<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: GeekLink</title>
    <description>The latest articles on DEV Community by GeekLink (@geeklink).</description>
    <link>https://dev.to/geeklink</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4217762%2Fe89cf08e-8500-4f86-80e7-91de06f7d89b.png</url>
      <title>DEV Community: GeekLink</title>
      <link>https://dev.to/geeklink</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/geeklink"/>
    <language>en</language>
    <item>
      <title>We Open-Sourced Our AI Subtitle Translation Engine: Every SRT Cue In, Every Cue Out</title>
      <dc:creator>GeekLink</dc:creator>
      <pubDate>Mon, 02 Nov 2026 06:34:41 +0000</pubDate>
      <link>https://dev.to/geeklink/we-open-sourced-our-ai-subtitle-translation-engine-every-srt-cue-in-every-cue-out-n0k</link>
      <guid>https://dev.to/geeklink/we-open-sourced-our-ai-subtitle-translation-engine-every-srt-cue-in-every-cue-out-n0k</guid>
      <description>&lt;p&gt;If you've ever pasted an SRT file into an LLM and asked for a translation, you've probably seen this: you send 400 subtitle cues and get back 397. Two short lines got merged into one sentence, one line disappeared, and from that point on every translation sits on the wrong timestamp. The output looks fine at a glance, and you only notice when the dialogue on screen no longer matches what people are saying.&lt;/p&gt;

&lt;p&gt;We hit this problem constantly while building GeekLink's subtitle translator, so we pulled the core engine out and open-sourced it under GPL-3.0: &lt;strong&gt;&lt;a href="https://github.com/GeekLinkDev/jev-subtitle-translator" rel="noopener noreferrer"&gt;jev-subtitle-translator&lt;/a&gt;&lt;/strong&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why LLMs mangle subtitles
&lt;/h2&gt;

&lt;p&gt;LLMs are trained to write fluent text, and subtitles aren't fluent text. A single sentence often spans three cues, a cue can be just "Yeah." or "—no!", and a model that's trying to be helpful will happily join fragments into natural sentences. Ask it to "keep the format" and it mostly will, until a batch of 40 cues comes back with 38 items and nothing tells you which two are gone.&lt;/p&gt;

&lt;p&gt;Prompt engineering alone doesn't fix this. You need structure the model can't drift away from, plus validation that doesn't trust the model's output.&lt;/p&gt;

&lt;h2&gt;
  
  
  How the engine handles it
&lt;/h2&gt;

&lt;p&gt;The pipeline looks like this:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Parse SRT and preserve every cue
        ↓
Translate ID-tagged batches
        ↓
Validate returned IDs and non-empty translations
        ↓
Retry only missing or empty IDs
        ↓
Run deterministic checks and Jev review
        ↓
Rebuild the SRT from the original cue list
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A few design decisions do most of the work:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Timestamps never reach the model.&lt;/strong&gt; The SRT is parsed locally. Subtitle numbers, order, and timestamps stay on your machine. Only the dialogue text and a stable ID per cue are sent to the model.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Structured output, not prose.&lt;/strong&gt; Each batch is translated with JSON Schema structured output through OpenRouter, so you can use GPT, Claude, Gemini, DeepSeek, Grok, or any other model that supports strict schemas. Local OpenAI-compatible servers (Ollama, LM Studio, vLLM, llama.cpp) work too.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Retry by ID, not by batch.&lt;/strong&gt; After every response, the engine accepts only expected IDs with unique, non-empty translations. If a batch of 40 returns 35 valid results, the next request contains only the 5 missing cues. You don't pay to re-translate the whole batch, and the good results aren't thrown away.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Rebuild from the original cue list.&lt;/strong&gt; The final SRT is assembled from the parsed source, not from whatever the model returned. One failed line can't shift or delete the cues after it. If a line still can't be translated after retries, it stays in place as an empty subtitle and gets flagged in the report, so the failure is visible instead of hidden.&lt;/p&gt;

&lt;h2&gt;
  
  
  Jev: automatic QC that flags, not rewrites
&lt;/h2&gt;

