TL;DR: There are three ways to subtitle anime, and the right one depends on your source. If you have raw Japanese with no subtitles, transcribe the audio with Whisper and translate it; if the episode already has subtitles burned into the picture, pull them back out with OCR; if you already have an .srt or .ass file, you only need to edit and re-time it. For finished fansub typesetting — karaoke, styled signs, OP/ED — you still export to .ass and polish in Aegisub.
Which kind of anime subtitles do you need?
Three sources, three workflows. Before you touch anything, check what you are actually starting from — it decides the whole approach:
- Raws — a Japanese episode with no subtitles at all. You transcribe the audio, then translate it into your language.
- Hardsubbed — subtitles baked into the picture (common in re-encodes and stream rips), with no separate track. You OCR them back into editable text.
- Softsubbed — a separate
.srtor.assfile alongside the video. You already have the text; you only edit and re-time.
The fastest path depends entirely on your source: transcribe raws, OCR hardsubs, or edit an existing subtitle file — never redo work the episode already contains. Pick the wrong workflow and you burn hours recreating text you could have pulled out in minutes.
Subtitle raw Japanese anime (transcribe + translate)
This is the fansub case: a raw with no subtitles. It is two steps — transcribe the Japanese audio, then translate it. Both run locally on your Mac with Whisper speech recognition and AI translation; nothing is uploaded.
Two things about anime specifically will wreck a naive run, and both have a fix:
Music-heavy openings and endings make Whisper invent subtitle lines. Over an OP, ED, or insert song, speech recognition tends to hallucinate lyrics or repeat phantom lines. A voice-activity-detection pass that mutes non-speech segments keeps those ghost lines out, so your timeline is not littered with garbage over the theme song. Still spot-check the OP/ED afterward.
Anime lives and dies on consistent names and honorifics. Character names and suffixes like -san, -chan, and -senpai have to read the same way in episode 1 and episode 12. A terminology list that pins each name and honorific keeps the AI from spelling a character three different ways across a cour.
Treat the output as a strong first draft, not a finished sub — anime leans on wordplay, jokes, and cultural references that still want a human review pass. GeekLink does the transcribe, translate, and review-editor steps in one local workflow.
Extract subtitles burned into anime (OCR)
Many re-encodes and stream rips ship with the subtitles burned straight into the video, with no separate track to copy. To reuse, re-time, or re-translate those, OCR reads the characters back off the frames.
Burned-in anime subtitles have no text track to copy; OCR reads the characters straight off the video picture and turns them back into an editable .srt. It works on Japanese, Chinese, and English burned-in text, so a hardsubbed release becomes an editable file again. See the full walkthrough in how to extract hardcoded subtitles with OCR.
Subtitle a whole season at once
Anime comes in cours — 12 to 13 episodes, sometimes 24. The real time sink is not any single episode; it is doing the same setup twelve times over.
The work in anime is volume, not any one episode — batching a full cour through transcription and translation at once turns a week of repetitive setup into a single queue. Import the whole season, run transcribe → translate → export across every episode, then spend your time reviewing instead of re-configuring. Everything stays on your machine.
Soft subs vs hard subs (and why fansubs use ASS)
Two decisions come up once the text is ready: soft versus hard, and which format.
- Soft subs (toggleable) — a separate
.srtor.assfile the viewer can turn on or off, and that you can fix later without re-encoding. Best while you work and for sharing editable files. - Hard subs (burned-in) — baked into the video, always visible. Needed for the final clip or for platforms that strip subtitle tracks (social shorts, some players).
Use soft subs while you work and to share editable files; burn in hard subs only for the final export or a platform that drops subtitle tracks. On format: .srt is universal but plain, while .ass (Advanced SubStation Alpha) carries the fonts, positioning, colors, and karaoke that fansubs rely on for signs and songs. Keep the file in .ass when you need styling.
GeekLink and Aegisub: who does what
Worth being straight about the division of labor. Aegisub is the fansub standard for typesetting — karaoke timing, styled signs, complex ASS effects — and it is free. GeekLink does not replace that, and you should not expect it to.
Let GeekLink do the volume work — transcription, OCR, translation, re-timing, batching a whole season — then export to .ass and finish the typesetting and karaoke in Aegisub. They are complementary, not competitors: one clears the heavy, repetitive front of the pipeline locally on a Mac (Apple Silicon), the other does the artisanal styling at the end. For plain dialogue subs you may never open Aegisub at all; for signs-and-songs typesetting, that hand-off is exactly the point.
FAQ
Can AI translate Japanese anime accurately?
Whisper transcription plus AI translation gives a solid first draft for ordinary dialogue, but anime leans on wordplay, honorifics, and cultural references that a machine will flatten. Treat it as a fast draft you review and fix, not a finished sub — the AI saves you the typing, not the judgment.
How do I stop Whisper from adding subtitles over the opening song?
That is a music hallucination: over an OP, ED, or insert song, speech recognition invents lyrics or repeats phantom lines. A voice-activity-detection pass mutes non-speech segments so no lines are generated over the music. After the run, spot-check the OP and ED anyway.
What subtitle format do anime fansubs use?
Mostly .ass (Advanced SubStation Alpha), because it supports positioning, fonts, colors, and karaoke for signs and songs. Use .srt for simple dialogue that only needs to appear at the bottom of the screen, and .ass when you need styling or typesetting.
Can I extract the subtitles that are burned into an anime episode?
Yes, with OCR. It reads the on-screen text out of the video frames and rebuilds an editable .srt, for Japanese, Chinese, or English burned-in subtitles — so a hardsubbed release becomes a file you can re-time or re-translate.
Do I still need Aegisub if I use GeekLink?
For plain dialogue subtitles, usually not. For fansub-style typesetting — karaoke, styled signs, complex positioning — yes: export from GeekLink to .ass and finish in Aegisub. GeekLink handles the transcription, OCR, translation, re-timing, and batching; Aegisub handles the artistic styling.