Don't use parallel nodes like this...
Вставка
- Опубліковано 18 кві 2023
- Something you need to keep in mind when using parallel nodes in DaVinci Resolve.
--------
Grade faster with my Voyager LUT Pack:
procolor.ist/voyager-lut-pack
Get my free Kodak 2383 film print LUT for DWG and ACES here:
procolor.ist/freelut
Get all of my freebies with my free Color Grading & Look Design Starter Pack!
procolor.ist/starter-kit
Check out my ebook, The Colorist's 10 Commandments:
procolor.ist/ebook - Розваги
Ive changed my workflow many times in my 2.5 years or so on UA-cam, largely thanks to your insights! Its not easy being a Colourist who also does YT where you are expected to be 100% correct every time. Whilst my workflow and node tree is different from yours I am still modifying and learning all the time from your tips. I'm 54 years old and many years as a Pro Colourist and still love learning more!! Look forward to next insights!! Cheers, Darren.
Lovely comment. It would be delightful, perhaps even collaborative, as the two of you engage in a conversation and elucidate the advantages of one decision tree over another.
Love your videos 👍 thank you for bringing me here. You both are true heroes.
Thank you brother! You're one of the real ones, which explains why so many of us love your content! Excited to keep growing and learning with you!
Yes please do some collaborations!!!
you 2 are the GOATs... we appreciate you guys sharing your knowledge and expertise Darren and Cullen!
This is one of the reason why I watch your channel. Your complete honesty is well appreciated along with your thought processes for color grading. You're awesome dude!
Precisely why I promote Cullen and a very few other UA-cam experts.
Nothing beats integrity.
That’s just one of the many reasons why I love watching your channel Cullen.
Thank you 👍
Love that Cullen is one of those teachers who can be straight up about making a mistake. So unusual in so many teachers! Thanks cullen!
Thanks for the shoutout Cullen, and thanks for listening. That's what those online communities on youtube and forums are for. helping each other!
Don't worry:) I based my grading style on a mistake and I made it in two feature films:) You're the second colorist to admit a mistake, I knew another youtuber who apologized for misunderstanding Saturation boost and explained how it works. Then he disappeared, I don't know if that discouraged him, but he was a smart guy - Avery Peck.
Yeah I know that video. That guy’s great
Yup, I remember that video as well. I started following him before Cullen. Avery Peck is one of the greats in this field. I wish he put out more videos. Maybe him and Cullen could do a collaboration.
Cullen: I made a mistake.
Me: *Panics*
Cullen: Nobody died.
Me: *Relaxes*
Cullen: It was a color grading mistake.
Me: *Panic intensifies*
A true and outstanding professional. Cullen you are the best DvR educator bar none. Sorry Darren, FYI I've got you in P2 and close behind is Negative Phil. He has a dark side I find a bit distracting and he just disappears every now and again 🤷
Thanks for pointing that out! No worries Cullen, these type of videos are definitely useful to know what NOT to do.
Your library of content is critical while I'm transitioning to Resolve. Nobody, and I mean nobody has taught color on this platform like you IMO. Thank you!
Thanks Cullen, this is one more reason why you are so respected in this field. Great example for many go follow!
Kudos! It’s good to be aware of these kind of behaviors in Resolve. The parallel combiner node works in a certain way. It’s best to read the manual to fully understand what it does.
Mistakes is such a harsh word. I prefer to think of this as continually iterating towards better solutions.
THANK YOU. Last night when trying to sleep I was wondering about my node tree. I had texture pop as a parallel with primary nodes! Your lesson saved me on this film as I had just begun to grade. Much appreciated Cullen.
This is another example of how Cullen demonstrates his professional integrity. It generates more trust in his personal depth. Thank you!
I'm not super enthused by the math behind the parallel node. Resolve identifies the state of the image (let's call that IN) before your node graph is split prior to the node parallel mixer, then at the end of the structure when two images OUT1 and OUT2 are getting recombined, it creates a new image by taking IN + (OUT1-IN) + (OUT2-IN).
This seems to usually give results that look qualitatively acceptable, but it's honestly pretty hard to interpret just from a math standpoint. If you applied an exposure adjustment on one branch of a parallel mixer and contrast in the other, I'm not entirely sure what I should expect or if the operations are even photometric anymore without doing some math on a sheet of paper. Sometimes I'd probably want the parallel node mixer to do something more like IN * (OUT1 / IN) * (OUT2 / IN) 🤔. As a result I just use serial nodes with mostly commutative operations whenever I can and pull keys from upstream if needed.
