Source: raw/How_to_Auto_Translate_Videos_Into_ANY_Language_With_AI_Full_Tutorial.md (YouTube xa8Hsacu7fM, uploaded 2026-10-07) and raw/How_to_Translate_Videos_With_AI_for_FREE_177+_Languages_HeyGen_Tutorial.md (YouTube jw-ivrqe5qA, uploaded 2026-09-29). HeyGen plan and credit figures also from raw/Top_5_AI_Tools_for_Video_Translation_in_2026.md (YouTube RwkFWOcDqMM, uploaded 2026-09-28). Creator: HeyGen official channel (@heygen_official); presenter not named in the transcripts URL: https://www.youtube.com/watch?v=xa8Hsacu7fM · https://www.youtube.com/watch?v=jw-ivrqe5qA Duration: 9:21 (auto-translate tutorial) · 8:45 (free tutorial) Platform: YouTube Skill level: operator

HeyGen’s video translator takes one finished recording and produces dubbed versions in other languages. It transcribes and translates the speech, re-voices it with a voice based on the original speaker, and can optionally re-render the mouth to match (lip sync) and add translated captions. Two HeyGen-made tutorials walk the same flow: import, source and target language, translation engine, brand glossary and advanced settings, then either direct translate or a proofread pass. Both are vendor tutorials, so treat “free” and quality claims as HeyGen’s own.

Key Takeaways

  • What it changes: spoken dialogue (transcribe → translate → re-voice in a clone of the original speaker), optionally lip movements, and optionally translated captions.
  • What it does not change: text baked into the picture. Graphics, logos and hard-coded source captions must be replaced separately in editing.
  • Language count (vendor claim): “more than 177 languages and dialects”, and the count includes regional variants (e.g. Spanish for Spain, Mexico, the United States, Latin America).
  • Base vs localized variant: the base/general option preserves more of the original speaker’s accent and identity; a localized option prioritizes an accent native to the region. The presenter recommends base for interviews, documentaries, reaction and personality content; localized for training, onboarding, product demos and single-market marketing.
  • Up to 10 target languages per job. Both tutorials say to test one language first, then batch the rest, so a mistake is not repeated across 10 versions and their credits.
  • Three translation engines:
    • Audio only — replaces the dialogue, mouth unchanged. Fastest and cheapest. For narration, screen recordings, slides, podcasts, or when the face is not visible.
    • Speed — adds lip sync; for straightforward front-facing footage: face visible, speaker within roughly 45° of the camera, one person speaking at a time.
    • Precision — highest-quality lip sync; for side angles, camera changes, speaker switches, and hands or microphones partly covering the face.
  • Credit cost on paid credit plans (as stated 2026-09-29): audio only 4 credits/minute, speed 6, precision 10. The comparison video’s example agrees: one 10-minute video into 3 languages with precision = 300 credits.
  • Brand glossary (set before an important translation): add terms manually or upload a CSV. Force translate fixes how a term appears in the target language; Don’t translate protects a term; the free tutorial adds pronunciation control for names. Use per-language glossaries, since terminology that works in Spanish may not be what you want in Japanese or German.
  • Advanced settings:
    • Dynamic duration lets translated segments stretch or compress by up to 20%, so speech sounds natural; it can change total runtime slightly. Both tutorials recommend keeping it on.
    • Output to original video specs preserves source resolution (720p, 1080p, 4K) — higher-tier plans only.
    • Enhanced voice can improve clarity and volume but may slightly change tone; compare before using it on a batch.
  • Translate vs Review and edit (Proofread): Translate sends straight to final processing. Proofread creates an editable project first and is only on supported paid plans; free accounts may see only Translate. The comparison video lists Pro, Business, Team and Enterprise as the plans with proofread ^[ambiguous] (the caption run “Pro, business team, and enterprise” could also mean a combined “Business/Team” tier).
  • Inside Proofread: dialogue is split into segments, each with original timing, translated script, assigned speaker and generated voice. You can reassign speakers, mark narration as off-screen (so the voice is not applied to a visible person), edit words, rewrite sentences, add a missing line between timestamps, adjust timing, fix pronunciations, and swap the voice (a cloned voice, a HeyGen library voice, or another voice in the project). After any text, timing or voice change, click refresh voice for that segment. A Preview tab plays translated audio against the original footage before the lip-sync render; a Result tab holds the finished translation.
  • Generate step: choose whether to include captions and whether the output is lip-synced or audio only. Each language is processed as its own separate video, and the language versions can be organized into one collection. HeyGen notifies you on completion; the page does not need to stay open.

