SRT Caption Toolbox

The SRT Caption Toolbox helps you move from Premiere transcripts to subtitle files you can review, improve, translate, and import back into Premiere.

The tools work with SRT files because Premiere's current UXP API cannot yet read or change the text on Caption Tracks directly. Each step creates a visible file or report, so you stay in control of what is imported into the project.

New to Automation Agent?

These recipes use Automation Agent for Adobe Premiere. The finished workflows are available in the panel's Library under Examples > Transcript Handling.

The Eight Library Workflows

Export Selected Clip Transcripts To SRT

Select one or more footage or file items in the Project panel, then run Export Selected Clip Transcripts To SRT.

The workflow:

• exports the complete source transcript of every selected clip;
• reads and normalizes each clip's language from its Premiere transcript;
• creates one SRT per usable transcript;
• applies your chosen line length, line count, and reading-speed limits;
• optionally saves precise word timing for later Caption Toolbox steps;
• skips existing output files unless you explicitly enable overwrite;
• continues with other selected clips when one item has no usable transcript.

It does not export only the portion used in a sequence. Use the active-sequence workflow for that.

When Premiere provides one clear transcript language, the output filename includes it, for example Interview.de-DE.srt. If that language information is missing or inconsistent, the workflow does not guess: it creates Interview.srt and identifies the missing language information in the report.

Keep Save precise word timing for later tools enabled if you may adjust the captions afterward. The workflow places small companion files in an automation-agent-timing/ folder next to the SRT files. You do not need to open or import them. Reflow SRT Captions finds them automatically and can then rebuild captions at the actual spoken words instead of estimating the timing.

Translation does not reuse source-language word timing directly because translated words have different lengths and positions. However, if you reflow the source SRT first, translations created from that improved SRT inherit its better-prepared caption time ranges.

Generate SRT From Active Sequence Source Transcripts

Run Generate SRT From Active Sequence Source Transcripts when you want captions for the current edit rather than for complete source clips.

The tool combines the source transcripts of the clips used in the active sequence and places their words at the corresponding Timeline times. It can use all unmuted audio tracks or only one chosen track. Track numbers match the Timeline labels: enter 1 for A1, 2 for A2, and so on.

The language is read from the contributing Premiere transcripts. A single shared language becomes the filename suffix. Mixed languages or clips without reliable language information are listed in the report, and the output uses the sequence name without a potentially misleading language suffix.

Before running it, check that the intended dialogue is audible and mute any transcribed material you do not want in the subtitles. Music and sound-effects clips without usable transcripts are simply skipped, so they can remain on unmuted tracks. If you want tighter control, choose a single dialogue track instead of combining all tracks.

The tool creates the SRT only when it can match the included dialogue reliably:

• normal forward playback at 1× speed is supported;
• muted tracks and disabled clips are left out;
• clips without usable transcripts are skipped and listed by clip name, track, and Timeline range;
• as long as at least one clip contributes usable dialogue, skipped music or effects do not prevent the SRT from being created;
• duplicate representations of the same spoken material are included only once;
• overlapping, different spoken passages are listed for review instead of being combined automatically;
• if an available transcript has unreliable timing, the report identifies the clip and no potentially incorrect SRT is created;
• if no clip contributes usable dialogue, no empty SRT is created.

The report shows which clips were used, which were skipped, and where overlapping dialogue needs your attention. Premiere does not expose Solo state or all volume, routing, effect, and silence information through the current API, so check the audible track state before combining all tracks.

Analyze SRT Readability And Timing

Use Analyze SRT Readability And Timing to find common subtitle problems without changing the SRT. Select the SRT and report folder directly in the setup dialog. The analysis does not need a language code. It checks line count, line length, caption duration, reading speed, gaps, overlaps, frame alignment when requested, and file structure.

Under Caption settings, choose:

• Standard settings (recommended) for a quick general editing check;
• a profile found in Documents/Automation Agent/Caption Profiles for recurring client or platform requirements;
• Choose another profile file… for a profile stored with a project, client, or team.

If a saved profile cannot be read, the dialog identifies the affected file and explains why it is not available in the list. The same notice appears in Create Or Edit Caption Profile, so a broken or newer profile does not simply disappear without explanation.

Leave Review or adjust settings before running disabled when you simply want to use the selected settings. Enable it to see a second dialog containing only the values relevant to this analysis. That dialog is prefilled from the selected profile, with Standard values filling any fields the profile does not define. Changes apply only to this run and never modify the saved profile.

The review dialog can adjust text length, line count, reading speed, minimum and maximum caption duration, delivery frame rate, the minimum-gap check, and optional frame-boundary checking. The general 80-ms gap is an editing preference and produces warnings. To use a specific requirement, first choose milliseconds or frames; the following dialog asks for the exact value. A millisecond- or frame-based minimum selected in a profile or for the current run is treated as a requirement. Frame-based checks always require an explicit delivery frame rate because an SRT does not contain one.

