Already have an SRT? Import it — skip recognition | AI Video Factory

If CapCut or another editor already gave you cues that sit on the mouth, another speech or OCR run burns free quota and can mangle names. The studio imports SRT / VTT / YAML. When you already have captions, import them. Do not listen to the track or read the picture again.

Confirm you already have a clock

Import takes text and timestamps, not style. If the clock is messy or full of typos, fix it in the original app, or recognize in ten-minute captions. Import when the axis is usable and you need a restyle, a translation, or a dub.

No caption file, only voice or burned-in type: speech or OCR. Do not drop a burned-in MP4 as a caption file.

After import, do only the next step

1. Open captions in the studio and drop a local MP4 / MOV. The file stays in the browser. Preview and burn-in both need this picture.
2. Choose Import captions and pick SRT / VTT / YAML. Check the cue count against the source file.
3. Style only: open style, set line length, position and the bar. Do not rerun recognition. Vertical: vertical layout.
4. Translation only: translate on the same clock; only a translation run is counted. Path: bilingual captions.
5. Dubbing only: dub on the same clock; only a dubbing run is counted. See timeline dubbing.
6. Preview before export. Short video burns in; channels that accept soft captions take SRT — SRT or burn-in.

The free plan includes one caption recognition per month; import does not use that run. Translation and dubbing still count on their own. Details: Features and Pricing.

Check only three things

• Format: SRT, VTT or timeline YAML. ASS / a CapCut project is not a caption import.
• Count: scan names and empty cues after import. Zero cues: open the source file first, do not jump to recognition.
• Picture: a different video than the clock puts type on the wrong shots. Match the files before you restyle.

What usually breaks

Transcribing a clock that already works: quota gone, names rewritten. Import then recognize again is wasted work. Sending a burned-in file to OCR as “import” stacks a second cue list. Export a cover separately — covers from the clip.