorial pick for training teams that want one workflow for multilingual rollout. It translates video into Mais de 110 línguas with natural dubbing, includes Clonagem de voz VoiceREAL™, facultativo LipREAL™ sincronização labial, and a built-in proofreading editor so humans can refine output in real time. - Vozo AI’s AI Dubbing: https://www.vozo.ai/dubbing
Useful when your priority is fast, natural voiceover replacement. It supports Mais de 60 línguas e Mais de 300 vozes de IA realistas, O texto é um texto de apoio, concebido para corresponder ao tom, ao ritmo e à emoção. - Vozo AI’s Voice Studio (Video Rewrite): https://www.vozo.ai/video-rewrite
Ideal when translation reveals a script issue or you need to simplify phrasing for readability. It lets you rewrite and redub voiceover using a text-based editor without re-recording. - Vozo AI’s Lip Sync: https://www.vozo.ai/lip-sync
Helpful when you dub training with a visible instructor, interviews, or multi-speaker scenes and want mouth movements to match the new audio. - Vozo AI’s Audio Translator: https://www.vozo.ai/audio-translator
Good when you have separate audio tracks or want to translate and preserve the original speaker’s voice, tone, and emotion.
Other platforms in the research set include Smartcat, Verbalate™, and ScreenPal, which offer variations of AI translation, subtitling, and dubbing.
OCR software and APIs (for extracting visual text)
If you need workflows where visual text extraction is step one, OCR is often the starting point:
- Google Cloud Vision (GCV): reported 96.7% OCR accuracy for lecture slide extraction (academia.edu)
- Tesseract: open-source OCR supporting Mais de 30 línguas (eecs.berkeley.edu)
- Abbyy FineReader: commercial OCR for documents and images
Software de edição de vídeo
For burned-in text replacement and motion graphics recreation:
- Adobe Premiere Pro (professional editing and overlays)
- DaVinci Resolve (free, professional-grade editing)
- After Effects (motion graphics and animated text)
Interactive video platforms
If you want overlays that can be updated without re-rendering the whole video:
- Mindstamp (clickable hotspots, branching logic, dynamic overlays, analytics)
CAT tools (for translation consistency)
For professional localization teams and translators, CAT tools help manage translation memories (TM), termbases (TB), and enforce consistent terminology:
- Estúdio SDL Trados
- MemoQ
- Wordfast
Step-by-step instructions for translating on-screen text
Below is a practical workflow for translating on-screen text in training videos, from discovery to final export. I’m including realistic time ranges so you can plan resourcing.
Phase 1 (estimated time: 1 to 5 hours per 10 minutes of video): text identification and extraction.
Phase 2 (estimated time: 2 to 10 hours per 1,000 words): translation and quality assurance.
Phase 3 (estimated time: 5 to 20 hours per 10 minutes of video): re-integration and video localization.
Fluxo de trabalho passo a passo
Identify every on-screen text element
Start with a frame-by-frame review. Your goal is completeness.
Capture titles, lower thirds, labels, callouts, slide text in screen recordings, chart and diagram labels, and brief flashes of text during transitions.
Build a timestamped list and note the exact text, appearance time and duration, font family (or closest match), color and size, approximate position, and animation type (fade in, slide, type-on).
Dica de especialista: animated sequences are where teams miss text most often, especially text that appears for less than a second.
Use OCR to extract text, then validate it
OCR speeds up the inventory process, especially for slide-heavy training. Export frames or short segments containing text as high-resolution images (PNG or JPEG), run OCR (Google Cloud Vision or Tesseract), then manually verify the output.
Verification matters because OCR accuracy drops with low resolution, stylized fonts, motion blur, or complex backgrounds (stacks.stanford.edu).
Preprocessing helps: grayscale conversion, binarization, noise reduction, and correcting uneven lighting can improve OCR results (stacks.stanford.edu).
Data point: Google Cloud Vision has been reported at 96.7% accuracy for lecture slide extraction (academia.edu), but that is under favorable conditions, not worst-case motion graphics.
