@@ClipTalks5 What version of Blender are you using? The setup on gumroad was built for Blender 2.93 and works until 4.1. This is stated on the gumroad page too. Blender changed how the compositor worked in 4.2 and I haven’t updated the example on gumroad with support for it yet. But the technique works and the sample file works fine in previous blender versions.

Yes

@@statixvfx1793 i want to learn lot of things but i don't have so much money can you teach me i am from bangladesh

@@statixvfx1793 i want to learn fusion can you help me

@@statixvfx1793 hello

@@statixvfx1793 need some more tutorials

Any tips getting this data out to Blender?

@@mtscott44 You can instance small spheres to the point cloud if you're setting the point cloud to renderable. You can instance using the replicate3d node and the fbx exporter node to save it out.

@@statixvfx1793 Cheers.

At the time of recording, Manjaro. But ive since switched back to Xubuntu

@@statixvfx1793 Bro, please make more tutorials in Fusion.

If you dive into the pCustom node it turns out its a lot closer to (old) houdini pops actually. Combine it with sets and you do some pretty advanced stuff!

As soon as movement comes into play, davincis build-in temporal denoiser produces a blurry mess, unusable imo., but for static stuff its pretty good. It can't utilise the vector map, that's why the node based is much better but generating the vector map will add a bit of render time, for me it went from 0:46 to 1:02, still much worth it. When building this setup in blender, use blender 4.2, in the compositor side panel under settings: set it to GPU (it's 5x faster for me than cpu) But davinci is still a bit faster and easier to set up i get the best results by: -ideally denoise your render passes first individually in blenders compositor (don't denoise the Color passes, you would just loose details because they are usually noise free), then temporal denoise ("direct"and "indirect" passes (you could also add them together before denoising to save render time, if you don't need them seperated for post production)) -avoid animated noise seeds (adds more flickering) -increase resolution instead of samples for better quality (file size and denoising time is increasing tho, so you need to find the sweet spot for your system if vram isn't a limiting factor)

@@Lakus231 Thank you.

Super fine detail and transparencies are the trickiest bits. But using the velocity vectors to add a bit of extra motion blur sometimes helps. You can push and pull the pixels back and forth so at least "smear" some of the noise away. At some point its just whatever tricks you have combined that works. Never a single solution.

Ideally both, but its highly dependent on the shot. Like mentioned in the video, hair/fur and transparencies can cause issues and would have to be solved slightly differently.

if you want to do it in blender, go with version 4.2 or above (in the compositor side panel under settings: set it to GPU, for me it's 5x faster than cpu) But davinci is even faster and easier to set up. results looked identically.

Thank you, I just got busy with real world stuff. I have a pretty big backlog of lots of cool fusion things id love to share tho....need to find the time..

@statixvfx1793 so I use Fusion within the Davinci Resolve editor - looks like all the nodes you referenced here are there as well. Is it still possible to use Fusion outside of resolve?

@@TabletopToolbox Yes it is, and its the preferred way for a lot of things. Especially CPU and RAM usage. Fusion Standalone is much less hungry than in Resolve. Granted, there's a few nodes thats only available in ResolveFusion, like surfacetracker and reduce noise.

i use blender for that , just export the scene to fbx and do every thing in blender very easy

You can smooth the mesh by blurring the position/displacement image :) Granted that also means it'll interpolate between actual tracking points so the mesh will be less accurate.

@@statixvfx1793 I will try that, thx again

It’s a bit of a hack, but it helps :) Glad you enjoyed it.

Thanks, the real power comes from balancing both temporal and spatial denoising and then re-graining where needed. This is especially true for any works where you have to integrate cg into plates.