&lt;p&gt;A complete, well-aligned file can still contain bad translations, so the engine adds a QC pass. Local deterministic checks run first and catch empty lines and structural problems without calling any model. Then &lt;strong&gt;Jev&lt;/strong&gt; reviews every source–translation pair and answers one question: &lt;em&gt;does this line need human review?&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;It's tuned to flag things like dropped meaning, reversed negation, changed names, numbers, dates, or units, and truncated or repeated text. It looks at neighbouring cues in the batch, so a sentence that legitimately spans several cues isn't flagged. The rules favour precision: style differences pass, so the report stays short enough that you'll actually read it.&lt;/p&gt;

&lt;p&gt;Jev doesn't rewrite anything. A flag means "look at this line", not "this line is wrong". The review rules live in &lt;a href="https://github.com/GeekLinkDev/jev-subtitle-translator/blob/main/src/jev_subtitle_translator/qc.py" rel="noopener noreferrer"&gt;&lt;code&gt;qc.py&lt;/code&gt;&lt;/a&gt; if you want to see exactly what it checks or tune it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Quick start
&lt;/h2&gt;

&lt;p&gt;You'll need Python 3.10+ and an OpenRouter API key. On macOS or Linux:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git clone https://github.com/GeekLinkDev/jev-subtitle-translator.git
&lt;span class="nb"&gt;cd &lt;/span&gt;jev-subtitle-translator
bash run_web.sh
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Open &lt;code&gt;http://127.0.0.1:8000&lt;/code&gt;, upload an SRT, pick your languages, and click &lt;strong&gt;Translate + QC&lt;/strong&gt;. You can also QC an existing translation without re-translating it. You get the translated SRT plus a JSON report with line-level translation and review status.&lt;/p&gt;

&lt;p&gt;Prefer the terminal? There's a CLI for pipelines:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;export &lt;/span&gt;&lt;span class="nv"&gt;OPENROUTER_API_KEY&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"your-api-key"&lt;/span&gt;

.venv/bin/jev-subtitle-translator translate input.srt &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--source-language&lt;/span&gt; en &lt;span class="nt"&gt;--target-language&lt;/span&gt; de &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--model&lt;/span&gt; openai/gpt-4o-mini &lt;span class="nt"&gt;--output&lt;/span&gt; translated.srt
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That writes &lt;code&gt;translated.srt&lt;/code&gt; and &lt;code&gt;translated.srt.qc.json&lt;/code&gt;. Point &lt;code&gt;--base-url&lt;/code&gt; and &lt;code&gt;--qc-base-url&lt;/code&gt; at a local server to run everything locally.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where this fits
&lt;/h2&gt;

&lt;p&gt;This repo covers one job: translating and reviewing existing SRT files. If you're starting from a video, the &lt;a href="https://geeklink.dev/subtitle-translator/?utm_source=devto&amp;amp;utm_medium=blog&amp;amp;utm_campaign=jev" rel="noopener noreferrer"&gt;GeekLink desktop app for Mac and Windows&lt;/a&gt; is built on the same engine and adds OCR extraction of hardcoded (burned-in) subtitles, local Whisper transcription, and a bilingual editor for fixing lines side by side.&lt;/p&gt;

&lt;h2&gt;
  
  
  Contributions welcome
&lt;/h2&gt;

&lt;p&gt;The most useful contributions are real subtitle files that break things: odd SRT formatting, language pairs where Jev misfires, or models that keep dropping IDs. Open an issue or PR on &lt;strong&gt;&lt;a href="https://github.com/GeekLinkDev/jev-subtitle-translator" rel="noopener noreferrer"&gt;GitHub&lt;/a&gt;&lt;/strong&gt;. And if you've been translating subtitles with a one-off script and your fingers crossed, give the &lt;a href="https://github.com/GeekLinkDev/jev-subtitle-translator" rel="noopener noreferrer"&gt;repo&lt;/a&gt; a try and a star.&lt;/p&gt;

