I’m sorry, Kyle, but despite the goodwill and positivity you consistently show on the forum, I find Topaz’s innovations over the past year (October 2025–October 2026) rather underwhelming.
Nyx XL: An extremely niche model. And when it comes to Nyx, scaling things up often means scaling up the hallucinations along with them.
Proteus Natural: A welcome improvement over Proteus, certainly, but we’re still dealing with a non-diffusion model. One can reasonably assume it’s largely a further training iteration of its predecessor. Hardly a revolution.
Hyperion 2 & 2.5: HDR manipulation is inherently subjective, and the capabilities of these models overlap considerably. Worse still, you can already achieve remarkably convincing results with Hyperion 1 by applying scene-by-scene dynamic contrast and brightness adjustments using free third-party tools.
Like many others in the community, I believe we’re long overdue for a model that genuinely breaks new ground.
We’ve grown accustomed to seeing AI models dramatically outperform their predecessors, sometimes delivering twice the performance or quality within just six months. Against that backdrop, Topaz’s apparent stagnation is becoming increasingly difficult to ignore.
We can no longer afford to claim that a handful of patches take time, in an era when tens of thousands of lines of code can be reviewed and debugged in just few hours.
Unfortunately, in my case, I’ve stopped waiting and started building my own solutions, even if that means spending twice as much as the monthly subscription.
At least I get to decide where the innovation goes.
If the scene detection option is not checked, Topaz processes the correct number of frames. That’s why it’s hard to understand that something so simple to fix wasn’t fixed immediately after I reported the bug. And now, for several versions, enabling this option is completely useless.
I definitely don’t want to waste hours on dividing the movie into dozens of scenes and then stiching them back.
Same here. Computer still completely freezing and requiring a force restart when trying to use Aion frame interpolation. Incredibly frustrating, since Aion is by far the best frame rate increase model.
What has been frustrating since old TVAI V5 up to now is the bugs that keep getting carried over from one version to the next, and with every new release, even more seem to be added. Ok some gets fixed finally, some not really final and new bugs were “released”. At this point, you almost hesitate to install an update because you never know what new problems you might run into.
Maybe in the past it was due to a lack of manpower or lack of willingness? But times have changed. With today’s AI tools, you could finally perform a thorough, systematic analysis of your entire codebase.
When I paste my logs into ChatGPT, it immediately identifies dozens of potential errors and issues. Some may be harmless, but others are serious and could even lead to crashes.
You have the complete source code, put it into Claude and fix the bugs once and for all
Or more advanced: Topaz could deploy AI agents that autonomously generate Topaz Video outputs using all available models, across all codec formats, fps color spaces etc, and a wide range of resolutions. After each encode, the AI would automatically analyze the output errors (logs), identify issues, and fix it in the code before running the next test, tousand times until it’s bugfree and then no Customer here must upload any logs anymore. Crazy world we live in, isn’t it? But this is actually possible.
I saw someone asked if the Nvidia tensor unlock program will be helpful for SLP 2.6 in the old 1.7.1 topic. Since that topic is already locked, I will share my test result here. By using the NV UEFI Unlocker on my RTX 4070tis with Windows 10 LTSC, the speed increased from 0.74fps to .094fps. While I use MSI afterburner controls the power limitation which makes the power increased from 200W to 215W(without power control it increased 50W after GPU unlocked).
Anyway, anyone want to try it need to take own risk though technically it should not break anything.
I’ve been following this topic as well. In the “unlock” tests for the RTX 4070 Ti, there was an increase from 84.8 TFLOPS to 169.4 TFLOPS (a 100% gain) for FP16/FP32 operations. The SLP model is the same, yet you’re only seeing a 27% increase—that’s a significant discrepancy. Either this was only achieved in theory, or there are additional fine-tuning steps required. Unfortunately, since I don’t own that card myself, I’m unable to run the full tests.
I’ve noticed that the pace of development and meaningful updates seems to have slowed down considerably.
Is this because the technology is approaching its current limits, making further breakthroughs increasingly difficult?
Or has Adobe’s acquisition introduced internal organizational issues, shifting priorities, or other obstacles that are preventing the development team from focusing on innovation?
I’d genuinely like to know what’s happening behind the scenes and what we can realistically expect going forward.
It was me, and I tried it today as well. I use a 4070 Ti Super. 2x SLM upscale speed, 720p to 1440p, went from 7.1-7.3 seconds per frame to 6.1-6.3 seconds per frame. Shaved off exactly 1 second per frame. So what previously would have taken 48 hours to process would now take 41 hours. I expected the improvement to be a bit more, but this is still a meaningful improvement for long exports.
(Sigh…) I’m still having an issue where the stand-alone app is leaving behind weird duplicate files, named random series of numbers. I have to delete them manually every time I encode. And I would like to be able to import and batch more than 10 clips without the software crashing. I work w a high volume of media and frequently have the need to transcode hundreds of clips at a time. Having to break that down into batches of 25 clips is killing me. Finally the batch option does not work correctly even after I followed Topaz’s workflow instructions to the letter. I still have to go and assign my preset to every clip individually. Selecting them all and doing this yields an undesired result…the software replaced my preset settings w random settings. I have a current-gen box, specs below, so I know it isn’t hardware related. Why isn’t the app more stable by this point? CPU: Intel Core Ultra 7 270K Plus (LGA 1851, 24 Cores / 24 Threads)
The issue with the oddly-named duplicate files should be solved if you disable ‘Enable externally viewable exports’ in Preferences. This won’t affect your ability to view live output on the Export queue screen.
If temp files (long string of numbers file) remains after the processing has completed, it likely means the app doesn’t have permission to delete them. I’d recommend checking your folder permissions or trying to run the app as administrator.
As far as batch, can you reach out to help@topazlabs.com with a screen recording of your workflow when it does not work correctly? We’d like to see if we can help here.
We recommend loom, it’s free and easy to use. Once your Loom recording is ready, you’ll be able to get a link with the video, please reply back with the link
Feel free to use any other preferred method and send the video file attached to your email replace. If the file is too large to attach, you can upload it here: Submit File to Dropbox
I have processed multiple short videos over the last two days using SLP, RheaXL, and Proteus. So far, no crashes. (fingers crossed). MacOS 27.2 and Topaz 1.7.2 patch. I have had bluetooth enabled the entire time. I have tried changing sound output selections and have had no crashes. Changing the display preset no longer causes an immediate crash.
I’m not very familiar with the detailed background of the macOS or AMD issues, but even in a Windows environment, rolling back the Topaz app frequently fails to restore its previous behavior. I’m not entirely sure of the exact technical reason, but I suspect that changes are being made to core application files/system dependencies beyond just the version update itself. If anyone has more insight into this, I’d appreciate your thoughts.
To reliably avoid this issue, taking regular system images is the safest and most logical approach. Whether you need to take a system image with every single update depends on the release cadence of new versions, but these days Windows Updates actually occur more frequently. If you restore a system image and then install the version immediately prior to when the issue started, most problems are resolved. And if an issue should still persist, you can simply restore the system image once again.