Free tier (vendor claims, as stated)

  • Free account, no credit card required; the free plan gives “limited access to video translation” — “a free trial of the feature, not an unlimited free translation.”
  • The tutorial does not state the free allowance and says to check the allowance shown in your workspace because limits change.
  • The comparison video (2026-09-28) states the free plan allows up to 3 translated videos in total, maximum 3 minutes each.
  • Proofread is not on the free plan.

Try It

  1. HeyGen dashboard → left sidebar Apps → Translate video.
  2. Import: upload MP4, MOV or WebM (up to 5 GB; max duration depends on plan), pick a video from your HeyGen workspace, paste a full YouTube URL, or a Google Drive link set to “anyone with the link”.
  3. Set the source language manually if you know it (auto-detect is available). Source should contain one primary spoken language; mixed-language recordings can produce a wrong transcript or a failed translation.
  4. Pick one target language and region variant first.
  5. Pick the engine: audio only if the face does not need to change, speed for a clean front-facing talking head, precision for complex footage or when lip-sync accuracy matters most.
  6. Open Advanced: attach the brand glossary, keep dynamic duration on, decide on enhanced voice and original specs.
  7. Before clicking, re-check source language, target language and region, engine, and the estimated usage shown on screen.
  8. On paid plans use Review and edit; otherwise Translate.
  9. Watch the full result before downloading: translation, voice, names and product terms, timing, captions, and any moment the face turns away. If something is wrong, first check whether the cause is the source audio, the chosen dialect, or a difficult section of footage before regenerating the whole project.
  10. Once one language is approved, return to the original project and run the remaining languages with the same settings.

Source-video checklist (from both tutorials): clean speech without background noise; face visible and reasonably close to the camera; no overlapping speakers; music and complex sound effects can hurt the result.

Implementation

Tool/Service: HeyGen video translator (web app, Apps → Translate video). The same capability is exposed to coding agents as the heygen-translate skill — see skills. Setup: free HeyGen account to test; a paid plan for Proofread and original-spec output. Cost: credits per minute of source video per target language: 4 (audio only), 6 (speed), 10 (precision). Creator plan from $29/month with 600 credits (comparison video, 2026-09-28) — the source notes one 10-minute video into 3 languages on precision uses half that allowance. Integration notes:

  • Budget per language, not per video: each target language is a separate render.
  • Glossaries are reusable across future videos, which keeps product names consistent.
  • Accuracy caveat from the comparison video: have a fluent speaker verify anything customer-facing or legally sensitive.

Open Questions

  • Free-tier allowance is stated only in the comparison video (3 videos, 3 minutes each); the tutorials decline to give a number. Not verified against HeyGen’s pricing page.
  • Which exact plans include Proofread and “output to original video specs” — the sources say “supported paid plans” and “higher-tier plans” without a definitive list.
  • Language count: these 2026-09/10 videos say 177+ languages and dialects (regional variants included). Existing wiki articles (Avatar V, skills bundle) cite 175+ languages. Both are lower bounds, so this is not logged as a contradiction, but the counting basis differs.
  • Maximum source duration per plan is not stated.
  • How lip sync handles more than one visible face in the precision engine beyond “speaker switches” is not demonstrated.