</description>
      <category>opensource</category>
      <category>python</category>
      <category>ai</category>
      <category>subtitles</category>
    </item>
    <item>
      <title>GeekLink Tutorial: Extract Hardcoded Subtitles and Create Bilingual SRT Files</title>
      <dc:creator>GeekLink</dc:creator>
      <pubDate>Fri, 30 Oct 2026 17:22:07 +0000</pubDate>
      <link>https://dev.to/geeklink/from-burned-in-text-to-bilingual-srt-how-subtitle-ocr-local-whisper-and-ai-translation-fit-21lm</link>
      <guid>https://dev.to/geeklink/from-burned-in-text-to-bilingual-srt-how-subtitle-ocr-local-whisper-and-ai-translation-fit-21lm</guid>
      <description>&lt;p&gt;You have a raw video with subtitles burned into the picture, and you need them as an editable subtitle file. Maybe you want to translate a lecture, study a drama with the original and the translation side by side, or re-publish a clip for a new audience. You open the file and there is no subtitle track to extract. The text is just pixels.&lt;/p&gt;

&lt;p&gt;Sometimes it is worse: there are no subtitles at all, only speech. And either way, you usually also want the result in a second language.&lt;/p&gt;

&lt;p&gt;This post walks through how I handle both cases in &lt;strong&gt;GeekLink&lt;/strong&gt;, a desktop subtitle app for Mac and Windows. It goes from a raw video to a bilingual SRT file, or a video with the subtitles burned in, in six steps.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnuckoy454nc3dw8bkypy.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnuckoy454nc3dw8bkypy.png" alt="GeekLink start screen with the three task cards" width="800" height="520"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The short version of how it works
&lt;/h2&gt;

&lt;p&gt;Burned-in subtitles are read off the frames with OCR inside a subtitle region you confirm, then merged into timed cues. When there are no subtitles, local Whisper transcribes the audio instead, and either result goes to AI translation as a bilingual subtitle you can review before exporting.&lt;/p&gt;

&lt;p&gt;That's all the theory you need. Here is the actual workflow.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: Import the video
&lt;/h2&gt;

&lt;p&gt;When GeekLink opens, it asks &lt;strong&gt;"What do you want to do today?"&lt;/strong&gt; Pick the card that matches your video:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Extract hardcoded subtitles (OCR)&lt;/strong&gt; if the subtitles are already on screen.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Transcribe speech&lt;/strong&gt; if the video has no subtitles, only audio.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Click &lt;strong&gt;Start&lt;/strong&gt;, then drag your video onto the drop zone (&lt;strong&gt;"Drag to import videos"&lt;/strong&gt;) or click &lt;strong&gt;Pick videos&lt;/strong&gt;. You can drop several files at once. They all land in the &lt;strong&gt;Library&lt;/strong&gt;, ready to be processed one after another.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0sly3w23ckxc2a2deec1.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0sly3w23ckxc2a2deec1.png" alt="Importing a video into the GeekLink Library" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Not sure which mode you need? The app asks &lt;strong&gt;"Where are the subtitles in your video?"&lt;/strong&gt; Choose &lt;strong&gt;Already on screen&lt;/strong&gt; for burned-in text or &lt;strong&gt;No subtitles, audio only&lt;/strong&gt; for speech.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Step 2: Select the subtitle region
&lt;/h2&gt;

&lt;p&gt;For burned-in subtitles, click &lt;strong&gt;Run Text Recognition&lt;/strong&gt;. GeekLink opens the &lt;strong&gt;Check subtitle lines&lt;/strong&gt; page and analyzes where the subtitles are. In most videos it finds them on its own and shows &lt;strong&gt;"Subtitle lines found - just start extraction"&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Scrub through the timeline once to make sure nothing important is cut off:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;If a suggested line is actually a logo, watermark or title, select it and click &lt;strong&gt;Delete&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;If the subtitles are somewhere unusual, draw a box around them yourself.&lt;/li&gt;
&lt;li&gt;If some lines were missed, open &lt;strong&gt;Can't find all subtitle lines?&lt;/strong&gt; and choose &lt;strong&gt;Scan more of the video&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F55mlqu381sj17ph9hd9n.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F55mlqu381sj17ph9hd9n.png" alt="Subtitle region picker in GeekLink" width="800" height="520"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Restricting OCR to the subtitle area is what keeps shop signs, on-screen titles and channel logos out of your subtitle file. It is worth the ten seconds.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: Recognize the text (OCR or local Whisper)
&lt;/h2&gt;