Add context notes before translation
Translation quality rises when linguists understand purpose and context. For each text segment, include what it refers to (for example, “label for power button”), what the learner should do with it (instruction versus concept), and whether it must match a UI term from the product.
Cross-reference with the spoken script so the visual text and narration stay aligned.
Conselhos de segurança: for medical, safety, or compliance training, a human review of extracted text is mandatory. OCR mistakes can turn into training errors.
Choose a translation method (HT vs. MTPE)
You have three practical options.
- Human Translation (HT): best for high-stakes training, nuanced messaging, or culturally sensitive content. Research notes HT is superior for contextual accuracy and appropriateness (al-kindipublishers.org). Error rate reported at 4.5% (aviewint.com).
- Machine Translation Post-Editing (MTPE): Neural Machine Translation (NMT) for a first pass, then professional post-editing. NMT is fast and affordable, but requires human quality control (aviewint.com). MTPE can increase productivity by up to 37% compared to translating from scratch (aclanthology.org).
- AI translation with built-in human refinement: Vozo AI’s Video Translator (https://www.vozo.ai/video-translate) fits well because it combines AI translation with a proofreading editor that supports real-time refinement, which is practical for training teams that need both speed and quality.
Dica de especialista: avoid using public NMT tools for confidential corporate training, because user content may be used for model training in some services (atanet.org). For internal onboarding, compliance, or customer data, treat privacy as a core requirement.
Enforce glossary and style guide rules
On-screen text is often short, which makes terminology consistency even more important. One inconsistent term on a diagram can undo trust in the whole module.
Lock key terms in a termbase (TB), use a translation memory (TM) to keep recurring phrases identical, and apply the same style rules as other training materials (capitalization, formality, measurement units).
This is especially important for compliance and technical training where ambiguity is costly (Translated.com).
Run linguistic quality assurance (LQA)
At minimum, use a native speaker reviewer for accuracy and completeness, grammar and fluency, cultural appropriateness, and tone alignment with training intent.
This is also where you catch text expansion problems early. Plan for 20 to 30% longer text than English in languages like Spanish and German (verbalate.ai).
Conselhos de segurança: in critical fields, LQA should include a subject matter expert in the target language, not only a linguist.
Re-integrate translated text using the right technique
This phase is where teams realize that “translation” is also design and engineering. You typically mix three techniques depending on the training format and constraints.
Legendas e legendas ocultas: Generate and translate subtitles, export to SRT ou VTT, and keep files in UTF-8 encoding to support multilingual characters (Translated.com). Readability guidelines often cited for training include a maximum of 37 characters per line e two lines max, plus a maximum of about six seconds on screen (ajsp.net). Also decide whether you need open captions (burned in) or closed captions (toggleable) (interproinc.com).
Burned-in text replacement (graphic overlays): Mask or remove the original text, recreate translated text as a new layer, match the original font/color/positioning, and replicate animation timing. This is where hard-baked motion graphics can require frame-accurate adjustments and re-rendering (Compass Languages).
Dynamic text overlays (interactive video): Use platforms like Mindstamp to add translated overlays, hotspots, and branching logic. Keep phrases concise (Storykit), use readable sans-serif fonts and high contrast (Mindstamp), place overlays so they do not block key visuals (Storykit), and time them so learners can read comfortably (Mindstamp suggests long enough to read twice).
Forward-looking note: Vozo AI’s Visual Translate (announced March 12, 2026 via TMCnet) is designed to detect, translate, and preserve layout, style, and animations directly from the video file. In an alpha phase, it reportedly reduced localization time by over 96% for a multinational manufacturing company (TMCnet, March 12, 2026).
Dica de especialista: design for localization from the start. Keeping text in editable layers avoids labor-intensive rework (Compass Languages).
Export, test in your LMS, and do an end-to-end review
Export in formats needed for distribution (often MP4), plus LMS-specific codecs and packaging if required, including SCORM compatibility when needed.
Do a full-context review: a native speaker watches the full video, all on-screen text is translated and readable, subtitle timing constraints are respected, overlays do not obscure critical visuals, and everything is tested across target devices, operating systems, and LMS platforms.
