1
Load media
Open a video (e.g. MP4, MKV, AVI) or an image (PNG, JPG, WebP, BMP). FFmpeg and OpenCV prepare frames and metadata.
WatermarkRemoverLiteVideo
For Windows · Your media stays on your PC
Remove watermarks from video and image with high-quality inpainting
WatermarkRemoverLiteVideo clears logos, stamps, and burned-in captions from video and still images with natural, convincing fills powered by LaMa. Edit on a full timeline, refine masks by hand or with smart assist, then export your clip—processing stays local, so your files are not sent to the cloud for this workflow.
Demo video
Add a
demo.mp4 file in the assets/ folder to show your own demo here, or check the browser console if the file fails to load.
Demo video showing what WatermarkRemoverLiteVideo can do — before/after results, mask refinement, and clean exports.
Workflow
How the application works
The core is an inpainting pipeline: a mask is generated over the watermark region and the LaMa model fills that area by predicting texture and color consistent with the surroundings. For video, processing runs frame by frame (with quality options and temporal smoothing to reduce flicker).
2
Define the mask
Manual: draw an ROI rectangle over the watermark; choose a solid mask, adaptive text detection, or soft alpha. Auto: corner heuristics (top-hat, Canny, Otsu) for typical edge watermarks.
3
LaMa AI
The inpainting engine reconstructs the patch. You can set the maximum patch side (512–1024 px or unlimited) and passes for opaque logos.
4
Export
The output video is encoded with FFmpeg (CRF, preset, copied or AAC audio). By default output goes to output/videos next to the executable or project root.
Product
Key features
Real capabilities from the app: single LaMa engine, video and image workflows, and fine-grained quality controls.
Video mode with a timeline
Temporal editor: multiple watermark clips, drag ranges, resize edges, double-click to jump to clip start. Ideal when the mark appears only in segments or moves.
Image mode: before / after
Load an image, process, and compare before/after previews. Export to PNG or JPG when you are happy with the result.
Manual: fixed ROI or multiple over time
Fixed ROI: one region for the whole video. Multi + time: different rectangles per segment; the mask can be re-estimated if the ROI changes (motion or scene variation).
Auto detection (corners)
Heuristic detection aimed at corner watermarks: fine text, high contrast, or semi-transparent overlay, using classical vision (no extra ML for the mask).
Adjustable inpainting quality
Maximum patch size to the model (512 px fast, 768 balanced for 1080p, 1024 for fine logos/4K, or unlimited for best quality with more VRAM). Optional multiple passes for opaque logos. Temporal patch smoothing to reduce flicker between frames.
Diagnostics console
Progress messages, device (CPU/GPU), PyTorch backend, resolution and FPS of the loaded video, and export paths for troubleshooting.
Spanish and English UI
Built-in translations for toolbars, dialogs, and tooltips in both languages.
Executable and local dependencies
Project ready for build_exe.bat (PyInstaller). Models and FFmpeg can be bundled; rebuild the .exe after changes to models or paths.
Privacy by design
Models and processing run locally. You do not need to upload your videos to a server to use LaMa in this app.
Use cases
What kinds of watermarks you can treat
LaMa is strong at filling regions, but results depend on background complexity and whether the mask fully covers the mark. Automatic detection targets corners; for central or moving marks, manual mode or time-based clips usually work better.
Often works well
- Logos or text in corners (auto or manual).
- Semi-transparent marks over textured backgrounds (alpha / detect mode).
- Static text or ROI adjusted frame-by-frame on the timeline.
- Small channel badges or “subscribe” overlays in a fixed area.
- 1080p video with 768–1024 px patches for fine detail.
Limitations and good practices
- Very busy backgrounds behind the mark may show artifacts; try more passes or a larger patch.
- Marks covering a very large area are harder: reconstruction has less visible context.
- Auto-detect does not guarantee marks outside corners; use manual in those cases.
- Respect copyright and licenses for the content you process.
Before you install
System requirements for the downloadable app
These notes are for end users of the Windows build from WRLV.rar. You normally do not install Python, pip, or FFmpeg yourself.
- Operating system
- Windows 10 or Windows 11, 64-bit.
- Memory (RAM)
- 8 GB minimum; 16 GB+ recommended for HD/4K.
- Graphics (GPU)
- NVIDIA GPU with enough VRAM recommended; CPU-only is much slower.
- Disk space
- Several GB free for app, models, and videos.
- Extract and run
- Keep folder layout after unzip; run the .exe from the extracted folder.
- Exported videos
- Default:
output/videosnear the app.
More from AmericaPixelGames
Other software developed by AmericaPixelGames
If you want to explore more tools from the same studio, here are two additional projects published by AmericaPixelGames.
FFStudio
FFStudio is a desktop video editor created for creators who need a clear workflow for clip selection, crop editing, reframing, and export.
FlexJS
Another AmericaPixelGames software project with its own dedicated website and documentation.
Games by AmericaPixelGames
Also explore some games from the studio
These titles come from the public metadata of AmericaPixelGames game pages and show another side of the studio's work.
Nexus Blasters
A retro-inspired arcade space shooter where elite pilots battle hostile forces, collect cosmic stars, and push humanity one step closer to recovery.
View game details
Wayra
A rhythm-charged runner built around music, reflexes, and flow, where every collected element enriches the soundtrack as the pace rises.
View game details
Musical Path
A music-learning racing game that mixes theory, memory, and agility, helping players identify notes, degrees, and emotions while they play.
View game detailsDownload for Windows
Want the Windows program ready to run? Get WRLV.rar from the download page.