Foreword: My friends have been telling me enthusiastically about a CAD program for Linux that’s on par with AutoCAD and uses AutoCAD’s standard file formats. According to its author, this program—OpenCAD Studio ( https://github.com/HakanSeven12/OpenCADStudio)—was developed primarily using AI (Claude). Silly question: Can AI be used for some development in CinGG? It's a silly question because I don't know anything about programming or AI, but I'm curious about it.
I think AI can absolutely be used for programming. The main question is still the quality of the code it produces. AI coding is developing incredibly fast. These days, even local models can already produce complete working applications. Of course, that does not mean the result is always good. The code still needs to be reviewed, tested, and maintained by someone who understands the project. I am not a programmer myself, so AI has become a real lifesaver for me. Best regards, Andrey пн, 21 сент. 2026 г. в 11:16, Andrea paz via Cin <cin@lists.cinelerra-gg.org
:
Foreword: My friends have been telling me enthusiastically about a CAD program for Linux that’s on par with AutoCAD and uses AutoCAD’s standard file formats. According to its author, this program—OpenCAD Studio ( https://github.com/HakanSeven12/OpenCADStudio)—was developed primarily using AI (Claude). Silly question: Can AI be used for some development in CinGG? It's a silly question because I don't know anything about programming or AI, but I'm curious about it. _______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
Op 21-09-2026 om 10:16 a.m. schreef Andrea paz via Cin:
Can AI be used for some development in CinGG? It's a silly question because I don't know anything about programming or AI, but I'm curious about it.
Yes, it can. But the first functional steps are typically to define features or algorithms that lack and build a prototype from that. For example, last week I have been trying to build a continuous timeline video editor that would allow to place timecoded and not-yet-timecoded content and get an aligned perspective for multicam material. I think it would be absolutely worthwhile to get this functionality into Cinelerra, and having some AI assistance on handling the custom widget toolkit Cinelerra uses can be a great assistant. -- Stefan
Hi guys, Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing. Cheers Sam Am 21.09.26 um 11:05 schrieb Stefan de Konink via Cin:
Op 21-09-2026 om 10:16 a.m. schreef Andrea paz via Cin:
Can AI be used for some development in CinGG? It's a silly question because I don't know anything about programming or AI, but I'm curious about it.
Yes, it can. But the first functional steps are typically to define features or algorithms that lack and build a prototype from that. For example, last week I have been trying to build a continuous timeline video editor that would allow to place timecoded and not-yet-timecoded content and get an aligned perspective for multicam material.
I think it would be absolutely worthwhile to get this functionality into Cinelerra, and having some AI assistance on handling the custom widget toolkit Cinelerra uses can be a great assistant.
_______________________________________________ Cin mailing list --cin@lists.cinelerra-gg.org To unsubscribe send an email tocin-leave@lists.cinelerra-gg.org
Here are more screenshots: Am 21.09.26 um 15:31 schrieb Sam via Cin:
Hi guys,
Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing.
Cheers
Sam
Am 21.09.26 um 11:05 schrieb Stefan de Konink via Cin:
Op 21-09-2026 om 10:16 a.m. schreef Andrea paz via Cin:
Can AI be used for some development in CinGG? It's a silly question because I don't know anything about programming or AI, but I'm curious about it.
Yes, it can. But the first functional steps are typically to define features or algorithms that lack and build a prototype from that. For example, last week I have been trying to build a continuous timeline video editor that would allow to place timecoded and not-yet-timecoded content and get an aligned perspective for multicam material.
I think it would be absolutely worthwhile to get this functionality into Cinelerra, and having some AI assistance on handling the custom widget toolkit Cinelerra uses can be a great assistant.