&lt;h3&gt;
  
  
  If the video has burned-in subtitles: OCR
&lt;/h3&gt;

&lt;p&gt;Click &lt;strong&gt;Start extraction&lt;/strong&gt;. GeekLink reads the subtitle region frame by frame and turns repeated readings of the same line into a single cue with a start and end time. Click &lt;strong&gt;View Live Results&lt;/strong&gt; to watch lines appear while it runs.&lt;/p&gt;

&lt;p&gt;The defaults work for most videos. Two settings are worth knowing about, under &lt;strong&gt;More settings&lt;/strong&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Detection interval (frames)&lt;/strong&gt; controls how often frames are checked. Lower numbers are more precise but slower.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ignore flash text&lt;/strong&gt; drops text that only appears for a split second, which is usually a misread during a fade.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnov3z853govd4p2v8t21.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnov3z853govd4p2v8t21.png" alt="Recognized lines in the View Live Results panel" width="800" height="520"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  If the video has no subtitles: local Whisper transcription
&lt;/h3&gt;

&lt;p&gt;Switch to the &lt;strong&gt;Speech&lt;/strong&gt; tab. Under &lt;strong&gt;Speech recognition settings&lt;/strong&gt;, set the &lt;strong&gt;Speech language&lt;/strong&gt; and choose a &lt;strong&gt;Whisper model&lt;/strong&gt;. &lt;strong&gt;Small&lt;/strong&gt; is the recommended balance. &lt;strong&gt;Medium&lt;/strong&gt; and &lt;strong&gt;Large V3 Turbo&lt;/strong&gt; are more accurate if your machine has the memory. Then click &lt;strong&gt;Run Speech&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Transcription runs locally on your computer, so your footage does not get uploaded anywhere. If you already know you want a translation, check &lt;strong&gt;Also translate to another language&lt;/strong&gt; and use &lt;strong&gt;Run Speech + Translate&lt;/strong&gt; to do both in one go.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fu5cwn2y59ojhw475y90a.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fu5cwn2y59ojhw475y90a.png" alt="Speech recognition settings with Whisper model selector" width="800" height="502"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Tip:&lt;/strong&gt; skim the source text before you translate. Fixing a misread name once in the original is much faster than fixing it in every translated line.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: Translate with AI
&lt;/h2&gt;

&lt;p&gt;Open the &lt;strong&gt;Translate&lt;/strong&gt; tab. Under &lt;strong&gt;Translation settings&lt;/strong&gt;:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Check &lt;strong&gt;Source language&lt;/strong&gt; and set your &lt;strong&gt;Target language&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;Choose a translation &lt;strong&gt;Engine&lt;/strong&gt;: ChatGPT, Claude, DeepSeek, Gemini or Grok. (The AI engines are part of GeekLink Pro.)&lt;/li&gt;
&lt;li&gt;Optionally, open &lt;strong&gt;More translation settings&lt;/strong&gt; and fill in &lt;strong&gt;Custom translation instructions&lt;/strong&gt;. This is the place for character names, product terms and tone, for example "keep honorifics" or "translate X as Y."&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Click &lt;strong&gt;Run Translate&lt;/strong&gt;. The translation is lined up with the original timing cue by cue, so you end up with a bilingual subtitle rather than a loose block of translated text. When it finishes, GeekLink runs an AI quality check and flags lines that deserve a second look.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuh79cgxknzwqk1p7v6gp.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fuh79cgxknzwqk1p7v6gp.png" alt="Translation settings in GeekLink" width="800" height="520"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: Review in the bilingual editor
&lt;/h2&gt;

