How to make a video in a language your fonts do not support
Give Vizard Agent the script in its own language. It checks which installed typefaces actually render that script, downloads and installs a proper one if none do, spells numbers phonetically so the voice reads them correctly, and proofs the captions on a real frame.
What is the short version?
Producing a video in Thai, Arabic, Hindi or Korean fails in two dull places long before anything creative happens: the captions render as boxes because no installed font covers the script, and the voice reads "100W" as something that is not a word.
- Go to Vizard Agent and give the script in its own language.
- Describe the voice you want in that language.
- Flag the numbers and units, which are where it goes wrong.
What do you need before you start?
The script as it should appear, and a description of the voice. Vizard Agent handles the typography and the pronunciation, but it needs the text in the target language rather than in English for translation, because a translated script and a written one read very differently aloud.
- The script. Written in the language, not translated.
- The voice. Age, register, pace, accent.
- The numbers. Flagged, with how to say them.
- The hook. Both spoken and on screen.
- The footage. Or the images to build from.
What do you type into Vizard Agent?
Write the script in the language and say how it should sound. Vizard Agent chooses the typeface and tests the pronunciation itself, so what it needs from you is the register — a casual creator voice and a corporate announcer are completely different requests.
Prompt
Variants worth knowing:
- "Not an announcer." Default voices skew formal.
- Flag the numbers. "100W" needs saying, not spelling.
- Say where text must not go. Over the product, usually.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real Thai-language product video. Two unglamorous problems dominate the whole session: getting the numbers and units said properly by the voice, and finding a typeface anywhere on the machine that renders the script at all.
- Inspects the uploaded footage and extracts frames to see it.
- Checks the original audio and the voiceover options.
- Searches for voice candidates in the target language.
- Generates the voiceover on two different engines for comparison.
- Evaluates both takes for pronunciation and tone.
- Regenerates with the number written out phonetically so the unit is read correctly.
- Generates further candidates and evaluates them again.
- Checks the silences in each voice file and gets word timings.
- Paces the voiceover to match the footage length and re-checks it still sounds natural.
- Searches for and downloads a music bed and measures every stem's loudness.
- Checks which fonts on the system support the script — and finds none suitable.
- Installs proper typefaces for the language and verifies the files.
- Transcribes the paced voiceover for precise subtitle synchronisation.
- Extracts full-resolution frames and inspects the layout and safe zones.
- Renders preview frames with the subtitles and inspects how the script actually renders.
Step six is the fix nobody anticipates. A voice reading "100W" in Thai produced something wrong, so Vizard Agent wrote the unit out phonetically in the script fed to the voice — the same trick that gets a brand name pronounced right, applied to a number.
What does the result look like?
The source footage this page is written from measured 1920x1080 at 60fps and ran 8.21 seconds — a short product clip cut into a vertical UGC piece with a Thai voiceover, a spoken and on-screen hook, and Thai captions in a typeface installed for the job.
What makes it work is that nothing in it looks like a translation. The captions render properly, the numbers are said the way a Thai speaker would say them, and the voice sounds like a person recommending something rather than an announcement.
When does this not work well?
Language support is uneven across every part of the pipeline, and the gaps sit in completely different places from one language to the next. Vizard Agent reports plainly what it cannot do, rather than rendering a row of empty boxes and hoping you will not look closely.
- Rare scripts. Fewer voices, fewer usable typefaces.
- Mixed scripts. Latin brand names inside another script.
- Numbers and units. Almost always need phonetic spelling.
- Line breaking. Some scripts do not break on spaces.
- Machine translation. A written script reads badly aloud.
How do you fix a result that came back wrong?
Say what you heard, or what you saw on screen, and be specific about which. Vizard Agent keeps every voice take it generated, the installed typefaces and the caption timings, so a pronunciation fix or a typeface change does not disturb the edit underneath.
- "That number is said wrong." Re-spelt phonetically and re-recorded.
- "The captions render as boxes." A supporting typeface is installed.
- "Too formal." Re-cast with a more casual voice.
How does Vizard Agent compare to doing it yourself?
By hand this is where non-English projects stall: the editor's machine has no font for the script, the voice tool mangles the units, and the captions have to be checked by someone who reads the language. Vizard Agent installs the typeface and proofs the render.
| By hand | Vizard Agent | |
|---|---|---|
| Typeface | Whatever is installed | Checked, and a proper one installed |
| Numbers | Read wrong, or avoided | Spelt phonetically for the voice |
| Voice choice | One take, accepted | Two engines compared and evaluated |
| Captions | Exported and hoped for | Proofed on a rendered frame |
Common questions
Which languages does this cover? Any with an available voice. Vizard Agent checks font support separately from voice support.
What if no font supports the script? It installs one. Vizard Agent downloads a proper typeface for the language and verifies it.
Why do numbers go wrong? Voices read digits, not meaning. Vizard Agent spells the unit out phonetically instead.
Can I write the script myself? Yes, and you should. Vizard Agent reads what you write rather than translating from English.
Will the voice sound natural? It compares candidates. Vizard Agent generates on more than one engine and evaluates them.
Can it match the footage length? Yes. Vizard Agent paces the voiceover and checks it still sounds natural afterwards.
What about right-to-left scripts? Handled separately. Vizard Agent deals with directionality and line breaking for those.
Will the captions fit? It proofs them. Vizard Agent renders a real frame and inspects the layout before committing.
Can it caption in two languages at once? Yes. Vizard Agent handles a second track, though the frame gets crowded.
Does the brand name still work? Give it phonetically. Vizard Agent applies the same approach to names as to numbers.
Why not just translate an English script? Because a translated script reads like a translation when spoken aloud. Writing it in the language, in the register people actually use, is the difference between a local video and a dubbed one.
Can I reuse the setup next time? Yes. Vizard Agent keeps the typeface and the voice choice for subsequent videos.
Does it check the finished video? Yes. Vizard Agent inspects rendered frames for the caption rendering and the safe zones.