_______________________________________________ Cin mailing list --cin@lists.cinelerra-gg.org To unsubscribe send an email tocin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list --cin@lists.cinelerra-gg.org To unsubscribe send an email tocin-leave@lists.cinelerra-gg.org
That is INCREDIBLE!! Thanks so much for working on this, it must have been an incredible amount of work even to this point! Best, Glen On 2026-09-21 09:45, Sam via Cin wrote:
Here are more screenshots:
Am 21.09.26 um 15:31 schrieb Sam via Cin:
Hi guys,
Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing.
Cheers
Sam
Am 21.09.26 um 11:05 schrieb Stefan de Konink via Cin: Op 21-09-2026 om 10:16 a.m. schreef Andrea paz via Cin: Can AI be used for some development in CinGG? It's a silly question because I don't know anything about programming or AI, but I'm curious about it.
Yes, it can. But the first functional steps are typically to define features or algorithms that lack and build a prototype from that. For example, last week I have been trying to build a continuous timeline video editor that would allow to place timecoded and not-yet-timecoded content and get an aligned perspective for multicam material.
I think it would be absolutely worthwhile to get this functionality into Cinelerra, and having some AI assistance on handling the custom widget toolkit Cinelerra uses can be a great assistant.
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org _______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
--
Phyllis and I have been working on a new version of Cinelerra for several months now.
WOW!!! That's incredible news! Best regards, Andrey пн, 21 сент. 2026 г. в 16:45, Sam via Cin <cin@lists.cinelerra-gg.org>:
Here are more screenshots:
Am 21.09.26 um 15:31 schrieb Sam via Cin:
Hi guys,
Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing.
Cheers
Sam
Am 21.09.26 um 11:05 schrieb Stefan de Konink via Cin:
Op 21-09-2026 om 10:16 a.m. schreef Andrea paz via Cin:
Can AI be used for some development in CinGG? It's a silly question because I don't know anything about programming or AI, but I'm curious about it.
Yes, it can. But the first functional steps are typically to define features or algorithms that lack and build a prototype from that. For example, last week I have been trying to build a continuous timeline video editor that would allow to place timecoded and not-yet-timecoded content and get an aligned perspective for multicam material.
I think it would be absolutely worthwhile to get this functionality into Cinelerra, and having some AI assistance on handling the custom widget toolkit Cinelerra uses can be a great assistant.
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
It is incredible! Without Sam using AI, it would have taken years to get to QT6 which will remove a big hurdle in getting the flexibility for users to better set up Cinellera for their individual needs. FYI, the credit is all Sam's and I mostly just provided moral support and got to do a little testing with AppImages. I will continue supporting the original GG version as I can. On Mon, Sep 21, 2026, 07:58 Andrea paz via Cin <cin@lists.cinelerra-gg.org> wrote:
What a wonderful Surprise! Thanks, Sam and Phillys. I can try to build and test... _______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
A note on how this version came about, and what to expect from it This is an experimental build. Please do not use it for real projects. Most of this work was done with Claude Code. For me, for this kind of work on a large codebase, it has been the most productive tool by far. It started as an experiment, mostly for fun: I wanted to see how far one can get. Out of that grew the new version, and along the way I realised what this technology can actually do. The result makes no claim to be free of bugs. The goal was to develop and port quickly, and the version certainly still has many bugs and a lot of headroom. The existing version will continue to be maintained. The wider goal is faster development and, above all, making the work easier. We are a small team with hardly any developers. This technology lets us add things we have been missing for years, and that is why we use it. What is new The user interface now runs on Qt6 instead of the old X11 toolkit. The code base is still Cinelerra-GG 5.1; Qt6 replaces the X11 layer underneath, and the old code path is kept in the tree. Concretely: - Native Qt6 main panels: timeline with waveforms and thumbnails, viewer, compositor with tool overlays, resource panel with bins, and a dockable Deliver panel for rendering. - Redesigned dialogs: Preferences, New Project, Set Format, Proxy, Transcode, Export Project, Export EDL, Batch Render, Create BD and Create DVD, Record, Backup. - Native