&lt;p&gt;Click &lt;strong&gt;Open subtitle editor&lt;/strong&gt;. Set the view to &lt;strong&gt;Bilingual&lt;/strong&gt; to see the original and the translation together over the video, with the subtitle list beside it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcqc96t3j25axwrtg85kz.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fcqc96t3j25axwrtg85kz.png" alt="GeekLink bilingual subtitle editor" width="800" height="520"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;This is where you make the subtitles actually readable:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Edit text in place.&lt;/strong&gt; Click any line to fix a word in either language.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fix timing and line breaks.&lt;/strong&gt; Right-click a line to &lt;strong&gt;Merge into previous&lt;/strong&gt;, &lt;strong&gt;Merge into next&lt;/strong&gt; or &lt;strong&gt;Split&lt;/strong&gt; it at a chosen time point. To fill a gap, use &lt;strong&gt;+ Insert subtitle&lt;/strong&gt; on the waveform.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Jump to flagged lines.&lt;/strong&gt; Lines tagged &lt;strong&gt;Low conf?&lt;/strong&gt;, &lt;strong&gt;Not translated?&lt;/strong&gt; or &lt;strong&gt;Check translation?&lt;/strong&gt; are the ones to check first. &lt;strong&gt;Review flagged&lt;/strong&gt; walks you through them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Find &amp;amp; replace&lt;/strong&gt; across the whole file when a name is spelled wrong everywhere.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Changes are auto-saved. Play the video through once at normal speed. If you can follow the bilingual version comfortably, your viewers can too.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 6: Export SRT or burn the subtitles in
&lt;/h2&gt;

&lt;p&gt;Click &lt;strong&gt;Export&lt;/strong&gt; and pick:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Format:&lt;/strong&gt; &lt;strong&gt;SRT&lt;/strong&gt; for a subtitle file you can load in any player or editor. &lt;strong&gt;Toggleable subtitle&lt;/strong&gt; keeps a subtitle track inside the video that viewers can turn on and off. &lt;strong&gt;Burned-in subtitle&lt;/strong&gt; renders the text permanently into the picture.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Language:&lt;/strong&gt; &lt;strong&gt;Source&lt;/strong&gt;, &lt;strong&gt;Target&lt;/strong&gt; or &lt;strong&gt;Bilingual&lt;/strong&gt; (source + target).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Scope:&lt;/strong&gt; &lt;strong&gt;Current video&lt;/strong&gt;, or &lt;strong&gt;Marked videos&lt;/strong&gt; if you processed a batch.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fowh550il9aubq3s2zfwf.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fowh550il9aubq3s2zfwf.png" alt="GeekLink export dialog" width="800" height="520"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;For burned-in output, open the &lt;strong&gt;Style Editor&lt;/strong&gt; first to set the font, font size, outline, text color and bottom margin. Then click &lt;strong&gt;Render burned-in subtitles&lt;/strong&gt;, or &lt;strong&gt;Batch render all videos&lt;/strong&gt; for a whole folder.&lt;/p&gt;

&lt;h2&gt;
  
  
  Quick checklist
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Confirm the subtitle region before extracting. That removes most OCR noise.&lt;/li&gt;
&lt;li&gt;Set the speech language yourself instead of relying on auto-detect, especially if the video opens with music.&lt;/li&gt;
&lt;li&gt;Put names and recurring terms in &lt;strong&gt;Custom translation instructions&lt;/strong&gt; before translating.&lt;/li&gt;
&lt;li&gt;Review flagged lines first, then watch once at normal speed before exporting.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Get GeekLink
&lt;/h2&gt;

&lt;p&gt;GeekLink runs on &lt;strong&gt;Mac and Windows&lt;/strong&gt;. OCR extraction and Whisper transcription run locally on your own computer. There is a free plan, so you can run your own videos through the whole workflow before deciding whether you need more.&lt;/p&gt;

&lt;p&gt;Pricing:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Free&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;$12.99/month&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;$99/year&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Lifetime Pro: $129&lt;/strong&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Download it at &lt;a href="https://geeklink.dev/?utm_source=devto&amp;amp;utm_medium=blog&amp;amp;utm_campaign=tutorial_hardsub_bilingual" rel="noopener noreferrer"&gt;geeklink.dev&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If you work with subtitles in a language pair that gives you trouble, let me know in the comments which one. I'm curious which combinations people are dealing with most.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>tutorial</category>
      <category>subtitles</category>
      <category>productivity</category>
    </item>
  </channel>
</rss>
