Hi everyone,

a new Qt6 alpha is online, built on 2026-10-29. It brings zoom grips on the timeline scrollbars, hints in the footer, a first version of remote control for AI assistants, and the fixes for what you reported since the alpha of 2026-10-27. Thank you, Terje, Andrea, Rob and Andrew, for testing and for the detailed reports.

Download (sorted by date, the newest on top): download.cinelerra-gg.org

The source is public: git.cinelerra-gg.org, branch main, 59 commits since the alpha of 2026-10-27.

1. Zoom grips on the timeline scrollbars

Premiere, Kdenlive and VEGAS let you zoom right on the timeline scrollbar, and Cinelerra can do that now, too. Each end of the scrollbar thumb has a round grip. Drag a grip and the view zooms smoothly, while the other end stays where it is. When you let go, the grip stays under the mouse. Drag the middle to scroll as before, and a double-click shows the whole project.

The vertical scrollbar beside the tracks has the same grips. There a grip changes the height of all tracks at once, audio and video alike, and a double-click fits all tracks into the window.

The footer now tells you what the mouse can do: over a clip, over the time ruler, over a scrollbar. While you drag a clip, it says what the drop will do, move it, or with Shift, overwrite. If you don't want the hints, a right click on the footer turns them off. The E key toggles the editing mode again.

Video, 1:32, English with subtitles: zoom grips and footer hints
Video, 1:32, English with subtitles: zoom grips and footer hints

2. Remote control for AI assistants (MCP), an early version

Professional editors are adding a way for AI assistants to work in the edit. DaVinci Resolve Studio has a built-in MCP server for this. Adobe has an official connector for quick video jobs in the cloud, while control of the Premiere Pro or Final Cut Pro timeline still takes add-ons from the community. Cinelerra now has its own MCP server, built into the program, free and open source.

It is off by default. Turn it on in Settings, Preferences: search for remote control and tick Allow AI tools and scripts to control Cinelerra (local socket). Cinelerra then opens a socket in a private directory, only for your user. Every call is written to api.log in the configuration folder, and every change is one undo step. Then register Cinelerra with your assistant. For Claude Code that is one line:

claude mcp add cinelerra -- /path/to/CinGG-qt6-alpha-20261029-x86_64.AppImage --mcp

Codex works the same way, with an entry in ~/.codex/config.toml. The README in the repository has both setups.

The video shows a very simple example, just to see what it can do: load a clip, cut out two ranges, add Brightness/Contrast and look at the frames on both sides of the cut. The terminal on the left is re-enacted, the Cinelerra on the right is the real program, driven through the socket with the same calls a real session made.

The assistant can do a lot more:

Prepared, but not available yet: rendering, inserting, moving and trimming clips, and automatic editing. Scene detection comes next. This is the first step, and more will follow.

Video, 2:46, English with subtitles: remote control for AI assistants
Video, 2:46, English with subtitles: remote control for AI assistants

The film in both videos is Tears of Steel, (CC) Blender Foundation, mango.blender.org, licensed CC BY 3.0. All videos with .srt subtitles: qt6-alpha/videos.

3. Fixes from your reports

Terje (GNOME):

Andrea:

Rob (Debian 13, openbox, two screens):

Found in our own tests:

Andrew, Terje: VAAPI with ffmpeg 9 is still open. I will come back to you with a test build.

How this is made

As before, the code is written with Claude. What I rely on is testing: automated tests that run the real program, code reviews of each larger change, and my own check of every feature before it is pushed. That does not make it free of bugs, so please keep the reports coming: bug.cinelerra-gg.org or this list.

Sam