plugin GUIs with an effect inspector for the first 13 effects, plus in/out markers and labels on the ruler. Known issues - Loading long files with an audio track can crash the program. Reproduced this week, a fix is in progress. - Rendering is software only, OpenGL is disabled for now. - Seven plugins are not built yet. - Translations are incomplete, several of the new panels are English only. - Loading a file from the command line crashes; use File → Load. Getting it I will create a new AppImage in the course of this week. What helps me most Try the new dialogs, the plugin panels and the effect inspector, and tell me what is missing or in the way compared to the old windows. That is the feedback I can act on fastest. Known issues from the list above need no report. For bugs, please include the build date from the AppImage name, the steps that lead to the problem, and the file /tmp/cinelerra_<pid>.dmp that the program writes when it crashes. How the AI work is done A word on process, because "AI-written" rightly raises the question of who checks the code. The recent changes follow a fixed routine: a written plan, implementation, review by a second AI instance against the plan, unit tests, headless runtime tests against the running program, and a screenshot of every new panel before anything counts as done. On top of that comes extensive manual testing, and the concepts themselves are refined together with the AI, with a human in the loop at every step. I will be honest that the early phases of the experiment ran with less rigour than that. Nothing is merged into the existing version by this route. One last word This is an experiment, and I mean that literally. If you feel uneasy about using software written with AI, that is a perfectly good reason to stay with the existing version, which was built the conventional way and continues to be maintained. This build is here for those who want to see where this goes. Sam Am 21.09.26 um 16:31 schrieb Phyllis Smith via Cin:
It is incredible! Without Sam using AI, it would have taken years to get to QT6 which will remove a big hurdle in getting the flexibility for users to better set up Cinellera for their individual needs.
FYI, the credit is all Sam's and I mostly just provided moral support and got to do a little testing with AppImages. I will continue supporting the original GG version as I can.
On Mon, Sep 21, 2026, 07:58 Andrea paz via Cin <cin@lists.cinelerra-gg.org> wrote:
What a wonderful Surprise! Thanks, Sam and Phillys. I can try to build and test... _______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list --cin@lists.cinelerra-gg.org To unsubscribe send an email tocin-leave@lists.cinelerra-gg.org
пн, 21 сент. 2026 г., 17:59 Sam via Cin <cin@lists.cinelerra-gg.org>:
A note on how this version came about, and what to expect from it
This is an experimental build. Please do not use it for real projects.
Most of this work was done with Claude Code. For me, for this kind of work on a large codebase, it has been the most productive tool by far. It started as an experiment, mostly for fun: I wanted to see how far one can get. Out of that grew the new version, and along the way I realised what this technology can actually do.
The result makes no claim to be free of bugs. The goal was to develop and port quickly, and the version certainly still has many bugs and a lot of headroom. The existing version will continue to be maintained.
The wider goal is faster development and, above all, making the work easier. We are a small team with hardly any developers. This technology lets us add things we have been missing for years, and that is why we use it.
What is new
The user interface now runs on Qt6 instead of the old X11 toolkit. The code base is still Cinelerra-GG 5.1; Qt6 replaces the X11 layer underneath, and the old code path is kept in the tree. Concretely:
- Native Qt6 main panels: timeline with waveforms and thumbnails, viewer, compositor with tool overlays, resource panel with bins, and a dockable Deliver panel for rendering. - Redesigned dialogs: Preferences, New Project, Set Format, Proxy, Transcode, Export Project, Export EDL, Batch Render, Create BD and Create DVD, Record, Backup. - Native plugin GUIs with an effect inspector for the first 13 effects, plus in/out markers and labels on the ruler.
Known issues
- Loading long files with an audio track can crash the program. Reproduced this week, a fix is in progress. - Rendering is software only, OpenGL is disabled for now. - Seven plugins are not built yet. - Translations are incomplete, several of the new panels are English only. - Loading a file from the command line crashes; use File → Load.
Getting it
I will create a new AppImage in the course of this week.
