Local-first video cleanup SDK

Plan the wipe.
Then run it.

Detect subtitles, watermarks, logos, and timestamps into a reviewable, time-aware WipePlan. Inspect remove/keep tracks, follow progress, cancel safely, and inpaint locally.

Get started View on GitHub

quickstart.py

from videowipe import WipeEngine, WipeRequest

with WipeEngine(task="clean") as engine:
    plan = engine.plan(WipeRequest(
        video="input.mp4",
        output_dir="plan/",
    ))
        

Features

Plan first. Pixels second.

One local engine for detection, review, and inpainting — embeddable in Python workers, CLI workflows, and your own product.

Reviewable WipePlan

Detection produces a plain-JSON plan with remove/keep tracks, time ranges, and precise masks. Review or edit the decision before any pixels are changed.

Pluggable backends

Default STTN runs temporal plans locally. The validated ProPainter adapter and --external-command support static plans without time ranges.

Observable SDK

plan() and run() report structured progress and accept cooperative cancellation. Reuse one engine across a batch or worker.

Pipeline

Detect. Review. Execute.

Separate the cleanup decision from the expensive inpainting step.

Detection

Find text overlays and build candidate tracks. Balanced and sensitive modes densely recheck detector-backed remove ranges; coarse fallbacks stay explicit.

Plan review

Inspect remove/keep actions, temporal segments, and precise masks. Optional OCR, intent, and local-agent selection help choose the right targets.

Local execution

Execute temporal plans with STTN. ProPainter and other file-based external models accept static plans without time ranges. Source binding keeps mask decisions reproducible.

Usage

Python SDK or CLI

Embed the structured engine or use the command-line adapter over the same pipeline.

python — plan and run

from videowipe import CancellationToken, WipeEngine, WipeRequest

def report(event):
    print(event.phase, event.completed, event.total)

with WipeEngine(task="clean") as engine:
    plan = engine.plan(
        WipeRequest(video="input.mp4", output_dir="plan/"),
        on_progress=report, cancellation=CancellationToken(),
    )
    result = engine.run(WipeRequest(
        video="input.mp4", output_dir="result/", plan=plan,
    ), on_progress=report)
          

terminal — quick clean

# Generate a reviewable WipePlan
videowipe clean input.mp4 --preview -o plan/

# Execute the reviewed plan
videowipe clean input.mp4 \
  --plan plan/wipe_plan.json -o result/

# Or run the full pipeline directly
videowipe clean input.mp4 -o result/

# Docker (CPU)
docker run --rm -v "$(pwd)":/data \
  ghcr.io/kkenny0/videowipe:latest \
  clean /data/input.mp4 -o /data/result/
          

3.10+

Python version required

Reviewable plan contract

Manual masks needed

GPL-3.0

Open-source license

Get started

Build your cleanup pipeline.

Clone the source, choose ONNX Runtime or PyTorch, and embed the same local engine used by the CLI and Web adapter.

Install from source View source on GitHub

Support

Help keep VideoWipe maintained.

If VideoWipe saves you time on subtitle, watermark, or text-overlay cleanup, support helps keep model packaging, Docker images, detection tuning, and documentation maintained.

Support the project

Plan the wipe.Then run it.