The status is that now there is CUDA/Metal/CPU support. The blender addon is getting closer to an early final version. The solver is getting a lot of improvements and bug fixes so it’s also getting stable. There are some things that I’d like to implement for the first version that are not yet done but for now it’s quite nice already. 
Looking good! Don’t be afraid to share an early beta with a select group of testers, many add-ons are released with very strong algorithms but UI’s that are more… eh… dev-oriented than creator-oriented 
A clear and memorable UI that’s understandable without lots of training or manual reading makes all the difference in the add-ons popularity after launch.
Thanks. That’s actually the plan. So hopefully when it’s released to the public it’ll be easy but powerful
Awesome! I’ll definitely try this out when you release it. 
Will you include the Linux version in the Open Beta?
I’ll try to have Linux support
You’re doing an amazing job. If you need testers, I’d be happy to be one
And my axe!
A small side project
Are you smoothing out the effects of denoising, or did you write a better temporal denoiser? Based on extra motion render pass or AI? Kinda like Neat Video (no extra pass) or TurboTools (republish noisy render)?
In the first pass I’m first using custom temporal denoiser to overall reduce the noise and preserve details using multiple frames. Then I’m applying the oidn denoiser. In the second pass I’m smoothing out the denoiser results so it’s more stable. The algorithm works on all direct and indirect channels separately so later I’m compositing to have cleaner passes. For now it was a small proof of concept so nothing special but in the future I’m planning to add more sophisticated algorithms and maybe my own ai denoiser (though oidn seems to be very good)
This video shows some first results and comparisons of my own temporal video denoiser. For now it’s just a very early prototype but on the above demo scenes it shows more details and quality than even neat video, though take that with a grain of salt as I’ve done only limited testing.
You have my respect for writing a denoiser sir! I run Neat, love their parallellism on the GPU, but am often confused by the training phase if I haven’t used it in a while. It’s a lot better and quicker than TurboTools imho.
It’s not easy to judge your results, but it looks impressive. Youtube compression introduces quite a lot of new compression artefacts… You’ll have to WeTransfer us the proRes Would 3D noise be similar to this video noise?
There should be a onedrive link to less compressed video in the description, though I’ll do better ones in the future. I’ve just added also the option to save the removed noise for later regrain of the footage for VFX needs in the compositing. Also from my limited CGI denoising it seems to work pretty well too, especially when the noise seed was set to change in each frame so there is more data. I’m also planning to combine the video denoiser algorithm with the EXR denoiser I’ve made for more control. So stay tuned for more results. Overall more testing and work is needed but the initial results seems to be okey.
Shut up and take my money! I paid 62€ for Neat (discounted) and 38$ for TurboTools, I’d pay somewhere in that ballpark for a blender native temporal denoiser. If it works in Mac and has decent parallellization
Link to ProRes file:
https://1drv.ms/v/c/d3137fc45fb8a229/IQCAkQJ9vbbwR6-pN3HhKdSsAds77enb7cwXeOpJYTQsmsk?e=p4RR8T
A simple comparison of my own temporal multi layer EXR denoiser with blender’s OIDN and no denoising. The advantages that I see is that it has very little flickering and works overall quite well while not being too slow. The other thing is that it denoises all EXR layers which is a benefit when you want to do a composit later in Fusion or other programs as the blender default setup only gives a combined denoised layer. It’s still a very early stage but it seems to be working surprisingly well.
ProRes file link:
https://1drv.ms/v/c/d3137fc45fb8a229/IQCTp48BkZ9aSayXe47NYrVTARumoHTnCNmmi8i-Qqs3i04?e=8wy3ZZ
So after a lot of time and upgrades I present you another trial of my video denoiser. Still not perfect but the results seems to be even better than last time and quite useful. Especially considering that neat video has a noise pattern learning process and mine is almost a one click solution (of course you can play with settings but it’s far simpler).
Great work! I often struggle with the Neat Video learning process, especially when my image doesn’t have an ideal flat area for the training sample… I often reuse a generic profile.
Could your denoiser be a part of Blender in terms of platform support and licensing? Or would you prefer offering it as an addon?
Thanks. Here I don’t know yet. On one hand I want to make it available for as many people as possible but on the other hand I would like to support its development more as it takes a lot of time and resources to do well. So I’m leaning more into making it as an add-on or standalone but with a price like 20$ or something around it so it’s cheap but would still allow me to focus more time on the development.
For more realism and control I’ve added new combustion mechanism and separated temperature and flames fields. Many other features were improved for more control and stability.
is there support for blender 5.0?