What helps me most
Try the new dialogs, the plugin panels and the effect inspector, and tell me what is missing or in the way compared to the old windows. That is the feedback I can act on fastest. Known issues from the list above need no report.
For bugs, please include the build date from the AppImage name, the steps that lead to the problem, and the file /tmp/cinelerra_<pid>.dmp that the program writes when it crashes.
How the AI work is done
A word on process, because "AI-written" rightly raises the question of who checks the code. The recent changes follow a fixed routine: a written plan, implementation, review by a second AI instance against the plan, unit tests, headless runtime tests against the running program, and a screenshot of every new panel before anything counts as done. On top of that comes extensive manual testing, and the concepts themselves are refined together with the AI, with a human in the loop at every step. I will be honest that the early phases of the experiment ran with less rigour than that. Nothing is merged into the existing version by this route.
One last word
This is an experiment, and I mean that literally. If you feel uneasy about using software written with AI, that is a perfectly good reason to stay with the existing version, which was built the conventional way and continues to be maintained. This build is here for those who want to see where this goes.
Sam Am 21.09.26 um 16:31 schrieb Phyllis Smith via Cin:
It is incredible! Without Sam using AI, it would have taken years to get to QT6 which will remove a big hurdle in getting the flexibility for users to better set up Cinellera for their individual needs.
FYI, the credit is all Sam's and I mostly just provided moral support and got to do a little testing with AppImages. I will continue supporting the original GG version as I can.
Thanks for this (building qt6 on unsupported OS/architecture is a big pain, so I like to have something less modern for my odd usecases). Thanks Sam, this is indeed Big Surprise - I see a lot of emulators or even FPGA implementations supercharged with "AI" tools. My other problem with AI (apart from resource use and how it shoveled into everything) is who control it - as those companies increasingly look for some profitability "free" options will dry up, leaving users with expanded codebase uneasy to navigate/fix manually. It was "a bit" of problem with original Cinelerra, too, I must admit. I am relatively happy that I am not only person poking those C files now. For now I live mostly on Macintosh forums, testing new software for PPC machines or emulators for/on said machines. There are some talks about upcoming harder times (due to all those wars etc), so stay safe from preventable dangers!
On Mon, Sep 21, 2026, 07:58 Andrea paz via Cin <cin@lists.cinelerra-gg.org> wrote:
What a wonderful Surprise! Thanks, Sam and Phillys. I can try to build and test... _______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
Am 21.09.26 um 20:00 schrieb Andrew Randrianasulu via Cin:
Thanks for this (building qt6 on unsupported OS/architecture is a big pain, so I like to have something less modern for my odd usecases).
Thanks Sam, this is indeed Big Surprise - I see a lot of emulators or even FPGA implementations supercharged with "AI" tools.
My other problem with AI (apart from resource use and how it shoveled into everything) is who control it - as those companies increasingly look for some profitability "free" options will dry up, leaving users with expanded codebase uneasy to navigate/fix manually. It was "a bit" of problem with original Cinelerra, too, I must admit.
I am relatively happy that I am not only person poking those C files now. For now I live mostly on Macintosh forums, testing new software for PPC machines or emulators for/on said machines.
There are some talks about upcoming harder times (due to all those wars etc), so stay safe from preventable dangers!