The frame-boundary check reports caption starts or ends that fall between frames at the selected delivery frame rate. It does not move or otherwise change any caption timing. This is useful when a delivery specification requires every caption boundary to land on the frame grid.

After the analysis, the result dialog shows an immediate overview: number of captions and problems, typical and maximum reading speed, the longest line, duration and gap ranges, counts by problem type, and a short list of the most important locations. When a frame rate is selected, it also shows frame measurements and frame-boundary results. Complete Markdown and JSON reports contain every finding and identify which values came from Standard settings, the selected profile, or changes made for this run. If you change or disable a profile rule for one run, the pass statement explicitly excludes that rule instead of claiming that the unchanged profile passed in full.

If the SRT belongs to the active sequence, you can optionally add review markers. The workflow creates one range marker per problem location and combines all findings for the same caption. Each marker contains the caption text, measured values, and configured limits, so you can use Premiere's normal marker navigation to review the affected places.

Only enable marker creation when the SRT times match the active sequence. Choose whether SRT time 00:00:00,000 corresponds to the beginning of the sequence or to the current playhead position. Existing sequence markers are left untouched.

Reflow SRT Captions

Use Reflow SRT Captions to create a separate SRT with completely rebuilt caption boundaries and line breaks. The workflow considers the surrounding passage and later captions instead of filling each old caption independently. This helps it avoid a locally convenient break that would leave a single word or another awkward fragment afterward.

Sentence endings are preferred over pauses and commas. The two punctuation percentages are minimum fill levels:

• At a sentence-ending preference of 60%, sentence endings become good candidates once the caption is at least 60% full.
• The workflow still checks whether another complete sentence fits and normally uses a later suitable sentence ending.
• It may choose an earlier ending when that produces a better overall sequence of captions.
• Commas are a weaker preference and therefore use a higher default such as 80%.

Select one or more .srt files directly in the setup dialog. The files can come from the Caption Toolbox or any other subtitle workflow; they do not need a special filename or folder structure. The same reflow settings are applied to every selected file.

Choose Standard settings (recommended), a saved caption profile, or another profile file stored with the project or team. Leave Review or adjust settings before running disabled to use those settings immediately. Enable it to review or change Reflow's text length, line count, and minimum/maximum caption duration for this run. Reflow uses those values while rebuilding captions. It also checks the finished SRT against a saved reading-speed or minimum-gap requirement and reports whether it passes, but it does not shorten wording or force gaps to satisfy those requirements. Frame-alignment requirements are not checked by Reflow and remain clearly listed as not checked in the report.

Matching precise word timing from automation-agent-timing/ is used automatically. Without it, the tool estimates word positions inside the existing caption times and identifies the result as estimated in the report.

The source files are never changed. Leave Output folder empty to create a reflowed folder next to each source file, or choose one common output folder for the whole batch. Output files keep their original names, and every result gets JSON and Markdown details in a reports subfolder. Existing output files are skipped by default; enable Overwrite existing output files only when you intentionally want to replace them.

Condense SRT For Readability With Workflow AI

Use Condense SRT For Readability With Workflow AI when captions contain more text than viewers can comfortably read in the available time. Select one or more SRT files, choose an output folder, and set the maximum reading speed you want.

Choose Standard settings (recommended), a saved caption profile, or another profile file stored with the project or team. Leave Review or adjust settings before running disabled to use it immediately. Enable the option to review or change Condense's text length, line count, reading speed, and minimum/maximum caption duration for this run. Gap, frame-rate, and frame-alignment requirements are not applied by Condense and are listed in its report instead of being silently treated as completed.

The tool avoids rewriting text when timing alone can solve the problem:

1. Captions that already meet the reading-speed limit stay unchanged.
2. For a sentence with one unusually dense caption, the tool first redistributes the complete original wording across that sentence's available time.
3. Only when the complete wording still cannot fit does Workflow AI receive that sentence and its immediate text context for careful shortening.
4. The tool—not Workflow AI—places an accepted result back inside the original sentence time range.

This distinction matters: a successful run may improve many captions without sending their text to a model at all. The source SRT is never changed. Each output is written as *.condensed.srt with Markdown and JSON reports showing captions left unchanged, redistributed without rewriting, rewritten with Workflow AI, or retained because a safe result was not available.

A PARTIAL result is still a usable SRT. The report shows every retained original group and every rewritten passage as an Original/New comparison. It also highlights possible changes to numbers, prices, URLs, or product codes. Automated checks cannot prove that every nuance survived shortening, so review all rewritten groups before delivery.