You are a really cool guy Cullen! Thank you for sharing your experience and knowledge, and for setting such a great example for other teachers and human beings in general ❤🙏🏻
Kelly, a great addition to professional color correction
I noticed this some time ago as well but I never thought it was an oversight on your part. Ha
The template node structure I’ve been enjoying doesn’t look exactly like yours but it follows the same logic. All primaries horizontal at the top, a vertical row of secondaries all sourced from the input and feeding into the same parallel mixer, and then everything downstream of that is any sort of halation, glow, texture, grain, vignetting, etc.
Thank you so much as always for all of your insights and honesty!
Thank you for sharing it with us so we can all improve on our color grading journey! 🙏
Thanks for tip Cullen. I've noticed that almost 2 years ago, so stop using parallel nodes for that type of application. Even tho, you've teach me a lot of things too. Always trying to be beter :) thanks to have you in youtube
Thank you so much for all your tutorials and knowledge 🌈
This was the first thing that made me noise when I started using your structure as a base. And i just stop applying FX in parallel, just split them into Saturation or HSV Saturation and an extra node or two for adjusting saturation details. Nothing more and in the final queue add FX as you show in the video. It continues to be a very effective structure of work.
Late to the party here, but I kind of found this out on my own by accident recently with NR and Sharpening. Now I see where it stretches and what to avoid in the parallel part of my grade, and for that I thank you.
Thanks for the tip Cullen!☺️
I love the honesty!
Thank you, you are my color GURU, you and @DarrenMostyn actually, learned a ton of things from you two, keep it coming. Now i realize I need to revisit Parallel nodes....lol
Learning is never done! Thanks for sharing this
Thank you, Cullen!
Thank you... I think it would be cool to have a video that touches on spatial operating tools and how to identify them
Great… thank you for bring this issue out sir. Lot of people might be using it and won’t be able to to notice because of hair amount of shift, now we know. ♥️
I am very proud that I’ve had the privilege to meet you before you left Austin! You are a badass!
Thanks for sharing Cullen!
Thanks for letting us know. Imo being able to recognize your mistakes and let people know that you made it and now you have fixed and learn from it is always a strength. It’s genuine and that is what you want someone you look up to to be.
We love you Cullen
Nice find!!
Chill man, your great. Thanks for your videos and this one too.
Also great to see that people that good make mistakes too so you don’t feel that dumb with your own.
Amazing, thank you for sharing!
thank you Cullen
Most Excellent tip! And creds to you too.
Truly the GOAT. You're amazing, thank you. How do I learn Resolve from you one on one?!
I was doing this in the right, i never feel that put before parallel looks right. But thank you very much for this video and honesty :)
You're The Blue's Clues of color grading
Another weird bug is possible - if you change image to monochrome in WB node, for example. The color is still there for the bottom nodes, so if you use mask and start to make gain adjustments, color suddenly starts to pop in the image.
So, Exp, Ratio, WB and then sequential additional nodes is a preferrable way to go, in my opinion. Unless you really need to key something that was pushed out of range.
I think that, if this desire is to keep the keys intact after exposure, contrast and balance changes, then it would be best to simply take the alpha from that node to another one serially after them.
you're my go to for color grading :]
OMG!! Justo lo que acaba de sucederme. Ahora lo arreglo!! Gracias!!!
thanks for sharing this knowledge
Hello Guys Thanks for all great inputs. i learned a lot from you can maybe some of you can make a complete breakdown of all nodes ind DR like whats the case and how when and why to use them like complete tutorial
i would appreciate the importance of a good fundament thank guys stay healthy keep doing the great stuff
Thank you, as always admirable. 💪🏼
What a true legend !
Thanks for sharing, keep on!
Yes! you are great!
I usually use my Parrell nodes for when i'm pulling keys. when i use blur or sharpening, Grain, or glow i use them at the end of node structure. As humans were always going to make mistakes the great thing is how Cullen talks about growing and learning. If it's one thing i have always implemented in my life and thats alway be learning we can never know enough
Thank you for your honesty and to stop for analyzing a comment. I’m sure this kind of videos will only increase your reputation as a professional. All colorists knows how challenging this path of color grading is.
Echoing everyone's sentiment that this was a great learning episode.
On the topic of parallel nodes, and having watched your template for a while, I wonder if there's another aspect worth diving into? The goal of having the secondary isolated from the other adjustments is absolutely critical. Frequent beginner mistake is to pull keys in the middle of the node graph only to find things fall parts during adjustments and get messy fast.
