03 Translation Pipeline & Progress
The progress page splits one video into cards, one card per step, executed top to bottom. How many steps you see depends on the project type and on which switches you turned on when you created it — the same type is 4 / 6 / 7 steps long by default — so don't measure your run against somebody else's step count.
1. What Each Step Does
| Step | What it is doing | When it shows up |
|---|---|---|
| Audio Extraction | Takes the sound out of the video | Always |
| Subtitle Removal | Erases the hardcoded subtitles already in the picture so they don't pile up under the new ones | No project you create today gets this step — the create form no longer asks about it. To erase subtitles, use the standalone Subtitle Removal tool in the left sidebar, see 07 Download · Subtitle Removal · Watermark Removal; only projects made years ago still run this step with the settings they were created with |
| Voice Separation | Pulls the voice out of the background music and ambient noise, so only clean vocals go into recognition and cloning | Always (every type except Video Subtitle) |
| Speech Recognition | Transcribes the speech into text and cuts a time range for every subtitle line | Always |
| Speaker Identification / Speaker Diarization | Works out how many people speak in the video and who said each line | When Multi-speaker mode is on |
| Emotion Recognition | Analyzes the emotion in the original voice so the dub can deliver it | When Emotion Recognition is on |
| Correction | A large model fixes the wrong characters that speech recognition produced | Video Subtitle, with AI Subtitle Correction still on |
| Translation | Translates the source text into the target language (this step runs in the cloud and needs internet access) | Every type except Video Subtitle |
| Audio Generation | Reads the translation out subtitle line by subtitle line in the voice you chose | Every type except Video Subtitle |
| Subtitle Position | Decides where in the picture the subtitles go | Expert / Quick types |
| Subtitle Burn | Draws the subtitles into the picture | Video Subtitle |
| Video Merge | Combines the new dub + subtitles + original picture into the finished video | Always |
The step count each type actually shows with the default settings:
- Expert Video Translation / Quick Video Translation: 7 steps.
- Video Dub: 6 steps (multi-speaker is forced on, Emotion Recognition is off by default).
- Video Subtitle: 4 steps (nothing related to dubbing).
It always runs in two stages: all seven cards are on the page, but the first stage only goes as far as Translation. Once that step completes, the progress page shows "Paused for review - edit subtitles in editor" and the editor opens by itself. The remaining three steps (Audio Generation → Subtitle Position → Video Merge) start only when you click "Generate Video" in the editor. That is by design, not a hang — dubbing costs money, so checking the translation first is cheaper than finishing the run and redoing it. This type has no "run it all in one go" switch: if you don't want the pause, use Quick Video Translation instead — the Batch Queue fills with that type as well, and it does not stop at the translation.
2. Where The Time Goes
Three things decide the total run time: how long the video is, how powerful your machine is, and the quality tiers you picked. In practice Voice Separation and Audio Generation take the most time, with Speech Recognition behind them; the translation step is usually the fastest one, because it goes through a cloud interface.
Treat run time as "the longer the video, the slower it runs, and it climbs in a straight line" — a 40-minute video isn't just forty times the work of a one-minute one, and failing halfway through costs a lot more too. So before you run a long piece of footage for the first time, run two or three minutes of it, check the voice and the subtitle style, then run the whole thing.
Voice Separation is the most memory-hungry step in the whole chain. When you run footage tens of minutes long, close your browser and your editing software first; when memory runs short this step fails outright instead of just getting slower. When you retry after that failure, the separation step carries on from the point it had already reached rather than redoing the whole step.
3. Failure, Retry, Cancel
- A step fails: that card stays in the failed state and shows you a message. Most of the time you don't have to start over — click "Retry" on that one step, and everything the earlier steps produced is kept.
- Change settings, then retry: if you changed the subtitle style or the voice in the editor and run generation again, the app re-runs everything from Audio Generation onward (Video Subtitle only re-runs Subtitle Burn). Change a single character of the subtitle text and the entire dub is regenerated — that has a cost consequence for you, see 04 Subtitle & Dubbing Editor.
- Cancelling halfway: clicking Cancel Pipeline really means you don't want this run — the intermediate output is thrown away and the next attempt starts from the beginning. That is a different thing from "retrying after a failure carries on where it left off".
- The message means nothing to you: the body of the error dialog is a single sentence in your own language; expand "View Technical Details" to see the actual cause. Paste that part back to us — it gets us to the cause far faster than guessing, see 11 When Something Goes Wrong.
4. Where You Land When It Finishes
Finishing doesn't always bring you into the editor; each of the four types has its own landing page:
- Expert Video Translation: when the first stage stops at "Translation" it opens the editor by itself; when you click "Generate Video" and the remaining three steps finish, you land on the read-only video preview page, which is where you watch the dubbed result — to keep editing, click "Edit" in that page's toolbar.
- Quick Video Translation: runs all the way through and lands on the read-only video preview page. To change subtitles or the voice, click "Edit" on the preview page to enter the editor.
- Video Dub / Video Subtitle: go straight into the editor when they finish.
Opening a project that is already finished from the project list lands you in the same place (Expert and Quick land on the preview page first). The finished video is target.mp4 in the project folder, and "Open Project Folder" in the top bar of the editor jumps straight to it.