DIGIT Batch SRT From Video

The single-file version of this is handy. This one is the version you want when you have twenty-seven videos sitting in a folder tree and the idea of running them one at a time makes you want to quit the project. DIGIT Batch SRT From Video recursively scans a folder, transcribes every video to subtitles with Gemini, and either saves an .srt next to each clip or collects everything into one project folder. Set it going before lunch, come back to captions.

It's built for exactly the boring production job: dozens of files, unattended, resumable. The skip logic is smarter than most - it checks for the specific output you asked for (a sidecar .srt, a burned-in _subtitled video, or both), so re-running after a partial failure only touches the files that actually failed.

How it works

For each video: ffmpeg extracts a mono 16kHz WAV, Gemini transcribes it with timestamps, then the output runs through the same post-processing pipeline as the single-file node - hallucination removal, line-length enforcement (Netflix's 42-char standard by default), optional frame padding and snap-to-frame. You can also burn the subtitles straight into the video with full styling control instead of (or alongside) saving files.

The inputs that matter:

• video_folder - top-level folder; it scans recursively. Required.
• file_types - all, or filter to mp4, mov, mxf, mkv, avi, m4v, qt when a folder mixes formats you don't all want.
• subtitle_output - srt_only, burn_in_only, or both.
• output_mode - alongside_video (.srt next to each clip, wherever it lives) or projekts_auto_srt (all output collected to one project folder).
• overwrite - off by default, so existing outputs are skipped.
• identify_speakers - on by default; labels SPEAKER 1, SPEAKER 2, etc.
• translate_to - transcribe then translate, preserving all timing.
• output_format - srt, vtt, ass, txt, or all.

For burn-in there's the styling block: font_name, font_size, font_color, outline_width, shadow_depth, position, margin_v. And model (gemini-2.5-flash default) plus delay_seconds (1.0) to pace the API calls.

Outputs: log (a per-file OK/SKIPPED/ERROR listing with relative paths), transcribed_count, and output_folder.

Installing it

Same pack as the rest of the DIGIT family:

cd ComfyUI/custom_nodes
git clone https://github.com/thedepartmentofexternalservices/comfyui-digit.git
cd comfyui-digit
pip install -r requirements.txt

ComfyUI Manager works too - search comfyui-digit. Two environment requirements matter here specifically: ffmpeg must be on your PATH (it's a system dependency, not a pip one), and you need the pack's standard GCP setup - gcloud auth application-default login, a Vertex AI–enabled project. Set gcp_project_id on the node or let it auto-detect.

Where it trips people up

Most issues are the two setup ones: ffmpeg missing (you'll see "ffmpeg audio extraction failed" in the log) and credentials not authenticated. Both are one-line fixes. Worth knowing before you start: transcription is per-call billed to your GCP account, and a folder full of long videos adds up, so check delay_seconds and spot-test one file first. The skip logic means you can always re-run and only the failures get processed - that's the safety net.