Automating subtitle correction and translation in DaVinci Resolve with AI

Neto Pompeu —

Automating subtitle correction and translation in DaVinci Resolve with AI

At Primo-Studio, we regularly produce audiovisual content: reports, live captures, promotional videos, documentaries. Almost systematically, we have to deliver a subtitled version — often in French, sometimes also in English for international clients or festivals.

Manual subtitling is slow. Really slow. When you add up transcription, spelling review, then translating from one language to another, you easily spend 3 to 5 hours for 20 minutes of video. On a project with several videos per week, that's economically unsustainable.

So we built two Python scripts that integrate directly into DaVinci Resolve and automate the task using AI. Here's how it works, and what it has enabled us to do.

The problem: the friction of subtitling

DaVinci Resolve handles the timeline and subtitle tracks very well. The built-in editor is decent. But two operations remain slow:

  1. Correcting French spelling on an auto-generated transcription — typos, missing accents, mis-detected proper nouns, technical words. Line by line, it drags on.
  2. Translating to English a finished French track. By hand, it takes several hours per project, and you must preserve the exact timecodes to keep the track aligned with the image.

Market tools that do this exist — but either they're expensive at recurring use, or they push the file out to an external platform (time loss + loss of control on format), or the translation quality is too generic for our level of requirement.

The solution: two Python scripts integrated into Resolve

DaVinci Resolve exposes a Python API that lets scripts manipulate the open project: read tracks, read subtitles, modify content, save. We wrote two scripts that install in Resolve's Fusion/Scripts/Edit/ folder. They then appear in the Workspace → Scripts → Edit menu.

Script 1: French spelling correction

It takes the active French subtitle track, sends each line to the Claude API, and gets back the same line with spelling corrections (accents, typos, agreements). Timecodes don't move. Structure stays identical.

In pseudo-code:

``` project = Resolve.GetProjectManager().GetCurrentProject() timeline = project.GetCurrentTimeline() subtitle_track = timeline.GetActiveSubtitleTrack()

for item in subtitle_track.GetItems(): raw_text = item.GetText() corrected = call_claude_api( prompt="Correct this French subtitle line...", content=raw_text ) item.SetText(corrected)

project.Save() ```

Claude returns the corrected text in a few hundred milliseconds per line. For a 20-minute video with about 200 subtitle lines, the script runs in two to three minutes total.

Script 2: FR → EN translation

Same principle, but this time we ask Claude to translate line by line into English. The script duplicates the French track, then fills the English track with the translations. We keep line-by-line control: we re-read, we tune nuances, we shorten if a line is too long on screen (limit ~42 characters per line for readability).

Pseudo-code:

``` track_fr = timeline.GetSubtitleTrack("FR") track_en = timeline.DuplicateTrack(track_fr, lang="EN")

for item_fr, item_en in zip(track_fr.GetItems(), track_en.GetItems()): text_fr = item_fr.GetText() text_en = call_claude_api( prompt="Translate this French subtitle to idiomatic English, max 42 chars...", content=text_fr ) item_en.SetText(text_en)

project.Save() ```

The key: we specify in the prompt the context (video subtitle), the maximum length, and the desired tone (idiomatic, not literal). Without that, you get a basic Google Translate output. With it, you get clean translation, ready for human review.

The real time saving

On a typical 20-minute video project:

  • Before: 3 to 5 hours of transcription + correction + translation
  • After: 30 to 45 minutes (script + human review + stylistic adjustments)

A 5× to 8× reduction in subtitling time. Over a normal month, that's 15 to 25 hours saved, redirected to higher-value work: editing, color grading, audio mixing, client relationship.

Why we build our own tools

This is where it becomes a broader topic. Many agencies use AI in consumption mode — they open ChatGPT, paste a brief, take the answer. It works, but it's limited.

We chose to embed AI directly into our business tools. The Resolve scripts are one example. We've also developed automations for Photoshop, our internal CRM, multilingual product description generation. Every time, the principle is the same: take an operational bottleneck, identify the precise place where AI can intervene, and code a small custom tool.

It's an upfront investment — a few hours to write and test a script — that pays back in one or two weeks of use, then generates value for years.

Why this matters for a digital agency in 2026

The agency market (communication / web / digital) is being reshaped. AI tools are upending production costs: what used to take ten hours takes one, what needed three people now needs one. Agencies that don't transform will find themselves billing ten hours of work that AI does in one — and the client will go elsewhere.

The question is no longer «will AI replace my craft?» but «can I use AI to do my craft 5 to 10 times faster, at higher quality, at a better price?»

This is the transformation we support at Primo-Studio. For our own agency first — we don't preach what we don't practice. Then for Guianese businesses that want to modernize their workflows, across all sectors: audiovisual, marketing, administration, retail, training.

Going further

On the same theme, see also:

Want to modernize your workflows with AI?

Contact us. Whether you're a production company, an agency, an institution or an SME in French Guiana, we help you identify the bottlenecks in your processes and automate them intelligently — with the right tools, without hacks and without excessive dependence on an external platform.