Your concerns are absolutely valid, and I ask myself the same question almost every day. I’m fully aware that many of the tools currently leading the industry are heavily subsidized. Claude Code, for example, currently costs around $200 per month, and even then I sometimes need more than the weekly quota allows. I’ve built an API cost calculator into my workflow to estimate what the same usage would have cost through direct API access. At this point, the theoretical cost is already in the tens of thousands of dollars. In the long run, I don’t think this level of subsidization is sustainable for providers such as Anthropic or OpenAI. They will either have to raise prices, improve efficiency significantly, or both, something we are already seeing from Chinese AI providers, with increasingly capable and efficient models such as the latest DeepSeek 4.1 Flash model. That is also why I’m actively testing alternative coding models and frameworks, including offerings from providers such as Qwen, Kimi, GLM, and DeepSeek. I’ve always been a strong supporter of open source (open weights); it’s simply in my DNA. If these tools give me the opportunity to improve Cinelerra, I’m going to use them. I use many of them through OpenRouter, where I can select specific data-center providers and, where available, use zero-data-retention policies. I have to say that open-weight models have reached a level of quality that I would not have expected only a few months ago. As a result, I’m gradually shifting more and more of my workload toward them. I’m also using OpenCode much more extensively and expanding it with the functionality I need for my own development workflow. One of the most interesting developments is that some AI companies have realized that scaling cannot depend indefinitely on simply adding more and more hardware. There is enormous value in making models smaller, faster, and more efficient while maintaining, or even improving, their capabilities. For highly complex coding tasks, models from Anthropic and OpenAI are still among the strongest options available in my experience. But I’m no longer dependent on them alone, I already have a Plan B, and that Plan B is improving rapidly. As for Cinelerra-GG Qt6 itself, I’m continuously reviewing and optimizing the code. I often run multiple independent code reviews on the same sections to identify problems, improve structure, and catch things that might have been missed in an earlier pass. I know the code will not always be perfect, especially during an active modernization effort. But at some point, you have to start. I could wait years and hope that new developers eventually join the project, or I can start doing the work now, improve it step by step, and learn along the way. I chose to start. It’s an experiment. It’s worth a try, in my opinion.
Hi, This is indeed a new frontier and there is great potential for both good and for harm. I will agree with Sam that for a person tasked with helping to maintain something like Cinelerra-GG with little programming skills then AI opens up some great possibilities. I am also not a programmer but AI has allowed me to improve AV Linux in various ways but I always try to remain with programming languages that I am at least somewhat familiar with so I can read and audit what it is doing while I'm testing. I think it is always valuable to have the AI also create debugging, testing and verification tools for your projects as well as the main development. When this gets to a public beta testing phase please let us know on the list, I would be interested in testing if time from other things allows. Thanks very much to Sam, Phyllis and all other contributors for your tireless work, it is very much appreciated! Best, Glen On 2026-09-21 15:04, Sam via Cin wrote:
Am 21.09.26 um 20:00 schrieb Andrew Randrianasulu via Cin:
Thanks for this (building qt6 on unsupported OS/architecture is a big pain, so I like to have something less modern for my odd usecases).
Thanks Sam, this is indeed Big Surprise - I see a lot of emulators or even FPGA implementations supercharged with "AI" tools.
My other problem with AI (apart from resource use and how it shoveled into everything) is who control it - as those companies increasingly look for some profitability "free" options will dry up, leaving users with expanded codebase uneasy to navigate/fix manually. It was "a bit" of problem with original Cinelerra, too, I must admit.
I am relatively happy that I am not only person poking those C files now. For now I live mostly on Macintosh forums, testing new software for PPC machines or emulators for/on said machines.
There are some talks about upcoming harder times (due to all those wars etc), so stay safe from preventable dangers!
Your concerns are absolutely valid, and I ask myself the same question almost every day.
I’m fully aware that many of the tools currently leading the industry are heavily subsidized. Claude Code, for example, currently costs around $200 per month, and even then I sometimes need more than the weekly quota allows.
I’ve built an API cost calculator into my workflow to estimate what the same usage would have cost through direct API access. At this point, the theoretical cost is already in the tens of thousands of dollars. In the long run, I don’t think this level of subsidization is sustainable for providers such as Anthropic or OpenAI. They will either have to raise prices, improve efficiency significantly, or both, something we are already seeing from Chinese AI providers, with increasingly capable and efficient models such as the latest DeepSeek 4.1 Flash model.
