How to install Drift on Ubuntu: AI video editor

Video editing on Linux has become much more accessible. Drift is a free desktop editor, based on Qt 6 and FFmpeg, that brings together a multi-track timeline, effects, audio, titles, automatic subtitles, and integration with AI agents. It is an interesting option for those who produce tutorials, technical videos, shorts, and demonstrations and want to work locally, without an account, subscription, or watermark. In this guide you will install Drift on Ubuntu, configure the AI packages, and learn the shortcuts that really help with your first project.

What Drift is and where it fits

The Drift is an open source desktop video editor under GPLv3. The proposal is to concentrate in the same project the timeline editing, texts, effects, transitions, audio treatment, masks, subject cutout, export and synchronized subtitles. The preview and export use the same compositor, which reduces the surprise of approving an image on screen and receiving another in the final file.

It does not replace the FFmpeg: it uses exactly that media foundation underneath. The difference is that Drift offers a visual interface for cutting, aligning, adding B-roll, adjusting audio and assembling a quick delivery. To compare other alternatives, see also the list of easy-to-install video editors on Ubuntu.

Recommended installation on Ubuntu: Flatpak

The path recommended by the project for Linux is Flathub. On a fresh Ubuntu, install Flatpak, add the repository once and install the official identifier org.cutwire.Drift:

sudo apt update
sudo apt install -y flatpak
sudo flatpak remote-add --if-not-exists flathub \
  https://dl.flathub.org/repo/flathub.flatpakrepo
sudo flatpak install -y flathub org.cutwire.Drift

# abrir o editor
flatpak run org.cutwire.Drift

After the first installation, the application will appear in the system menu. To update only Drift, use:

flatpak update org.cutwire.Drift

The installation command and the application ID are those published in the official README. If Flathub is already configured, the remote-add can simply be ignored.

Portable alternative: AppImage

Prefer not to use Flatpak or want to keep a portable copy? Download the AppImage from the latest release of Drift, and replace ARQUIVO-BAIXADO.AppImage with the actual name of the downloaded file:

mkdir -p "$HOME/Applications"
chmod +x "$HOME/Downloads/ARQUIVO-BAIXADO.AppImage"
mv "$HOME/Downloads/ARQUIVO-BAIXADO.AppImage" \
  "$HOME/Applications/Drift.AppImage"

"$HOME/Applications/Drift.AppImage"

This fixed location is especially useful for those planning to use the MCP: an agent configuration that points to ~/Applications/Drift.AppImage will remain valid after the window is closed. When there's an update, download the new AppImage, replace this file, and run it again.

First minutes: a workflow that avoids rework

1. Create a project and import the footage. First organize the main video, B-roll, images, audio tracks, and sound effects.
2. Build the rough cut on the main track. Use trim and split to remove beginnings, endings, pauses, and mistakes. Drift offers a multi-track timeline, snapping, ripple, undo, and track controls.
3. Fill in with context. Place images, B-roll, or titles in upper tracks; do not hide important information with an overlay.
4. Treat the audio before the visual finishing. Adjust the voice volume, use fade in and out, and only then add background music. The editor includes features such as EQ, compressor, gate, and voice tools.
5. Do a final pass with sound. Watch from beginning to end before exporting. It's the best time to catch a dry cut, a word lagging behind in the caption, or music that's too loud.

For terminal tutorials, a simple rule works well: cut the intervals where nothing happens, preserve small pauses between topics, and use zoom or highlighting only when there is something the reader needs to see. Speed is not synonymous with haste; it is clarity without wasted time.

Automatic subtitles with Whisper: how to install and use

Drift keeps heavy models and resources outside the main installation. This makes the initial download smaller, but it means some AI functions require an additional package. To generate captions, open Settings → Extra packs and install Auto Captions; this package contains the Whisper model used to turn speech into timed caption lines.

Then, open the captions panel, generate the captions from the audio track, and review each segment. The transcription is a draft: command names, distributions, acronyms, addresses, and technical terms deserve manual review. Fix the wording and the entry/exit point of each line before publishing.

For good mobile readability, keep at most two lines on screen, avoid huge sentences, and choose high contrast. Yellow subtitles with shadow or dark background usually work well over dark imagery; on light imagery, prefer a semi-transparent background or outline so you don't sacrifice readability.

If you see a message saying the Whisper model wasn't found, you don't need to look for model files on the internet or copy weights into a random folder: go back to Extra packs and install or repair the package Auto Captions. The documentation for the official addons states that these packages are downloaded on demand and verified by the application.

GPU acceleration: choose the right package

In addition to the subtitle model, there's the package AI Engine. It provides the ONNX Runtime in variants for CPU, NVIDIA/CUDA, and general-purpose GPU. On a machine with a compatible NVIDIA card, choose the NVIDIA/CUDA variant; in other scenarios, pick whatever option the package manager itself offers for your hardware. The advantage of the addons system is that it doesn't install a heavy AI runtime for users who only need cuts and transitions.

