Lada is a tool designed to recover pixelated adult videos (JAV). It helps restore the visual quality of such content, making it more enjoyable to watch.
- Recover Pixelated Videos: Restore pixelated or mosaic scenes in adult videos.
- Watch/Export Videos: Use either the CLI or GUI to watch or export your restored videos.
After opening a file, you can either watch the restored version of the provided video within the app (make sure you've enabled the Preview toggle) or export it to a new file.
Tip
If you've installed the flatpak then it should be available in your regular application launcher. You can also run it via flatpak run io.github.ladaapp.lada.
Otherwise, if you followed the Developer Installation section, use the command lada to open the app (ensure you are in the root directory of this project).
Note
If you installed Lada from Flathub and drag-and-drop doesn't work, your file browser might not support File Transfer Portal. You can fix this by:
- Switching or updating your file browser to one that supports it.
- Granting the app filesystem permissions (e.g., via Flatseal so it can read files directly).
- Using the 'Open' button to select the file instead of drag-and-drop.
Additional settings can be found in the left sidebar.
You can also use the command-line interface (CLI) to export restored videos:
lada-cli --input <input video path> --output <output video path>
Tip
If you've installed the app via Flathub then the command would look like this (instead of host permissions you could also use --file-forwarding option):
flatpak run --filesystem=host --command=lada-cli io.github.ladaapp.lada --input <input video path> --output <output video path>Alternatively, set an alias in your shell to simplify the command and match the non-flatpak version:
alias lada-cli="flatpak run --filesystem=host --command=lada-cli io.github.ladaapp.lada"Tip
If you installed via Docker, use:
docker run --rm --gpus all --mount type=bind,src=<path to input/output video dir>,dst=/mnt ladaapp/lada:latest --input /mnt/<input video file> --output /mnt/<output video file>Tip
Lada writes the restored video to a temporary file before combining it with the audio stream from the original file and saving it to the selected destination. The default location is /tmp.
You can change this by setting the TMPDIR environment variable. On flatpak, you can pass --env=TMPDIR=/my/custom/tempdir or use Flatseal to set it permanently.
For more information about additional options, use the --help argument:
Lada utilizes specialized models for the two main steps of the processing pipeline: Detection and Restoration. You can choose different models for each task.
Mosaic Restoration Models:
- basicvsrpp-v1.2 (Default) A general-purpose model trained on diverse video scenes. Delivers mostly good results.
- deepmosaics: Restoration model from the project DeepMosaics. Worse quality than basicvsrpp-v1.2.
Note
The DeepMosaics model should be worse in most/all scenarios. It’s integrated because the DeepMosaics project is not maintained anymore, and I wanted to provide an easy way to try it out and compare.
Mosaic Detection Models:
- v3.1-fast (Default): Fast and efficient.
- v3.1-accurate: More accurate than v3.1-fast, but slower. Not always better than v2.
- v2: Slowest of all but often provides better mosaic detection than v3.1-accurate.
You configure the models in the side panel, or when using the CLI by specifying path and type of the model as arguments.
Don't expect this to work perfectly, some scenes can be pretty good and close to the real thing. Other scenes can be rather meh and show worse artifacts than the original mosaics.
You'll need a Nvidia (CUDA) GPU and some patience to run the app. If your card has at least 4-6GB of VRAM then it should work out of the box.
The CPU is used for encoding the restored video so shouldn't be too slow either (current version does not ship with GPU video encoders). You can adjust encoder options to your needs on the CLI.
The app also needs quite a bit of RAM for buffering to increase throughput. For 1080p content you should be fine with 6-8GB RAM, 4K will need a lot more.
If you want to use watch the restored video in the GUI preview mode in real-time you'll need a pretty beefy machine or you'll see the player pausing until next restored frames are computed. GUI Preview mode will need less CPU resources as it will not have to encode the video but will use more additional RAM for buffering.
If your GPU is not fast enough to watch the video in real-time you'll have to export it first and watch it later with your favorite media player.
Technically running the app on your CPU is also supported where supported is defined as: It will not crash but processing will be so slow you wish you haven't given it a try.