The optional fields help Workflow AI make better editorial choices:

• Project context explains the production and audience.
• Known names and terminology helps interpret specialized wording.
• Text that must stay exactly unchanged protects source spelling for names, brands, URLs, product codes, legal text, or approved claims.
• Additional condensation instructions can describe tone and priorities within the goal of reducing reading speed.

The first version preserves each sentence's outer time range but may create consecutive captions with no empty time between them. Run Analyze SRT Readability And Timing afterward to verify final reading speed and gaps. Use Reflow SRT Captions when the main problem is line shape, line length, or caption division rather than excessive reading speed.

For Sentence groups per Workflow AI request, start with 25 for Codex CLI or Claude Code. For a local LM Studio model, start with 6 and stay around 4–10 until that model has proven reliable. Smaller batches are slower but easier for smaller models to return consistently.

Align Caption Track Changes to Sequence Cuts

Use Align Caption Track Changes to Sequence Cuts when the captions are already on a Caption Track in the active Premiere sequence. The workflow moves nearby caption starts and ends directly onto picture edits from one chosen video track. It changes timing only: caption text, styling, order, and count stay untouched.

Choose the Caption Track, the reference video track, and how many frames away from a cut a caption change may be. Track numbers follow Premiere's labels: 1 means V1, 2 means V2, and so on. Only boundaries between adjacent enabled clips on the selected video track are used. If that track contains one flat rendered clip, first run Detect Scene Edits In Selected Clips, review the resulting edits, and then run this workflow.

Minimum caption duration and minimum gap are also entered in frames. The tool uses the active sequence's frame rate, so you never need to convert those safety limits to milliseconds yourself.

This workflow is the deliberate exception to the Library's usual setup timing: its Caption Track choices are populated from the active sequence before the dialog opens. The dialog names that bound sequence. Cancel and restart the tool if you need to target a different sequence.

Before anything changes, the workflow shows a preview and asks for confirmation. It skips changes that would make a caption too short, create an overlap, or violate the selected minimum gap. Accepted changes are applied together, so one Undo restores them all. The workflow then reads the complete Caption Track again and verifies that Premiere applied exactly the planned timing. If the track is locked or Premiere refuses a change, the result tells you to use Undo instead of claiming success.

For the single grouped Undo step, enable Allow transaction boundary blocks under Permissions → Execution. The workflow creates preview and verification reports in the folder you choose.

Align SRT Caption Changes to Sequence Cuts

Use Align SRT Caption Changes to Sequence Cuts when subtitle changes that fall close to picture edits should land exactly on the corresponding cut frames. This is a focused timing adjustment: it keeps every caption's text, line breaks, order, and count unchanged.

Select one SRT and confirm that its time 00:00:00,000 matches the beginning of the sequence you intend to use. Then choose the video track whose edits should guide the captions. You can switch sequences while the setup dialog is open; the sequence active when you click OK becomes the target. Track numbers use Premiere's visible labels: enter 1 for V1, 2 for V2, and so on. The tool uses only boundaries between adjacent enabled clips on that one track. It reports muted tracks, disabled clips, gaps, and overlaps instead of guessing which edits should count.

The frame range controls which caption changes are candidates. With a value of 4, for example, a caption change up to four frames before or after a cut may move onto that cut. Caption changes farther away are intentionally left alone and are not treated as problems. If two changes compete for one cut, or a movement would make a caption too short, overlap another caption, violate the selected gap, or exclude precisely timed spoken words, the unsafe movement is skipped and listed for review.

Minimum caption duration and minimum gap are entered in sequence frames here as well. The tool uses the active sequence's exact frame rate and handles the SRT millisecond timestamps internally, including fractional rates such as 23.976.

The source SRT and Premiere sequence are never changed. When at least one safe movement exists, the workflow creates a new *.aligned-to-cuts.srt file in the selected output folder. Leave the folder empty to create an aligned-to-cuts folder beside the source SRT. A readable report shows how many caption changes moved, were already aligned, remained outside the chosen range, or need review. Matching precise timing files are found and updated automatically.

If the chosen track contains one flat rendered clip, first run Detect Scene Edits In Selected Clips with its option to apply cuts to the Timeline, review those detected edits, and then run this caption-alignment workflow. Scene detection is never started automatically.

Translate SRT Files To Multiple Languages With Workflow AI

In Translate SRT Files To Multiple Languages With Workflow AI, select one or more .srt files and enter comma-separated language codes such as de-DE, fr-FR, ja, or use Language name | tag for uncommon codes. Only the selected files are translated; other SRT files in the same folder are left alone.

Choose the kind of language work you need:

• Faithful translation (recommended) keeps the complete meaning, tone, factual claims, names, terminology, and calls to action without intentionally shortening or expanding the source.
• Marketing transcreation may adapt idioms, syntax, humor, and phrasing so the result has a natural brand voice and emotional effect in the target market instead of sounding literal.
• Follow my instructions applies no additional wording preset. Use the optional instructions field to describe the result you want.

