Have you heard about that thing where you displace the prev/next frame to match the current frame, using motion vectors? Then you average them three frames. I've seen a fusion youtuber talk about it, then replicated their setup. It's really hit and miss but sometimes it does the job.
I have seen some more advanced solutions that do something similar to what you're describing with smart vectors, (Sets a reference frame every few frames, blends between them with displacement), however I've never setup something like that myself as I haven't had the need yet. Would definitely be interesting to play around with though.
Optix temporal denoise since Blender 3.1 is amazing for animations, more people should be talking about it! No one seems to have tested it because there is no documentation. I got this far: Set render output to EXR_Multilayer; turn on vector and denoise_data layers and render an animation. Then open up python console and enter: bpy.ops.cycles.denoise_animation() It goes through each saved image file for each frame (so its best to test with just a few frames), and overwrites the layer "View Layer.Combined.RGB" with a denoised version. (I think you have to have an RTX card, Optix as the Cycles Renderer in System settings, and turn denoise on in render tab, select Optix, then turn off again - I don't know if all this is necessary, but when I initially tried with Cuda selected in system settings, and OpenImageDenoise in render tab, bpy.ops.cycles.denoise_animation() worked, but the results were poor. So I did everything I could to ensure Optix was being used, not any other denoiser, and got much better results. Open the EXR image sequence in Davinci Resolve, go to Fusion page, and on the MediaIn node, choose the layer RenderLayer.Combined.
The reason we project onto the model first is to stabilize the camera movement. If you put a time echo directly, the camera is zooming in, so it will echo with that movement (which we don't want). So basically, we need to remove the camera movement from the CG temporarily (using projections to stabilize and focus in on the texture), do the time echo (to average out the noise), and then re-add the camera movement back. Check out the video on my channel about UV Projections, it helps explain this a bit.
![Felix "Unfair" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/Oswujxm2Ag0/mqdefault.jpg)
If you missed the shockwave tutorial, this is the same scene:
ruclips.net/video/ErwClH-dQA0/видео.html
Have you heard about that thing where you displace the prev/next frame to match the current frame, using motion vectors? Then you average them three frames.
I've seen a fusion youtuber talk about it, then replicated their setup. It's really hit and miss but sometimes it does the job.
I have seen some more advanced solutions that do something similar to what you're describing with smart vectors, (Sets a reference frame every few frames, blends between them with displacement), however I've never setup something like that myself as I haven't had the need yet. Would definitely be interesting to play around with though.
Try Optix Temporal denoise in Blender 3.1 or newer. (see my comment above for how)
Optix temporal denoise since Blender 3.1 is amazing for animations, more people should be talking about it!
No one seems to have tested it because there is no documentation. I got this far:
Set render output to EXR_Multilayer; turn on vector and denoise_data layers and render an animation. Then open up python console and enter:
bpy.ops.cycles.denoise_animation()
It goes through each saved image file for each frame (so its best to test with just a few frames), and overwrites the layer "View Layer.Combined.RGB" with a denoised version.
(I think you have to have an RTX card, Optix as the Cycles Renderer in System settings, and turn denoise on in render tab, select Optix, then turn off again - I don't know if all this is necessary, but when I initially tried with Cuda selected in system settings, and OpenImageDenoise in render tab, bpy.ops.cycles.denoise_animation() worked, but the results were poor. So I did everything I could to ensure Optix was being used, not any other denoiser, and got much better results.
Open the EXR image sequence in Davinci Resolve, go to Fusion page, and on the MediaIn node, choose the layer RenderLayer.Combined.
Nice work!
Very informative. Thanks
Sir I have a doubt, why you didn't apply the time echo directly to the render ,
The reason we project onto the model first is to stabilize the camera movement. If you put a time echo directly, the camera is zooming in, so it will echo with that movement (which we don't want).
So basically, we need to remove the camera movement from the CG temporarily (using projections to stabilize and focus in on the texture), do the time echo (to average out the noise), and then re-add the camera movement back.
Check out the video on my channel about UV Projections, it helps explain this a bit.
@@CompositingAcademy thank you sir for your patience to reply me.
Thank you Alex. This is so informative 👍 Which version of Nuke has this denoiser ?
Glad you enjoyed! This is using Reduce noise, from Neat Video www.neatvideo.com
I believe it works in Nuke Indie, and the full version of Nuke.
Wish you a very happy birthday and prosperous year ahead alex .............. 💖💖
Great