The pipeline, and where errors creep in
Descript makes the timing trade-off explicit. Match timing fits the original speech timing and may add words or rephrase; Direct translation follows the wording more closely, and the dubbed audio may be sped up or slowed down to fit. Auto, the default, uses timing match for dubs. [1] Filmora's AI Translation has an Auto Match option, on by default, that aligns subtitles to the timeline. [6]
| Stage | What happens | Typical errors |
|---|---|---|
| 1. Transcribe | Speech recognition turns the original audio into text | Misheard product names, jargon, crosstalk; every later stage inherits these |
| 2. Translate | The transcript is translated, either literally or fitted to the original timing | Wrong terms, mistranslated idioms, lines that run longer than the original |
| 3. Synthesize | A stock or cloned AI voice speaks the translation | Odd emphasis, mispronounced names, speed changes to fit timing |
| 4. Lip-sync (optional) | The video is re-rendered so mouth movement matches the new audio | Artifacts with side profiles, hands or mics near the mouth, multiple speakers |
What each tool translates
| Filmora | Descript | CapCut video translator | |
|---|---|---|---|
| Outputs | Voice and text, or text only [6] | Translated captions, dubbed speech, lip sync [1] | Translated subtitles only; for voice, the page points to a separate AI dubbing workflow [9] |
| Languages | 23 source languages; 14 targets for voice and text; all recognized languages for text only [6] | Stock voices in 28 languages; recommended dubbing voices in 13 on Business and Enterprise [2] | Over 27 subtitle languages [9] |
| Lip sync | Videos under 5 minutes, voice and caption translation only [6] | Creator plan and above; one on-screen speaker at a time [1][3] | Not part of this tool [9] |
| Review before dubbing | Not described in the guide [6] | Translation proofread on Business and Enterprise only [1] | Edit subtitles before export [9] |
What it costs per minute
Filmora publishes the clearest rate: 16 credits per minute to translate voice and text, 5 per minute for text only. The same page also lists AI Translation at 5 per minute in another entry, so budget the higher figure for dubbing. [7] Credits from its smallest pack cost US$9.99 for 1,500, about US$0.0067 each, so a dubbed minute works out to roughly US$0.11; at the 100,000-credit pack (US$399) it is about US$0.064. [8]
Descript says dubbing on current plans uses AI credits but gives no per-minute rate; it suggests checking the Usage tab after a run. [2][4] Plans include 400 credits a month on Hobbyist, 800 on Creator and 1,500 on Business, and Creator and Business can buy 350 more for $35. [5][10] Re-translating after you edit the project uses more credits. [1] CapCut's video translator page describes the tool as free and lists no credit quota. [9]
- Worked example, Filmora: a 10-minute tutorial dubbed into Spanish and German is 2 x 10 x 16 = 320 credits. The Advanced plan's 1,000 monthly credits cover three such two-language videos (960 credits), with nothing left for captions or narration. [7][11]
- Worked example, Descript: run one dub of a short clip, read the Usage tab, and scale up before committing a long video, since the rate is not published. [4]
Lip sync: when it works and when it does not
Descript's lip sync guidance is the most detailed: use footage of one person speaking on screen at a time, facing the camera directly, well lit, with hands, microphones and hair away from the mouth. It is not supported with multiple speakers, multi-track sequences or avatars, very small faces or low-resolution footage give less crisp results, and rendering takes longer than translating. Run it as the final step, because later edits mean re-translating. [3]
Filmora limits lip sync to videos under 5 minutes, so a longer tutorial has to be split or dubbed without it. [6] For screen recordings with a small webcam bubble, lip sync adds little; a clean dubbed voice and accurate captions matter more.
Choosing a voice
Descript's dubbing step assigns each speaker label to a stock voice, and you can later change the speaker and regenerate. [2] A cloned voice of the original presenter sounds more continuous across languages but requires that person's consent; Descript's cloning rules and Filmora's credit cost for cloning are covered in our voice cloning guide.
CapCut's page claims 98% accuracy in its heading. That is the vendor's own unverified figure, and the same page still recommends reviewing the transcript, translation and timing before export. [9]
Review checklist before publishing a dubbed version
- Fix the source transcript first: names, product terms and numbers. Errors here carry into every language.
- Finish the edit before translating. Descript calls translation a finishing step and re-translating overwrites manual edits. [1][2]
- Have a fluent speaker review the translation, or at least the on-screen terms and calls to action. Use proofread where your plan has it. [1]
- Listen for speed changes where a direct translation was squeezed to fit, and for awkward phrasing where timing match added words. [1]
- Check numbers, units and dates in the new language; spoken and written formats differ.
- Watch lip-synced sections at full size for mouth artifacts, especially near microphones or hands. [3]
- Label the video as dubbed or AI-voiced in the description so viewers know, and keep the original-language version available.
- Generate captions for the dubbed audio and check they match what the voice actually says.
Questions before you decide
How does AI dubbing work?
It transcribes the original speech, translates the transcript, synthesizes the translation with an AI voice and, optionally, re-renders the speaker's mouth to match the new audio. Each stage inherits errors from the one before, so the transcript needs checking first.
How much does AI dubbing cost in Filmora?
Filmora lists AI translation of voice and text at 16 credits per minute. At its smallest pack, 1,500 credits for US$9.99, that is about US$0.11 per dubbed minute.
Does CapCut dub videos into other languages?
CapCut's video translator page covers translated subtitles in over 27 languages and says to use an AI dubbing workflow instead if you want to replace the spoken voice.
What footage works best for AI lip sync?
Descript recommends one person speaking on screen at a time, facing the camera, in clear light, with hands, microphones and hair away from the mouth. Filmora's lip sync works only on videos under 5 minutes.
Sources & verification
Product details and prices can change. Check the linked provider before buying.
- Descript Help: Translation overview Accessed 30 Oct 2026
- Descript Help: Dubbing Accessed 30 Oct 2026
- Descript Help: Lip sync Accessed 30 Oct 2026
- Descript Help: Media minutes and AI credits Accessed 30 Oct 2026
- Descript pricing Accessed 30 Oct 2026
- Filmora guide: AI Translation Accessed 30 Oct 2026
- Filmora AI credits rules Accessed 30 Oct 2026
- Filmora AI credits store Accessed 30 Oct 2026
- CapCut: Video translator Accessed 30 Oct 2026
- Descript Help: Top-ups Accessed 30 Oct 2026
- Filmora: buy video editor plans Accessed 30 Oct 2026
Sources link directly to providers. Product buttons may use separately labeled affiliate links. Read our disclosure.