Here are some speed performance numbers using Lada v0.7.0 on my available hardware to give you an idea what to expect (used libx264 codec with default settings; RTX 3090 results are limited by CPU encoding and could be faster with other settings):
| Video name | Video description | Video duration / resolution / FPS |
Lada runtime / FPS Nvidia RTX 3050 (Laptop GPU) |
Lada runtime / FPS Nvidia RTX 3090 (Desktop GPU) |
|---|---|---|---|---|
| vid1 | multiple mosaic regions present on all frames | 1m30s / 10920x1080 / 30 FPS | 3m36s / 12 FPS | 1m33s / 30 FPS |
| vid2 | single mosaic region present on all frames | 3m0s / 1920x1080 / 30 FPS | 4m11s / 21 FPS | 2m16s / 39 FPS |
| vid3 | half of the video doesn't have any mosaics present, the other half mostly single mosaic per frame |
41m16s / 852x480 / 30 FPS | 26m30s / 46 FPS | 10m20s / 119 FPS |
It may or may not work on Windows and Mac and other GPUs. You'll have to try to follow Developer Installation below and see how far you get.
Patches / reports welcome if you are able to make it run on other systems or have a suggestion how to improve the documentation.
The easiest way to install the app (CLI and GUI) on Linux is via Flathub:
Caution
The Flatpak version works only with x86_64 CPUs and Nvidia/CUDA GPUs. Ensure your system uses the official NVIDIA drivers, not nouveau.
CPU-only inference is technically possible but refer to the notes in the Status section first.
The app is also available via Docker (CLI only). You can pull it from Docker Hub with the following command:
docker pull ladaapp/lada:latestThis image has the same limitations as the Flatpak version: x86_64 CPU + Nvidia/CUDA GPU. To use your GPU, make sure to install the Nvidia Container Toolkit first.
If you prefer not to use Flatpak or Docker, have different hardware specifications, or are using a non-Linux system, follow the Developer Installation steps. Contributions are welcome if someone can package the app for other systems!
This section describes how to install the app from source.
Note
If you're on Windows and don't want to use WSL then follow the Developer Installation (Windows) instructions instead.
-
Get the source
git clone https://github.com/ladaapp/lada.git cd lada -
Install system dependencies with your system package manager or compile/install from source
- Python >= 3.12
- FFmpeg >= 4.4
Tip
Arch Linux: sudo pacman -Syu python ffmpeg
Ubuntu 25.04: sudo apt install python3.13 python3.13-venv ffmpeg
Ubuntu 24.04: sudo apt install python3.12 python3.12-venv ffmpeg
-
Create a virtual environment to install python dependencies
python3 -m venv .venv source .venv/bin/activate -
Install python dependencies
python -m pip install -e '.[basicvsrpp]' -
Apply patches
On low-end hardware running mosaic detection model could run into a timeout defined in ultralytics library and the scene would not be restored. The following patch increases this time limit:
patch -u .venv/lib/python3.1[23]/site-packages/ultralytics/utils/ops.py patches/increase_mms_time_limit.patch
Disable crash-reporting / telemetry of one of our dependencies (ultralytics):
patch -u .venv/lib/python3.1[23]/site-packages/ultralytics/utils/__init__.py patches/remove_ultralytics_telemetry.patch
Compatibility fix for using mmengine (restoration model dependency) with latest PyTorch:
patch -u .venv/lib/python3.1[23]/site-packages/mmengine/runner/checkpoint.py patches/fix_loading_mmengine_weights_on_torch26_and_higher.diff
-
Download model weights
Download the models from the GitHub Releases page into the
model_weightsdirectory. The following commands do just thatwget -P model_weights/ 'https://github.com/ladaapp/lada/releases/download/v0.7.1/lada_mosaic_detection_model_v3.1_accurate.pt' wget -P model_weights/ 'https://github.com/ladaapp/lada/releases/download/v0.7.1/lada_mosaic_detection_model_v3.1_fast.pt' wget -P model_weights/ 'https://github.com/ladaapp/lada/releases/download/v0.2.0/lada_mosaic_detection_model_v2.pt' wget -P model_weights/ 'https://github.com/ladaapp/lada/releases/download/v0.6.0/lada_mosaic_restoration_model_generic_v1.2.pth'
If you're interested in running DeepMosaics' restoration model you can also download their pretrained model
clean_youknow_video.pthwget -O model_weights/3rd_party/clean_youknow_video.pth 'https://drive.usercontent.google.com/download?id=1ulct4RhRxQp1v5xwEmUH7xz7AK42Oqlw&export=download&confirm=t'
Now you should be able to run the CLI by calling lada-cli.