GPU accelerates features based on models, but it doesn't automatically fix a bad cut or replace subtitle review. Before buying hardware or switching inference servers, it's worth understanding the difference between GPU and TPU and how they work graphics cards.

Cuts, rhythm, and transitions: tricks that make the video more fluid

• Cut on motion or on a change of idea. A cut feels more natural when it happens together with an action, keyword, or screen change.
• Use a fade when the intent is to wrap things up. For regular cuts, the most subtle transition is usually the best. Save wipes, glitches, and slides for block changes that truly call for attention.
• Don't speed up the entire recording. Speed up repetitive tasks, lengthy installations, or unexplained typing; keep at normal speed the section where a command or the interface needs to be understood.
• Leave some breathing room in the audio. Cutting right up to the last millisecond of speech produces jittery video. Short fades kill pops and add polish between music, voice, and effects.
• Make copies before major changes. Drift's linear history lets you undo, but a copy of the project before rhythm, color, or effects tests saves time.

Drift also includes speed ramps, reverse, fades, video and audio effects, visual templates, transitions, and music sync features. Start with one change at a time: test it, watch the transition in context, and only then add another layer.

Automating editing with MCP and AI agents

A standout feature of Drift is the integrated local MCP server. When you enable Settings → Agent access, compatible agents like Cursor and Claude Code can act on the open project: import media, position and trim clips, apply effects, capture an image of the composition, and start exports.

The option stays disabled each time the application is opened. When enabled, Drift shows a configuration snippet to copy. In HTTP mode, the Bearer token changes every session: copy the freshly generated snippet, keep it local, and never publish that token in repositories, screenshots, or posts.

For a local stdio connection, the official documentation shows the structure below. Adjust the executable path to your installation; if you followed the AppImage section, the path might be /home/SEU_USUARIO/Applications/Drift.AppImage.

{
  "mcpServers": {
    "drift": {
      "command": "/caminho/para/drift",
      "args": ["--mcp-stdio"]
    }
  }
}

A safe workflow for an agent is: consult the tool catalog, inspect the timeline, apply a small batch of changes, capture a preview, and check the result. Batches made with apply go into a single undo step, but they are not atomic: if one operation fails, the previous ones may have already been applied. For this reason, review the response and save the project at important points.

Those who use AI assistants in the terminal can turn repetitive tasks into a script: cut identified silences, position a bumper, create captions, apply the color style, and export a review version. The editorial decision — what is an error, a dramatic pause, or important information — remains human.

Headless mode: useful for automation and servers

Drift can also start without a window with --headless, serving MCP for automations. The documentation itself warns that rendering, capture, and export still need an OpenGL context. On a Linux host without a graphical interface, the official example uses Xvfb:

QT_QPA_PLATFORM=xcb xvfb-run -a drift --headless

Use this mode on a machine you control and keep the MCP port restricted to the host or to the trusted network. Without a proper graphical context, the service can edit the project, but capture and export may fail. For everyday editing use, the regular interface is simpler.

Export: do not skip this list

• Check that the beginning and end of the video have no black frames or cut audio.
• Read the subtitles at normal speed and on a small screen.
• Listen with headphones: the voice should remain clear when the music comes in.
• Export a short sample before the long file if you're testing a new chain of effects or GPU.
• Use the MP4 quality presets when you want speed; adjust advanced options only if you know which problem you're solving.
• Save the project and the source media. Drift can package a project with its media, preventing broken paths from ruining a future edit.

Common problems and quick solutions

“Whisper model not found” or subtitles don't start: open Settings → Extra packs and install the package Auto Captions. For other AI features, also install the AI Engine appropriate for the hardware.

The agent doesn't connect to the MCP: confirm that Agent access is enabled in the current Drift window. It starts disabled every time it opens. If using HTTP, update the token in the client with the freshly copied snippet.

The project opens, but headless capture or export fails: check the OpenGL context. On a host without a desktop, follow the example with QT_QPA_PLATFORM=xcb xvfb-run -a from the MCP documentation.

The caption is technically correct, but hard to watch: shorten the length of each line, pick a color that contrasts with the video, and adjust the entry timing. Captions are part of the rhythm, not a layer added at the end.

Conclusion

Drift is a promising choice for anyone who wants to edit video on Ubuntu with modern features without leaving the Linux environment. Install it via Flatpak for the most direct path, add the AI packages only when you need them, always review the captions, and use MCP with caution to automate the repetitive parts. Start with a short project: a few clean cuts, intelligible audio, reviewed captions, and a test export teach more than dozens of effects applied all at once.

Official sources: Drift on GitHub · latest release · MCP guide · official addons · Flatpak documentation