That is also why I’m actively testing alternative coding models and frameworks, including offerings from providers such as Qwen, Kimi, GLM, and DeepSeek. I’ve always been a strong supporter of open source (open weights); it’s simply in my DNA. If these tools give me the opportunity to improve Cinelerra, I’m going to use them. I use many of them through OpenRouter, where I can select specific data-center providers and, where available, use zero-data-retention policies.
I have to say that open-weight models have reached a level of quality that I would not have expected only a few months ago. As a result, I’m gradually shifting more and more of my workload toward them.
I’m also using OpenCode much more extensively and expanding it with the functionality I need for my own development workflow. One of the most interesting developments is that some AI companies have realized that scaling cannot depend indefinitely on simply adding more and more hardware. There is enormous value in making models smaller, faster, and more efficient while maintaining, or even improving, their capabilities.
For highly complex coding tasks, models from Anthropic and OpenAI are still among the strongest options available in my experience. But I’m no longer dependent on them alone, I already have a Plan B, and that Plan B is improving rapidly.
As for Cinelerra-GG Qt6 itself, I’m continuously reviewing and optimizing the code. I often run multiple independent code reviews on the same sections to identify problems, improve structure, and catch things that might have been missed in an earlier pass.
I know the code will not always be perfect, especially during an active modernization effort. But at some point, you have to start.
I could wait years and hope that new developers eventually join the project, or I can start doing the work now, improve it step by step, and learn along the way.
I chose to start. It’s an experiment. It’s worth a try, in my opinion.
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
--
Sorry for private mail. I was thinking along similar lines, too. That is, a way to sync the audio and video from all sources in batches during the pre-editing phase (in the Resources window) rather than on the timeline, as we do now. This idea has been on my mind ever since I read, on Mantis BT 448 (and comment 0003983), a request from a professional; he said that CinGG would be fully usable in a professional production environment if it had this feature. The idea of importing hundreds of “Dailies” into the timeline and then syncing them semi-manually was out of the question— it needed to be done automatically during the pre-editing phase. I've written down a lot of notes on how to do it, but I don't know how to put them into practice (and I think some of the steps are too complicated). All the major professional programs have this feature, for example, DaVinci Resolve: https://blog.frame.io/2017/05/15/sync-clips-in-davinci-resolve/
Hi Andrew My other problem with AI (apart from resource use and how it shoveled into
everything) is who control it
This is indeed a problem if you use the Claude code or Codex. However, if you have a high-end gaming graphics card, such as an Nvidia 3090, you can run local AI models such as the Qwen-3.8-27B model. I have had very good results with it. Best regards, Andrey
Hi Andrey, I bought a new PC when memory was still cheaper, and without graphics card, but AMD 8700G CPU and 64 G RAM, I can use several models purely local. gpt-oss gives me 15-20 tokens per second, converting sources. GLM is Ok too but slower, but I found the Qwen models too chatty. But there might be tuning setting that improve that. I use a terminal interface using AMD's lemonade, not yet a local agent like pi. MatN On Tue, 22 Sep 2026 09:50:59 +0300 Андрей Спицын via Cin <cin@lists.cinelerra-gg.org> wrote:
Hi Andrew
My other problem with AI (apart from resource use and how it shoveled into
everything) is who control it
This is indeed a problem if you use the Claude code or Codex. However, if you have a high-end gaming graphics card, such as an Nvidia 3090, you can run local AI models such as the Qwen-3.8-27B model. I have had very good results with it.
Best regards, Andrey
gpt-oss gives me 15-20 tokens per second
That's a good result for a test that only uses the CPU. вт, 22 сент. 2026 г. в 13:07, Mat via Cin <cin@lists.cinelerra-gg.org>:
Hi Andrey,
I bought a new PC when memory was still cheaper, and without graphics card, but AMD 8700G CPU and 64 G RAM, I can use several models purely local. gpt-oss gives me 15-20 tokens per second, converting sources. GLM is Ok too but slower, but I found the Qwen models too chatty. But there might be tuning setting that improve that. I use a terminal interface using AMD's lemonade, not yet a local agent like pi.
