Skip to main content

Translate Selected Transcripts

This workflow translates the transcripts of selected Premiere Pro project items into a target language — for multilingual captions, localized subtitles, or transcript-driven workflows in another language.

Because translation needs real language judgment on every run, this is a textbook case for a saved workflow with a Workflow AI step: the workflow handles all the transcript plumbing, and the AI does only the translating.

New to Automation Agent?

This recipe uses Automation Agent for Adobe Premiere. If you are here for the workflow solution first, that is fine. Use the product overview and Set Up Your AI for the setup path.

Fastest Path: Run The Library Script​

The recommended way to run this workflow is the finished script in the panel's Library: Translate Selected Transcripts With Workflow AI, under Examples > Transcript Handling.

What it does:

  1. You select the clip project items whose transcripts should be translated in the Project panel.
  2. You run the workflow and choose the target language and timing precision in a setup dialog.
  3. The workflow exports each transcript, splits it into stable segments, and sends only the text to your configured Workflow AI provider in batches.
  4. Every translated segment is validated — each requested segment must come back exactly once — and merged back into the Premiere transcript structure locally, preserving timing.
  5. After your confirmation, each processed clip's transcript is replaced with the translated version. Items without transcripts are skipped and reported, and temporary files are cleaned up.

Timing precision options: sentence timing translates smaller sentence-like units inside Premiere transcript segments for better sync; segment timing keeps the original Premiere segment granularity and is faster.

The model never receives or generates timing data — it translates text, nothing else. That constraint is built into the script, not into a prompt the AI might drift away from.

What You Need First​

  1. Selected clip project items with usable transcripts.
  2. A configured and tested Workflow AI provider — Codex CLI, Claude Code CLI, or a local LM Studio model. Guided Setup covers this.
  3. The Python Runtime Pack. The panel's preflight tells you if it is missing.

The workflow is tuned for CLI providers such as Codex or Claude Code by default: it sends larger transcript batches so the model can use broader context. It also works with local LM Studio models such as Qwen — for smaller local models, lower the batch size in the setup dialog.

Limit write access to the clips you actually want to translate:

  1. Create a dedicated Project bin, such as Translate, and move or duplicate the clip project items into it.
  2. Open the Execution Permissions tab.
  3. Enable Restrict write access in the Project section and add only that bin as the exception.

Keep in mind that the default run replaces each processed clip's transcript. If you want to keep the original-language transcript, duplicate the clips first and translate the duplicates.

Working Interactively Instead​

If you prefer a live agent session — for example to translate only parts of a transcript, use unusual terminology context, or review each translated passage in conversation — paste this prompt into your connected agent:

Using Automation Agent in Premiere Pro, translate the transcripts of the currently selected clip project items into [TARGET LANGUAGE].

- If no clip project items are selected, stop and ask me to select them.
- Skip items without usable transcript data and report them.
- Export each transcript JSON and verify it contains real words and timings before translating.
- Translate only the text. Preserve transcript timing, segment order, speaker metadata, and any IDs required by Premiere exactly.
- Keep names, brands, and technical terms unchanged unless I provide translations for them. Ask me for optional terminology context before you start.
- Show me a short sample of the translated text per clip and wait for my approval before importing anything.
- After approval, import the translated transcript JSON back into the matching clip project items and verify the import by re-exporting.
- Report per clip: translated, skipped, or failed.

Expect the interactive run to be noticeably slower and more token-hungry than the library script, because the agent has to work out the transcript export and import mechanics itself.

Limitations And Review Points​

  • Translation quality depends on the configured model. Review the result before publishing captions.
  • Idioms, wordplay, and culture-specific references need human judgment — treat the output as a strong draft.
  • The default library run replaces the source transcript; duplicate clips first if you need both languages in the project.
  • Names, brands, and technical terms benefit from explicit terminology context, exactly as in the proofreading recipe.