Visual Translate by Vozo - Translate text in your videos without recreating visuals

@lightfield Hi everyone — I’m Josie, the PM & designer behind Visual Translate at Vozo.

Really excited that Visual Translate is finally live after several weeks of development and early user trials.

Here are a few sample demos:

• DJI promo video

• A slide-based video

• A training video

• A Gemini intro video

• Geoffrey Hinton’s Royal Institution talk on AI

You can also check out a short How-to video showing how it works.

Over the past few weeks, users from different industries have already used Visual Translate to localize videos such as medical explainers, internal training, and safety instruction videos. It’s exciting to see it being used in real workflows.

Happy to answer any questions! Feel free to ask about how Visual Translate works under the hood, or tell us what kind of videos you’d like to translate.

@lightfield Translating the actual text inside videos feels like a huge missing piece in video localization. Subtitles and dubbing solve the audio layer, but the visuals are where a lot of the real information lives in explainer and slide videos.

Keeping the translated text editable while preserving the layout seems like the hardest part of that problem.

Curious how often the system has to recreate visuals versus just replacing text elements.

@lightfield Congrats on this to your team! Quick question: how does editable text ensure brand consistency across 110+ languages without manual tweaks every time?

@lightfield congrats on the launch bro! super useful tool

@lightfield Tried your product with Eng to Vietnamese translation and honestly, the accuracy surprised me a bit. It reads quite natural and gets the context right, which is not easy.

The only thing is the waiting time feels a bit long. If I’m testing a few times in a row, it starts to slow things down. Might be worth looking into scaling or optimizing the processing a bit so users don’t have to sit in the queue too long.

Overall though, this is really solid.

The project was officially green-lit in December, when we started the actual product design. Our founder CY, Tech Lead Fei, and I worked closely together to define the product direction and what Visual Translate should really be.

I also want to give a big thanks to our former engineer Yetong, who co-created the editor UX with me. Building the editing experience was one of the most challenging parts. There really wasn’t anything comparable on the market to reference.

Designing something this new has been incredibly exciting for me. I’m genuinely proud that we were able to create an editor that makes this workflow possible.

I also want to thank the engineers who made the product actually come to life.

Our frontend engineers Junjun and Kiya built the entire frontend workflow and the Visual Translate editor. As mentioned earlier, this is a completely new type of product — there wasn’t really anything similar on the market to reference. Many of the interactions and editing behaviors had to be designed and implemented from scratch.

Their work turned these ideas into a real, elegant editing experience. The editor you see today — smooth, intuitive, and flexible — would not exist without their craftsmanship and attention to detail.

I also want to thank our backend engineer Lucas. He built the backend architecture for this feature almost entirely by himself. At the same time, he was also responsible for several other feature developments across Vozo, which made the work especially demanding.

Despite that, he managed to design and implement a solid backend foundation for Visual Translate. We’re very grateful for his dedication and the reliability of the work he delivered.

Finally, I want to give another special thanks to Naro, who served as the engineering lead for this entire project.

Her role went far beyond frontend development. She was deeply involved in shaping the overall technical approach — from algorithms and pipelines to the product workflow. She also coordinated engineering progress across the team and kept the project moving forward.

Thanks to her leadership and persistence, the whole project was delivered on time and at a very high quality. We’re incredibly grateful for everything she contributed to making Visual Translate possible.

We’d also like to thank our QA engineers Menghan and Meixia, who were responsible for testing this entire feature.

As a brand new product, there were many aspects that needed to be tested — from product logic and interaction flows to algorithm performance and the quality of the generated results.

The first version of this feature was especially challenging. The system went through multiple rounds of model and algorithm updates, and the behavior could change significantly between iterations. That made the testing process far more demanding.

Thanks to their persistence and attention to detail, we were able to keep improving the system while maintaining stability. Their work played a crucial role in making sure Visual Translate could launch on time with reliable quality.

I’d also like to thank our website designer Ushuan. She led the upgrade of our entire website’s visual design and was almost entirely responsible for the design, build, and implementation herself.

For example, the landing page for Visual Translate was designed and built by her. She turned a complex product concept into a clear and elegant presentation that helps people quickly understand what this feature can do.

If you’re curious to see it, take a look here:
https://www.vozo.ai/visual-translate

@jessica_miller_7 Great question — especially since different languages can vary a lot in length. For example, Chinese text can become much longer when translated into English.

Our system analyzes the video frame, text length, and layout to compute a new layout that fits best. It can automatically adjust font size, reflow the text, and handle line breaks.