MatN
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
Fantastic Sam! Mat On Mon, 21 Sep 2026 13:31:52 +0000 Sam via Cin <cin@lists.cinelerra-gg.org> wrote:
Hi guys,
Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing.
Cheers
Sam
A huge job, Sam (and Phyllis)! You are Great! IgorBeg Il 21/09/2026 15:31, Sam via Cin ha scritto:
Hi guys,
Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing.
Cheers
Sam
Thank you all for the kind words. I don't want to keep you in suspense any longer—I'm sure you're eager to try out the new version—but I still need about one to two days before I can provide you with an AppImage to test the new Qt6 version. I still have a few bugs to fix before I can release this version for you to test. It won't be long now. Sam Am 24.09.26 um 10:09 schrieb Igor BEGHETTO via Cin:
A huge job, Sam (and Phyllis)! You are Great!
IgorBeg
Il 21/09/2026 15:31, Sam via Cin ha scritto:
Hi guys,
Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing.
Cheers
Sam
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
Sam and Phyllis, Thank you for driving Cin-gg forward! As you know, I'm not a programmer myself, but in my eyes using Qt6 is a great idea as it has been for several other video-related applications on Linux. I'm curious about how the AppImage looks and functions—whether it's hard-coded or adapts to the system's Wayland environment, and possibly which packages are needed to run it on GNOME. In the meantime, I have reviewed the history of Qt and Trolltech; how it started on a park bench in Trondheim in 1990 and evolved through Nokia and Digia ;) Terje J. H On 25.09.2026 01:47, Sam via Cin wrote:
Thank you all for the kind words.
I don't want to keep you in suspense any longer—I'm sure you're eager to try out the new version—but I still need about one to two days before I can provide you with an AppImage to test the new Qt6 version. I still have a few bugs to fix before I can release this version for you to test. It won't be long now.
Sam
Am 24.09.26 um 10:09 schrieb Igor BEGHETTO via Cin:
A huge job, Sam (and Phyllis)! You are Great!
IgorBeg
Il 21/09/2026 15:31, Sam via Cin ha scritto:
Hi guys,
Phyllis and I have been working on a new version of Cinelerra for several months now. Most of it has already been ported to Qt6. This means Cinelerra runs as a native Qt6 app. So I’ve now implemented everything that Cinelerra couldn’t do before, such as single-window handling, etc., and I’ve already built in many AI features, like speech-to-text, and set things up so that videos can be automatically edited in the future. This was actually supposed to be a big surprise. I’d like to invite a group of people to participate in alpha testing.
Cheers
Sam
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org
_______________________________________________ Cin mailing list --cin@lists.cinelerra-gg.org To unsubscribe send an email tocin-leave@lists.cinelerra-gg.org
Hi everyone, Here is a first AppImage for anyone who wants a look at where this is going: https://www.swisstransfer.com/dl/01a0f73b-69e8-709a-a57c-3e7c3d8e890b File: CinGG-qt6-20261001-x86_64.AppImage (222 MB, Linux x86_64) Please read this first: - This is an alpha. It is not even a beta yet. - There are known bugs, and much is not done yet. Some windows are native Qt6, others are still the old ones running inside Qt6. - Please do not use it for real projects. - It uses the same settings folder as your normal Cinelerra (~/.bcast5). Make a copy of that folder before you start it. To start it: chmod +x CinGG-qt6-20261001-x86_64.AppImage ./CinGG-qt6-20261001-x86_64.AppImage Reports are welcome, best with a screenshot and your distribution. Sam
participants (10)
-
Andrea paz -
Andrew Randrianasulu -
Igor BEGHETTO -
info＠bandshed.net -
Mat -
Phyllis Smith -
Sam -
Stefan de Konink -
Terje J. Hanssen -
Андрей Спицын