If your team edits on mobile or needs quick overlay adjustments, Vozo AI’s BlinkCaptions Video Editor can help with captions and overlay-style text on the go: https://www.vozo.ai/blinkcaptions
Pros and cons of the main localization methods
Method 1: Subtitles and closed captions
Prós
- Fast and affordable (colossyan.com)
- Strong accessibility benefits
- Helps with sound-off viewing (Mindstamp)
- Can improve completion rates (idearocketanimation.com)
Contras
- Adds reading load and can distract from visuals (colossyan.com)
- Does not fix cognitive dissonance when charts and labels remain untranslated (Translated.com)
Method 2: Burned-in text replacement (graphic overlays)
Prós
- Most seamless learner experience
- Eliminates visual-language mismatch
- Preserves professionalism and training clarity
Contras
- Most time-consuming and costly
- Hard-baked motion graphics can require extensive re-rendering (Compass Languages)
- Less flexible once exported
Method 3: Dynamic text overlays (interactive overlays)
Prós
- Easy to update translations without re-rendering the base video
- Supports engagement with hotspots and branching (Mindstamp)
- Provides analytics to optimize training (Mindstamp)
Contras
- Requires platform support and careful UX design
- Overlay placement and timing errors can harm comprehension
Common mistakes to avoid in on-screen text translation
These are the repeat offenders that inflate budget and reduce training effectiveness:
- Underestimating text expansion: Ignoring the 20 to 30% expansion range leads to cramped layouts and truncation (verbalate.ai).
- Ignoring hard-baked text early: Discovering burned-in labels late forces expensive re-editing.
- Poor readability: Bad contrast, tiny fonts, or busy backgrounds can violate WCAG-oriented accessibility practices (ajsp.net, Compass Languages).
- Inconsistent terminology: Skipping glossaries and style guides causes different translations for the same term (Translated.com).
- Direct machine translation without post-editing: Raw NMT can be inaccurate or culturally off (aviewint.com).
- No native speaker review: Final video review catches real-world issues that text-only review misses.
- Suboptimal subtitle formatting: Ignoring character-per-line and duration rules creates cognitive overload (ajsp.net).
- Obscuring important visuals: Poor overlay placement blocks diagrams or UI elements.
- Failing to test across devices and LMS: Layouts can break on mobile or inside LMS players.
- Ignorar as nuances culturais: Literal translation can confuse or offend in some locales.
Troubleshooting common issues
Text expansion causes layout issues
Problema: translated text runs off-screen, overlaps elements, or feels cramped.
Soluções:
- Adjust font size or weight slightly, keeping readability intact.
- Rephrase or condense with a linguist while preserving meaning.
- Redesign the layout with more negative space.
- Break into multiple lines, but avoid exceeding two lines for readability.
OCR inaccuracy (poor extraction)
Problema: OCR returns garbled or incomplete text.
Soluções:
- Improve image quality and export higher-resolution frames.
- Pre-process images (grayscale, binarization, noise reduction) to improve OCR results (stacks.stanford.edu).
- Manually transcribe hard cases.
- Try a different OCR engine (Tesseract versus Google Cloud Vision).
- Segment extraction into smaller chunks (word-level rather than full blocks) for stylized text.
Inconsistent terminology in translations
Problema: the same concept is translated multiple ways.
Soluções:
- Enforce a project glossary.
- Use CAT tools with TM and TB to auto-apply consistent terms.
- Add a post-editing pass focused only on terminology consistency.
Readability issues (subtitles and overlays)
Problema: text is hard to read due to font, color, or contrast.
Soluções:
- Follow WCAG-style contrast guidance (commonly cited target: 4.5:1 for normal text).
- Use clean sans-serif fonts (Arial, Helvetica, Lato) (Mindstamp).
- Add a semi-transparent background box or subtle drop shadow (Mindstamp).
- Adjust font size with expansion constraints in mind.
Synchronization or timing errors for overlays
Problema: text appears too early or late, or disappears too quickly.
