Digital Confusion - Faux DOF
- Thread starter erikals
- Start date
..a happy surprise is that Digital Confusion is fairly fast.
note 1... only 1 Digital Confusion filter can be used.
note 2... usually, it is better to add fake zDepth DOF in a video editor like Fusion.
note 3... Octane is Great for DOF renders - very fast.
note 4... i did try DPont DOF in an earlier video, but in this case it didn't work that well. (small shading bugs)
note 5... do Not use Oversample (it creates those thin black lines)
note 6... 1500x844 creates a better result versus 1280x720 (better AA and less artifacts) however, it will have less DOF.
note 1... only 1 Digital Confusion filter can be used.
note 2... usually, it is better to add fake zDepth DOF in a video editor like Fusion.
note 3... Octane is Great for DOF renders - very fast.
note 4... i did try DPont DOF in an earlier video, but in this case it didn't work that well. (small shading bugs)
note 5... do Not use Oversample (it creates those thin black lines)
note 6... 1500x844 creates a better result versus 1280x720 (better AA and less artifacts) however, it will have less DOF.
Last edited:
an alternative to Digital Confusion > DP DOF
however DP DOF is 3-4 times slower and OiDN doesn't like the data, so another denoiser should be used (Neat Video)
... to ensure no render errors, make certain the FGBlur BGBlur values match. (for example FGBlur 200% BGBlur 200%)
instead of running DP-Dof alone, you can also combine LW-Dof and DP-Dof
beware that it will need a Very high Antialias value in some extreme situations, (Heavy DOF)
so probably not worth it in most cases.
to conclude,
- Digital Confusion = Ok for rough renders
- DP-Dof = Ok-ish, but unfortunately OiDN App doesn't like it
- LW-Dof + DP-Dof = Way too slow
- making Dof in post instead is fast and smart, check for example the Frischluft plugin
- there are also other post filters/plugins foe After Effects / DaVinci Resolve
It would be good to make comparison between noise version and deflicker/denoiser version of the same render. Put as layers in PS and subtract them and enhance to see where are these fixed pixels (logarithmically rather than linearly?). Put as 2nd layer in PS on top of original to highlight them.
yes, it is a bit hard to see in the video.
so, uploaded the rendered sequence.
and it is probably best for people to run some tests themselves
so, uploaded the LW11 scene
i had more success later this night (an artist up late nights? can't be!)
using DP DOF i found better values for the render, but is slower than Digital Confusion to render of course
adding, Digital Confusion might be more forgiving in a render like this,
a medic animation flying through neurons, where only the camera moves.
Edit;
we can Kind of make it work, but in a case like that, Digital Confusion gives cleaner areas with AA than 1 for some reason.
but in other areas, it is best to use AA of 3. (hard to say what is better atm)
can't beat the DC render time tho'
unless one use Octane or use a zDepth DOF video editor plugin.
here, adding OiDN denoiser
adding Neat Video Denoiser is almost a must.
the render below actually also use a touch of LW Dof to make it all work.
rendertime ~2min on a super-slow laptop.
Last edited:
hm, i had forgotten about the plugin iDof - by Jeremy Hardin
might give it a go.
...Edit; gave iDof a run, couldn't get it to work properly.
there used to be an xDof plugin, but not available, and only 32bit.
Last edited:
I remember an old plug-in from Evasion3d, XDof, and it was a paid one. I never actually used it, as I usually do it in comp, but I remember people saying it was amazing.
And here the gallery
By the way, did you check the free dPontonier one? I test it yesterday pretty quickly but I feel it was pretty cool.
And here the gallery
By the way, did you check the free dPontonier one? I test it yesterday pretty quickly but I feel it was pretty cool.
WaveRider
New member
It's blazing fast...now. Probably didn't feel that way when I purchased it in 2003.
No idea about the standalone version.
Attachments
Tim Parsons
New member
The DOF in LW 2020 seems plenty fast. I have always done DOF in post but I think I might start doing some of it in camera now.
If DOF is fast it means it is cheat.. 
post-process cheat.
e.g. use Z-depth, subtract Z-focus, abs(), then use as Gaussian blur multiplier.
It could be done real-time even on CPU.
1920x1080x32 with matrix 7x7 it is 406 mln multiplications.
People have CPUs with 64 cores, so each core could do just 6 mln multiplications.
They can be done parallel by MMX/SSE/2/3/AVX.
If 4 multiplications are merged together, it'll be 1.5 mln CPU SSE/AVX vector multiply instructions per core or so.
post-process cheat.
e.g. use Z-depth, subtract Z-focus, abs(), then use as Gaussian blur multiplier.
It could be done real-time even on CPU.
1920x1080x32 with matrix 7x7 it is 406 mln multiplications.
People have CPUs with 64 cores, so each core could do just 6 mln multiplications.
They can be done parallel by MMX/SSE/2/3/AVX.
If 4 multiplications are merged together, it'll be 1.5 mln CPU SSE/AVX vector multiply instructions per core or so.
Last edited: