Hi Rob, thank you for testing again and for the precise details. They were enough to find the causes. Here is where each point stands. The fixes are done and will be in the next alpha AppImage. The one from 27 October has none of them yet. Floating windows that move only once. I found one cause. A floating panel had a 12 px resize border along its top edge, and its title bar is 29 px high. Grabbing the upper part of the title started a resize instead of a move. The border at the top is now 4 px, the corners keep 12 px. I reproduced it under openbox 3.6.1: before, a drag 8 px below the top edge resized the window, now it moves it. "Only once" itself I could not reproduce, three drags in a row moved the window in both versions. I also cannot test TwinView or separate X screens here. If it still happens with the next AppImage, please tell me where on the title bar you grab it. Independent windows. There is a new setting under Settings > Preferences
Appearance > Floating panels. "Stay above the main window" is the default and keeps the current behaviour. With "Separate windows" every floating panel is a window of its own, which the main window can cover, for example on your second monitor. The setting applies at once. In this mode the window manager moves the panel like any other window, and letting it go over the main window no longer docks it back, so your layout stays as it is. To dock a panel again, double-click its title.
Popups over the main window. Done. Every dialog now has a slightly lighter surface than the panels behind it and draws a thin frame of its own, so it stands out even without a compositor shadow. I checked the Preferences and ten other dialogs under openbox with the dark Onyx theme. Audio tracks drawn only half way. Found and fixed, and your description was the key. On first load the timeline drew the waveform from the audio index while the index was still being built, and kept that half-finished picture. Nothing told the timeline when the index was complete. That is why the second load was fine. Now a waveform drawn from an unfinished index is only provisional and is drawn again when the index is done. With the index deleted, the old version drew 82 to 93 of 578 columns of a test clip on first load, the new one all 578. AI. Fair enough, no debate from my side. Just so you know where things stand: the Qt6 port is built with an AI coding assistant, Claude Code, and every commit says so. With a team as small as ours the port would not have been possible otherwise. I see it as a tool, as Linus Torvalds does, who now uses Claude Code for kernel work too (https://www.phoronix.com/news/Linus-Torvalds-Claude-Code). I completely understand if you want nothing to do with it. The traditional version is still maintained, and fixes in the shared engine benefit both versions. Inside Cinelerra nothing of it is forced on you. The remote control is off unless you switch it on yourself (Settings > Preferences > Interface
Remote control), and it is a local socket, not a network port. Transcription uses a Whisper model, by default locally on your machine, and only runs when you choose Transcribe... yourself. Only its optional cloud engine sends audio to a service, and only with your own API key.
Build. Thanks, that helps. You built the traditional repository (goodguy/cinelerra.git at 3a30a189), not the Qt6 one, and only make install failed. Without its output I cannot say why. If you try again, make install 2>&1 | tee install.log keeps the output, and the last 50 lines are enough. One question already: did make install run as root or with sudo? Once the next AppImage is out, I would be glad to hear whether the separate windows work for your two monitors. Sam Am 28.10.26 um 23:23 schrieb Rob Prowel via Cin:
On 10/28/26 1:26 PM, Sam via Cin wrote:
Hi Rob,
welcome back, and thank you for this careful mail. Your concerns are fair, so let me go through them one by one.
What you see today is an alpha. Not everything is ported yet, so some things you rely on may still be missing or behave differently. That is exactly where reports like yours help us.
using appimage (CinGG-qt6-alpha-20261027-anylinux-x86_64.AppImage)
Windows on two monitors. The timeline is the main window. Every other panel (Viewer, Compositor, Resources, Effect Inspector and so on) is a dock that you can tear off into its own top-level window: drag its title bar out of the main window, or double-click it. Then put it on your second monitor. Window > Save Layout... keeps the arrangement, and Ctrl+Shift+F1 to F6 bring a saved layout back.
As long as the Qt MDI interface is used and floating windows are "children" of the timeline window I think it will be problematic for me. Upon testing I can only "move" a floating window once (when I pull it out of the timeline window) then afterwards it only responds to resize events. Also, child floating windows are not allowed to be "behind" the main timeline window which makes it too easy to accidentally merge the floater back into the timeline panel and screw up the timeline layout panel. I really do prefer separate/independent X11 windows for each function.
using openbox, xorg, and nvidia-570.207 driver. in twinview config when I'm editing content, and xaphod config any other time
Pretty versus power. Agreed, and that is why the keys came first. Another step in that direction is in the works: a remote control through which scripts and AI assistants can drive Cinelerra, every command and every effect, each step undoable. It is not in an AppImage yet.
You had me until you mentioned AI. Long passionate anti-AI rant, so I'll digress.
re - color scheme: just bring up the global configuration screen over top of the main app MDI panel and it's pretty obvious that something needs to be done to make the popup more distinguishable from the background.
Audio tracks drawn only half way. To reproduce it I need a few details: was it the traditional AppImage from August or the Qt6 alpha (and which date), what kind of clip (container, codec, length, number of audio channels), and at which point the waveform stops.
It happens with pretty much every media clip I load, upon first load. Once the thumbs exist in .bcast5 directory the clip loads properly upon subequent tries. If I delete the indexes and thumbs then it burps again. I've tried with both you-tube content (av1/opus/matroska) and with h264/aac content.
Latest august (traditional UI) cingg does not suffer the problem, just the Qt appimage I tried listed above.
Build. Please tell me which repository you cloned (the traditional git or the Qt6 one), the commands you ran, your distribution and compiler, and the last 50 lines of the output. If it was the Qt6 repository built with CMake: CMake has a recursion limit that the Qt6 deployment helpers can hit, and running cmake with -DCMAKE_MAXIMUM_RECURSION_DEPTH=5000 is worth a try. origin is: https://git.cinelerra-gg.org/goodguy/cinelerra.git
last pulled commit (HEAD) was: commit 3a30a189bfda351205615618fc2b206901bb6934 (HEAD -> master, origin/master, origin/HEAD) Author: Good Guy <good1.2guy@gmail.com> Date: Tue Oct 20 10:59:07 2026 -0600
Credit FFmpeg team and Andrew for patches for FFmpeg 9.0.2
generated standard configure program and ran with all defaults uses traditional gmake debian13: gcc 14.2.0-19
no output available because I deleted all artifacts and "git clean" on local repo after failure. build ran correctly. It was the install to /usr/local that barfed.
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org