yaofanguk/video-subtitle-remover

★ 10,864PythonAudience · generalComplexity · 4/5LicenseSetup · hard

Mindmap

mindmap
  root((video-subtitle-remover))
    What it does
      Remove subtitles
      Remove watermarks
      Reconstruct background
    AI modes
      STTN fast real people
      LAMA still images
      ProPainter fast motion
    Input types
      Full video
      Region coordinates
      Image batch
    Hardware
      NVIDIA CUDA
      AMD DirectML
      Apple Silicon
      CPU fallback

mindmap root((video-subtitle-remover)) What it does Remove subtitles Remove watermarks Reconstruct background AI modes STTN fast real people LAMA still images ProPainter fast motion Input types Full video Region coordinates Image batch Hardware NVIDIA CUDA AMD DirectML Apple Silicon CPU fallback

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Remove hardcoded subtitles from a downloaded video before re-editing or re-subtitling it in another language.

USE CASE 2

Strip text watermarks from a batch of images in one command.

USE CASE 3

Target only the bottom strip of a video to speed up subtitle removal without processing the whole frame.

Tech stack

PythonPyTorchCUDADirectMLDocker

Getting it running

Difficulty · hard Time to first run · 1h+

Requires Python 3.11+ and a separate GPU-specific installation path (NVIDIA CUDA, AMD DirectML, or CPU-only). Docker images available for each.

Use freely for personal and commercial projects as long as you include the license notice. Apache 2.0.

In plain English

Video Subtitle Remover (VSR) is a Python tool that uses AI to remove burned-in subtitles and text watermarks from videos and images. Burned-in subtitles (sometimes called hardcoded subtitles) are text overlaid directly onto the video frames, as opposed to subtitle tracks that can be toggled on and off. This tool detects where the text appears and fills in the background behind it using AI image reconstruction techniques, so the output looks like the text was never there. The tool runs entirely on your own machine with no third-party API required. It offers both a graphical interface and a command-line version. You can remove subtitles from an entire video automatically, specify exact screen coordinates to target only a particular area (such as the bottom strip where subtitles typically appear), or batch-process multiple images to remove watermarks. The README (written in Chinese) describes three AI-based fill methods with different trade-offs. The STTN mode works best on footage of real people and is the fastest option. The LAMA mode produces better results on still images and animated content. The ProPainter mode is the slowest but handles videos with a lot of fast movement more accurately. The tool supports NVIDIA GPU acceleration via CUDA, AMD and Intel GPU acceleration via DirectML on Windows, Apple Silicon on macOS, and CPU-only mode for systems without a compatible GPU. Pre-built packages for Windows are available as downloads for each GPU configuration. Docker images are also provided for each environment. Setup requires Python 3.11 or newer, and the installation steps differ depending on whether you have an NVIDIA, AMD, Intel, or no GPU. The README provides separate installation instructions for each path. The project is open source under the Apache 2.0 license, which allows free use in personal and commercial projects.

Copy-paste prompts

Prompt 1

I have an MP4 with burned-in subtitles at the bottom. Walk me through the exact command to remove them using video-subtitle-remover in STTN mode.

Prompt 2

I want to remove watermarks from 200 JPEG images at once using video-subtitle-remover. Show me the batch command.

Prompt 3

The subtitle removal output looks blurry on my animated video. Which mode, STTN, LAMA, or ProPainter, should I use and why?

Prompt 4

How do I install video-subtitle-remover with CUDA support on Windows without Docker?

Open on GitHub → Explain another repo

← yaofanguk on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.