-
Install everything mentioned in Install CLI
-
Install system dependencies with your system package manager or compile/install from source
- Gstreamer >= 1.14
- PyGObject
- GTK >= 4.0
- libadwaita >= 1.6 [there is a workaround mentioned below to make it work with older versions]
Tip
Arch Linux:
sudo pacman -Syu python-gobject gtk4 libadwaita gstreamer gst-plugins-base gst-plugins-good gst-plugins-bad gst-plugins-ugly gst-plugins-base-libs gst-plugins-bad-libs gst-plugin-gtk4Ubuntu 25.04:
sudo apt install gcc python3-dev pkg-config libgirepository-2.0-dev libcairo2-dev libadwaita-1-dev gir1.2-gstreamer-1.0
sudo apt install libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev gstreamer1.0-plugins-good gstreamer1.0-plugins-bad gstreamer1.0-plugins-ugly gstreamer1.0-pulseaudio gstreamer1.0-alsa gstreamer1.0-tools gstreamer1.0-libav gstreamer1.0-gtk4Ubuntu 24.04:
sudo apt install gcc python3-dev pkg-config libgirepository-2.0-dev libcairo2-dev libadwaita-1-dev gir1.2-gstreamer-1.0
sudo apt install libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev gstreamer1.0-plugins-good gstreamer1.0-plugins-bad gstreamer1.0-plugins-ugly gstreamer1.0-pulseaudio gstreamer1.0-alsa gstreamer1.0-tools gstreamer1.0-libav
#
# The gstreamer plugin gtk4 is not available as a binary package in Ubuntu 24.04 so we have to build it ourselves:
# get the source code
git clone https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs.git
cd gst-plugins-rs
# install plugin dependencies
sudo apt install rustup libssl-dev
rustup default stable
cargo install cargo-c
# now we can build and install the plugin. Note that we're installing to the system directory, you might want to adjust this and use GST_PLUGIN_PATH
cargo cbuild -p gst-plugin-gtk4 --libdir /usr/lib/x86_64-linux-gnu
sudo -E cargo cinstall -p gst-plugin-gtk4 --libdir /usr/lib/x86_64-linux-gnu
# if the following command does not return an error the plugin is correctly installed
gst-inspect-1.0 gtk4paintablesink- Install python dependencies
python -m pip install -e '.[gui]'
Tip
Ubuntu 24.04: The following is patch is only needed on Ubuntu 24.04 as its version of libadwaita is too old. For newer distributions where libadwaita >= 1.6 is available this should not be applied.
patch -u -p1 -i patches/adw_spinner_to_gtk_spinner.patchTip
If you intend to hack on the GUI code install also gui-dev extra: python -m pip install -e '.[gui-dev]'
Now you should be able to run the GUI by calling lada.
For instructions on training your own models and datasets, refer to Training and dataset creation.
This project builds upon work done by these fantastic individuals and projects:
- DeepMosaics: Provided code for mosaic creation during dataset creation/training. Also inspired me to start this project.
- BasicVSR++ / MMagic: Used as the base model for mosaic removal.
- YOLO/Ultralytics: Used for training mosaic and NSFW detection models.
- DOVER: Used to assess video quality of created clips during the dataset creation process to filter out low-quality clips.
- DNN Watermark / PITA Dataset: Used most of its code for creating a watermark detection dataset used to filter out scenes obstructed by text/watermarks/logos.
- NudeNet: Used as an additional NSFW classifier to filter out false positives by our own NSFW segmentation model
- Twitter Emoji: Provided eggplant emoji as base for the app icon.
- Real-ESRGAN: Used their image degradation model design for our mosaic detection model degradation pipeline.
- PyTorch, FFmpeg, GStreamer, GTK and all other folks building our ecosystem