However, in your node tree, you run both the secondary selective and the secondary correction on the parallel path, which is a bit unusual. To give an example - maybe you make another contrast adjustment in the mid-tones in node #4. It being on the secondary branch, that contrast adjustment moves things around based on the original, not based on the normalized/matched exposure and contrast you established on the primary branch. So your correction yields a different result for the same input. If you however, left the selective on the lower branch as a key only node, that then fed a correction on the top branch via a key input, you achieve the primary objective of isolating the selective from other grades, yet your secondary correction happens in the same context and mental model as your existing grade, which I think reduces mental overload because you don't have to adjust for the different contrast starting points.
The second benefit of not running the secondary correction through the same signal path as the selective is that you can pre-treat the selective in terms of gamma, saturation, denoise to help your selective quality and ease, without impacting the secondary correction. That can make the work better and easier.
I refreshed my memory on the parallel node via the Resolve manual. It doesn't state what the exact math is that is applied, but it looks like it simply iterates over the inputs and repeats each correction one at a time, assuming that they're all additive and independent. The manual does state the generally they assume that all inputs to the parallel feeds originate from the a singular prior node, thus as to simplify the math and assumptions. Now that doesn't seem any hard and fast rule, but maybe good to bear in mind. I've always found the parallel node a bit of a convenient feature, but also a black box as it's not explicit in how the operations get combined. The simple use case of simply stacking secondaries seems to be its original goal. But expanded or more liberal use, may be playing with fire as you saw with the spatial operations.
Thanks Cullen.
I've been using your parallel structure (or a version of it) for a while but I did strike the odd problem with it, which I fixed by using the method explained in this video. Was I ahead of myself? I have since developed a part parallel/part single line node structure that works better for me (mostly). Thank you for highlighting this issue, it helps to know that the solution I have chosen through my own efforts and trial and error is 'valid'.
Wow this is good to know :D I will modify my base node tree based on that ^^ Although I usually tend to do texture work either on the global timeline level or in a shot by shot basis at the very end of the node tree as serial nodes :)
Can you believe that I figured out this myself haha ,, anything texture or anything openfx I just make sure to do it after the parallel structure
Thank you.
Love the project title :D
This is awesome, thank you for sharing. A follow up question, is how can we tell if a change is a spatial operation?
If you don’t get it wrong every once and a while and own up to it, things seem off. Trust is everything. Thanks Cullen!
Top tip thanks
Now, I understand why not to put spacial effects into parallel nodes: it's just because 're-spaced' pixels get merged back with non-re-spaced pixels in the parallel mixer, creating unwanted artifacts.Such simple, and I did it wrong way too often, scratching my head why I had color banding and blocking, and other nasty effects from well exposed footage!
BTW, I've spent so many hours watching your videos - with every minute worth it!!!
Very useful finding! I also found that the parallel nodes are actually adding up the "modifications (formulas)" instead of "the end results". Perhaps it's too obvious to notice the differences. Let's say the upper node adds the exposure by 2 stops and the lower node in parallel doesn't touch exposure at all, it doesn't result in an image with exposure of one more stop by blending the two resulted images.
We all agree now It makes sense to use the lower node(s) in parallel to say selectively modify colour based on the unmodified hue or saturation so that the change is independent of the upper node(s).
Similarly, I wonder if there are scenarios where it could also make sense to place two spatial modification nodes in parallel so both spatial changes are independent of each other.
Nice of you to share this new knowledge! How does this apply to using Noise Reduction? Would you create a specific Noise Reduction Node before entering your parallel node structure?
Noise reduction should be on a serial node by itself.
I'm so disappointed... and hurt... and... who am I kidding?!? This is why we watch the channel! Your detailed explanations and thoughtful approach to both the art and science of color is exactly the example we all need to push further and improve our work. Also - I've been wondering specifically about this issue, so it's good to know I wasn't crazy. Great lesson!
Thank you so much for always keeping it real Cullen, on this topic I'd like to ask your thoughts on putting the NR before everything on a first node and than split the node tree and start with the structure you sugested? In the video you've pointed out to use it in the end, which way would u recomend?
I am also quite curious about that!
Interestingly, in a color managed project, I'm getting the same correct blur results regardless of where the blur is on my node tree before or after parallel nodes. In a node base manually managed project, I get the same result blurring just before or after a parallel node, however I get a much different result if the blur is before the IDT although still a full blur and not quite the same results you're seeing here.