This way, the translated text stays within the visual boundaries and keeps the video looking clean and natural.

@jessica_miller_7 Nice catch! This is where the magic happens. Give it a try and you’ll see how deeply our AI model understands the correct layout based on the surrounding context and text.

@jessica_miller_7 Great question, Jessica! I can tell you’re a localization expert 😄 Hope Josie's reply helps. And feel free to give it a try, would love to hear what you think!

@ali_goldberg Hi Ali, thanks for your kind words! Visual Translate Beta doesn’t support Hebrew yet, but our Translate & Dub feature does. Feel free to give it a try and let us know what you think :)

@ali_goldberg Thanks for the encouragement! We don’t currently support Hebrew for the Visual Translation feature because it’s a right-to-left language, which makes layout reconstruction more challenging. In some cases, visual elements may also need to be flipped to look correct.

That said, we’re definitely working to improve our AI models and system so we can support this important language in the future.

@josie_oy Thanks for highlighting this important feature! This will definitely help make it production-ready for users.

@josie_oy A round of applause for our product and R&D team! Thank you for making it possible for important information to travel across language barriers! 🎉

@flora07  In most cases, if the text is visible to the human eye, Vozo can detect and translate it.

Very small or faint text can sometimes be more challenging, and like any model we can’t guarantee perfect handling for every edge case. We’re continuously improving the detection and translation quality to make it more robust over time.

@flora07 One more thing worth mentioning: it’s not a one-shot process. If the model misses some text, you can select the region and trigger a more detailed detection just for that area.

This greatly increases the chances of capturing and translating the text correctly. More details are available in our docs.

@flora07 From our testing, small text is often detected surprisingly well. If anything gets missed, Visual Translate lets you manually select the area and trigger translation for that region.

@lily_liu8 Vozo can detect and translate explanatory text that appears inside videos.

However, we usually don’t automatically translate screenshots or UI elements embedded in the video. In many cases those are meant to stay exactly as they are.

If you do want them translated, you can manually select the text area in the editor and click “Regenerate” to translate it. Our editor is designed to be flexible, so you can easily adjust and translate elements that weren’t processed automatically.

@lily_liu8 Hi Lily, thanks for your question. as @josie_oy replied, we currently dont support automatically translate screenshot, but you can select to add. Here is a how-to video, hope it helps :)

@lily_liu8  Great question! Our model tries to infer whether text should be translated based on the context. For example, logos or text that belongs to real-world objects are usually left unchanged.

Screenshots can vary, so it may depend on the specific case. But you can always manually tell Vozo which areas you want or don’t want translated in the editor.

@dmitry_zakharov_ai Thanks!

We’re launching a new feature that translates on-screen text inside videos and makes it editable after translation.

This means you can now add localized subtitles, dubbing with lip-sync, and localize text on screen all in one place. In practice, it lets you fully localize a video instead of just translating the audio. On the other hand, HeyGen focuses on AI Avatar + Dubbing with LipSync for translation.

Give it a try :)

@dmitry_zakharov_ai Great question! Vozo focuses on video localization — translating voice, subtitles, and on-screen text so a video can be fully adapted for different languages.

HeyGen is more focused on AI avatars / digital humans. There’s some overlap around dubbing, but the core use cases are a bit different.

@sandy_liusy Thanks so much, Sandy, for your kind words! Really appreciate it! looking forward to hearing your feedback.

@sandy_liusy Thank you so much for the support Sandy, really appreciate it!

We’d love for you to give it a try, and we’re also very curious to see what kind of videos you might use it for or share with it. Looking forward to hearing how it works for you.

@thea5 Thanks for asking! For visual translation, slides and product demo–style videos work best. This includes content such as e-learning, training materials, and marketing videos.

At the moment, it doesn’t work perfectly for videos with animated backgrounds or moving text. Like those entertainment callouts. We’re actively working to improve those cases and bring a more universal experience to all users.

@thea5 Great question!

Right now the current version works best with slide-style and explainer videos, where a lot of key information appears visually on screen.

Think of scenarios like training materials, presentations, product introductions, financial briefings, or talking-head videos with text overlays. These formats are usually information-heavy, with slides, labels, diagrams, or callouts that stay on screen long enough for the system to detect and translate while preserving the layout.

Where it can still struggle today is with highly dynamic visuals, like moving text or complex animated backgrounds. We’re actively improving those cases so the experience becomes more universal across different video styles.