Soluções:
- Make frame-accurate timing adjustments in your editor.
- Review against audio and key visual cues.
- Extend display time for complex text so learners can comfortably read it (Mindstamp suggests long enough to read twice).
Corrupted characters in translated text
Problema: question marks or strange symbols appear, especially in non-Latin scripts.
Soluções:
- Assegurar UTF-8 encoding for SRT, VTT, and exported text assets (Translated.com).
- Use fonts that support the target script (for broad coverage, teams often choose fonts like Noto Sans).
- Confirm your operating system and software environment supports the target language.
Frequently asked questions (FAQ)
What’s the difference between translating spoken audio and on-screen text?
Spoken audio is translated via dubbing or subtitles. On-screen text is visual content inside frames (labels, charts, titles) and often requires graphic replacement or dynamic overlays, not just subtitles.
Is machine translation good enough for on-screen text?
NMT is fast and cost-effective, but raw MT output can miss nuance and precision needed for training. MTPE is strongly recommended, especially for critical information (aviewint.com).
How much does it cost to translate on-screen text?
Cost depends on video length, graphic complexity, number of languages, and whether text is editable or hard-baked. AI-driven solutions can reduce costs by 80 to 95% in some multilingual video production scenarios (colossyan.com), but burned-in graphic replacement remains labor-intensive.
How long does the process take?
It varies by density and complexity. Re-integration alone can take 5 to 20 hours per 10 minutes of video when done manually. Emerging generative workflows like Vozo AI’s Visual Translate aim to cut localization time by over 96% in some cases (TMCnet, March 12, 2026).
What are hard-baked subtitles or text, and how do you deal with them?
Hard-baked text is permanently embedded in the image. Translating it typically requires masking and overlaying translated graphics. The best fix is prevention: design with editable layers from day one (Compass Languages).
Can AI automate the entire process?
AI can automate transcription, initial translation, subtitle creation, dubbing, and some visual replacement. But human review remains crucial for accuracy, cultural relevance, and professional quality, especially in training.
How do you ensure translated text is readable?
Use clean sans-serif fonts (Arial, Helvetica, Lato), strong contrast guided by WCAG principles, and add drop shadows or semi-transparent background boxes when needed (Mindstamp, Storykit).
What file formats matter most?
- Legendas: SRT e VTT are widely supported (Translated.com).
- Graphics: PNG is common (especially with transparency).
- Codificação: UTF-8 is essential for multilingual character support (Translated.com).
How can you make original training videos easier to localize?
Design for localization (Compass Languages, verbalate.ai):
- Keep text in editable layers
- Avoid hard-baked text
- Allow space for 20 to 30% expansion
- Use modular graphics
- Write scripts clearly and avoid idioms
What role do interactive video platforms play?
Tools like Mindstamp let you add dynamic, clickable overlays that are easier to translate and update without re-rendering the whole video. They also provide analytics on learner interaction (Mindstamp).
A practical workflow recap and recommended tool stack
If there’s one operational takeaway, it’s this: translating audio alone is not enough. To scale learning globally, treat on-screen text localization as first-class work, with the same rigor as the script.
A reliable workflow looks like this:
- Inventory and extract every text element (manual review plus OCR)
- Translate with a method matched to risk (HT or MTPE)
- Enforce glossary consistency and run LQA
- Re-integrate using the right approach (subtitles, burned-in replacement, or dynamic overlays)
- Export, test in your LMS, and do a native-speaker final viewing pass
If you want a fast, practical starting point that combines translation, dubbing, subtitles, and human-in-the-loop refinement, Vozo AI’s Video Translator is a strong editorial pick for training teams managing multilingual rollouts: https://www.vozo.ai/video-translate.
Pair it with Vozo AI’s AI Dubbing (https://www.vozo.ai/dubbing) and Sincronização labial (https://www.vozo.ai/lip-sync) when you need natural voice and on-camera realism across languages.
Done well, translating on-screen text does more than localize a video. It removes friction, reduces mistakes, and gives every learner the same clarity, regardless of where they are or what language they speak.
