I vaguely remember a method of separating an image into high frequency (sharp) elements and low frequency (soft) elements, then being able to mess with each pass separately and composite back together. (probably using Ivar’s highpass matchbox)
Unfortunately my ageing brain has forgotten the details - any help would be appreciated.
Hiya Angus. Welcome. DIY frequency separation can be as simple as a median blur and a difference matte.
My new favorite technique is silhouette paint. It has these modes built in and is a brilliant implementation. Boris did a Logik Live and At 20 minutes in is a brief demo. This was life changing for me.
Cheers! I managed to find the answer via the FB group. The issue I was trying to fix though is interesting. The 2nd time I’ve seen it - lots of what appear to be “dead” pixels on footage shot with a slow shutter speed (to cause light streaks) on both an Alexa and some sort of Drone.
Hey, I’m not sure if this needed since you said your question was answered but just in case you’re wondering or for anybody else that stumbles across this: Crok_Separation is I what I use for frequency separation for beauty cleanup and I use Crok_Highpass to create a Highpass version of footage for tracking difficult shots. I’m not going to say that you can’t use Crok_Highpass for beauty but that is what Crok_Separation is setup to do. I hope that helps you out.
I want to share the way that I use High Pass separation. I was having a chat with @Ben and I thought that it would be worth posting on the channel.
I love it and I use it so so much that I believe everyone should have this trick in their tool belt.
So I make my own high pass filter. There are matchboxes that do it also (LS_Lumps, crok_highpass) but I have always made my own so I just never use those matchboxes.
Just use the a blur as an amount slider. Best to blur until the high frequency information disappears.
Then use the negative node to invert the image then use a comp node to blend back to the original at 50%.
The high pass image looks grey and sometimes low in contrast. I usually crank up the contrast and balance the exposure using the image window adjustments (shift-C, shift-E)
What you have now is the High Pass, all of the detail, and the blur which you can think of as the Low Pass, all of the colour, separate.
You can clone the detail without worrying about getting the wrong shade or you can smooth the colour without messing up the detail. You basically split up the image and work on the problem.
Combine the blur and your High Pass together again using a comp node set to Photoshop, Linear Light.
For example. If you wanted to paint out the flare in the sky. You can just copy brush paint the high pass and not worry about grabbing the wrong shade of the sky.
If you tried the exact same paint stroke on the unseparated image you would get this:
So there you go. If you split the image into High Pass and Low Pass you split your problem into two manageable parts which can be tackled separately.
One thing to look out for is edges. You will notice some high contrast edges getting a halo which needs to be considered when using this technique.
*EDIT - As pointed out by @Ben If this technique is done using Linear footage it will clip your highlights. So use ColourMgt and convert into a Log colour space first. Do the high pass and paint and then convert it back into Linear.
Great explanation.
I love these kind of topics.
When I use ‘highpass filtering retouch paint method’ I blur the image and after that a subtract from the original. Then an exposure node to see the details better. Thereafter I clone away in the highpass and at the end remove eposure node, Add back the highpass over the blurred image. And voila.
awesome. super post - many thanks for this - really really cool
And while we’re at it, you can also take that highpass and mix it back to the original with Overlay, SoftLight, HardLight and Spotlight for image sharpening.
Really rad technique! Love it. Saved me a bunch of time yesterday cleaning up a dirty wall that had a shadow pass through.
Question regarding your construction: I don’t quite understand linear light as a blend mode-- is there a reason to do the negative → 50% blend → linear light roundtrip over simply subtracting the blurred plate from source, and adding again? Is something lost in the simplicity of it?
Also: swapping the blur for a dollface gave me some slightly nicer results!
Any blur or median filter that removes the high frequency information will result in a good separation. It all depends on the circumstance.
I was taught this method which uses the Photoshop Linear Light blend mode. I assumed it was a technique that came from Photoshop. I had never really considered the alternative until this thread.
I was aware of the Grain Theft technique which subtracts the result of a de-noise from the original plate giving you nothing but the leftover grain.
They both work. I guess it depends which version you think will be easier to work with:
Top Left - original, Top Right - subtraction, Bottom Left - negative → 50% blend
Awesome stuff as always - thank you!
This technique totally saved my bacon on a car cleanup job this week. I’m so happy to have finally learned about it!!
Thank you to all for sharing your knowledge, with examples and setups, and then taking the time from your busy days to explain everything again and again. And again.
WTF…amazing man!! Thank you!!!
OMG I just tried this on a real job. OMG OMG OMG
Thanks Man!
Hey, I’ve got a lot of cleaning up to do today and the footage is pretty grainy. Is this the same technique you’d use to denoise (using Neat video) and then restore the noise at the end of the comp?
So far I’ve got the difference matte isolating the noise, but I can’t quite remember the comp function to restore back into the painted areas…?
Thanks for your advice!
I believe that “Grain Theft” technique is similar. Not using an obscure Photoshop filter thankfully.
There are a few ways to skin this cat.
Subtract the original from the denoise and then add the result back further up the chain to restore the grain.
Recently there was a post about how that gave less than desirable results in ACEScg.
Someone much smarter than me suggested doing a multiply and then divide further up the chain to fix this problem.
Sounds good but I havent tested this yet.
Ahhhh yeah, divide/mult sounds familiar! Thanks Rich. I’ve been Aviding for the last 6 weeks and I’m a bit Flame foggy!
I’ve tried it and it works a treat. Clever stuff.
impressive!
just used this technique along with a Median filter to help smooth out backgrounds light changes on a timewarped outdoor shot. Crazy light flickers got tamed really nicely.
Hi everyone,
Hope you’re all having a good day.
I have a cleanup job to do and make the actor with grainy skin pores look smoother and younger and get no look fake.
Will the highpass technique work for this?
The nuke guys are using the laplacian node for this to not totally remove the skin texture.
I think maybe one of these or a combination might also work beauty, washer, crok beauty, crok skin etc
Please correct me if I’m wrong … I’ve got to start this tomorrow…
Cheers