Additional wording instructions take priority over the selected approach for wording and language. They do not change the workflow's file structure or timing method: every expected caption group must still be returned. A transcreation is never automatically certified as culturally or commercially correct; someone familiar with the target market should review it.

Choose how closely the translated captions should follow the source timing:

• Keep original caption timing keeps every translated caption at the same start and end time as the source. This is the safest choice when timing must stay aligned with an approved edit.
• Adjust timing for target language may redistribute translated text within each source sentence's time range for more natural reading. These adjusted internal timings are estimates and should be reviewed.

Adjust timing for target language usually gives marketing transcreation more room because a phrase can move within its complete source sentence instead of being forced into each original cue.

Workflow AI receives the caption text, while the tool protects the SRT structure and timing. Before saving a language, it checks that every expected caption has a non-empty result in the correct order. Text you explicitly mark as unchanged is enforced. Suspected changes to URLs, email addresses, handles, product codes, and numeric values are flagged for review without overriding an intentional editor instruction. A completely unchanged faithful-translation result is stopped; an unchanged individual group or a transcreation result is reported for review because names and intentional source-language wording can legitimately remain unchanged. Other completed files and languages remain available.

For better names and terminology, you can optionally provide:

• Project context (optional)
• Known names and terminology (optional)
• Text that must stay exactly unchanged (optional, one entry per line)
• Additional wording instructions (optional)

None of these fields must be filled in for the workflow to run. Use Known names and terminology to help Workflow AI understand the material. Use the separate exact-text field only for spelling that must survive unchanged, such as a brand, product code, URL, or approved slogan. The tool finds that text in the source regardless of how you capitalized the entry, then protects the source's exact spelling. A report notice identifies entries not found anywhere in the source, which helps catch typos.

The workflow requires a configured Workflow AI provider and the Python Runtime Pack.

Choosing The Translation Batch Size

Sentence groups per Workflow AI request controls how much subtitle context is translated at once. You can normally start with:

Workflow AI provider	Start with	Typical adjustment
Codex CLI or Claude Code	25 sentence groups	Try 20–40 when you want more context or smaller retry steps.
Local LM Studio model	6 sentence groups	Stay around 4–10 until the specific model has proven reliable for the target language.

A larger value is usually faster and gives the model more surrounding text, but more subtitles must be repeated if that batch fails a quality check. A smaller value takes longer but is usually easier for smaller local models. Model size, language pair, and hardware all matter, so treat these values as starting points rather than limits.

Recommended Workflow

1. Proofread important source transcripts, including project context and authoritative names.
2. Export selected source clips or generate an edit-relative SRT from the active sequence.
3. Analyze the generated SRT and optionally add review markers to the matching active sequence.
4. Translate or transcreate the reviewed source SRT into all required target languages.
5. Condense any source- or target-language SRT whose reading speed is still too high.
6. Reflow each SRT when its caption divisions or line breaks need improvement.
7. When required, align nearby caption changes to reviewed cuts in the active sequence.
8. Analyze the final result, review the reports, and import the approved SRT files into Premiere.

Keep the automation-agent-timing/ folder next to your SRT files while you work. The tools maintain and use matching timing files automatically. Translations do not copy source-language word timing because the translated words differ; they either keep the source caption times or estimate adjusted timing within the corresponding source sentence.

Where To Find The Results

The workflows keep the files you need in the folders you choose:

• SRT files in the chosen output folder;
• optional precise word-timing companion files in automation-agent-timing/, used automatically by later Caption Toolbox steps;
• a readable summary plus any review or per-language reports.

Temporary transcript exports and Workflow AI working files are removed after a successful run unless you enable the troubleshooting option to retain them.

Current Limitations

• Caption Track text cannot yet be read or written directly through Premiere UXP. The separate direct alignment tool changes only Caption Track timing.
• Align SRT Caption Changes to Sequence Cuts creates a new SRT. Align Caption Track Changes to Sequence Cuts can instead adjust a selected Caption Track in one verified, undoable Timeline operation.
• The active-sequence workflow does not yet create conflict markers or offer interactive playhead-assisted conflict choices.
• Reverse playback, speed changes, time remapping, merged clips, ambiguous multicam material, and offline media can prevent a transcribed Timeline clip from being used reliably.
• Clips without usable transcripts are skipped; if no usable dialogue remains, no SRT is created.
• Automatic track merging cannot observe Premiere Solo state.
• Translation quality still requires editorial review. Marketing transcreation additionally requires someone who understands the target market, brand, and campaign intent.
• Condensation can verify structure, protected text, and reading speed, but not whether every nuance or emphasis remains editorially correct.

Related Workflows

• Agent-Proofread Selected Transcripts
• Translate Selected Transcripts
• The Script Library