Hi Cullen, I have been a colorist in Japan for more than 10 years. I regularly out my Noise Reduction on the first node and then I make a similar tree with parallel nodes. When I saw your tree before I was thinking why you don’t have an initial one or 2 nodes before your parallel tree?
I usually have to delete objects from many shots and usually I reserve one of the initial nodes for those objects. Also I keep a circular mask on my try an other masks that I know it can be useful to have.
Exactly my thought. NR upstream the tree… :-)
Is there a disadvantage to putting NR at end of tree?
this explains why I had a bad feeling about the parallel nodes structure…
What happens if on one branch I crank up saturation, and on the other one I lower saturation the opposite amount, do I end up with the original image?
That's the exact example when i ear from my professor these words "i'm here learning with you guys" 🤣 Anyway thanks for finding it out for us
I always knew there had to be a catch, just didn't know what it was
What about adding a vignette ? Can I do that in the secondary track with good result?
Hey Cullen! Thanks for the Update on your node tree.
So, if you put your sharpen node downstream, where would you put the noise reduction?
Is it also downstream? Thanks in advance!
Cheers Florian
Any link to the original video explaining this Node structure? (Or maybe better wait for the new improved version of it)
Hi Cullen and thank for being who you are as a person and professional. One question, you mentioned spatial as one. Isn’t a window or gradient spatial as well? Should these ones be avoided as well in a parallel node? Thank you.
Same question here. Logic tells me it shouldn't cause any artifacts if you're not using any spatial effect inside. If anything, its effect may be a bit weaker than in serial as its output is being blended with the clean signal.
That said, I really want to hear Cullen's thoughts on this!
Windows are okay, in my experience, unless you do monochrome in WB node.
does this apply when you use parallel nodes for face refinement?
Cool
"Anything spatial"... except for power windows, right? You have demonstrated the use of power windows to great effect, shaping light in the scene and this was done with a node in the lower branch of the parallel structure if memory serves. What other operations will you still use in the parallel branch Cullen? Thanks.
Cullen, you still haven’t talked about the 18.5 changes to the color page.
If you work in a colour management workspace, are you colour correcting or do you go straight to grading? In theory, colour management would be doing the colour corrections, right?
Can you stick the NR node at the very beginning? That’s what I’ve always pretty much done. NR is usually my very first node. I couldn’t tell you why. I just saw it on some tutorial forever ago when I started learning resolve and I just went with it. Have I been doing it wrong or what?
Hi,
I’m very impressed by this post, like all the others I’ve watched. Thank you for all the help I’ve gathered from those posts.
I have a question but don’t know if it’s the right place to ask it.
I exclusively shoot videos with my drone (DJI Mini 3 Pro) and I’m still wondering how to color grade my rushes.
I exclusively shoot in DJI D-Cinelike color profile and use 2 Color Space Transform nodes in my template. One, the first one in the chain, converts my clips from DJI D-Gamut - DJI D-Log (because I haven't found the description of the DJI D-Cinelike color profile) to DaVinci Wide Gamut - DaVinci Intermediate and the second one, the one at the end of my chain, converts my processed clips from DaVinci Wide Gamut - DaVinci Intermediate to Rec.709 - Gamma 2.4), well just because I don’t what else I should do.
So, my question is (if you would consider answering it) : is it a good way to proceed? And if it’s not the most stupid way to do it … should I leave these two nodes as the first and last nodes of each of the clips’ color management chains or should I consider using those two Color Space Transform nodes in the Timeline portion of my process (since all my clips come from my same drone)?
If you cannot answer these questions here, perhaps you can just advise me on where I could find help on this topic.
In any case, thanks again for the help and the inspiration I find on your channel.
Interesting;) soooo what Operations are ok on that nodes oO?
Does anyone have a list of effects and corrections that fall into the category that Cullen is referencing?
Anything texture/sharpening/smoothening/blurring/edge manipulations would fit in that "spatial" bucket of operations. Without having tested it, I'd assume that resizing, lens distortion, gate weave, and mattes (like film borders) are also bad ideas in a parallel node mixer. Not sure about Glow and fake Lens Flares, as those might depend on the implementation.
I also probably wouldn't use parallel mixers if your timeline color space's gamma was Linear. The fact that they're entirely additive makes them much more suited for a log state image.
So is it time to re-evaluate when a parallel structure is useful?
So, @Cullen Kelly Noise Reduction beginning or the end of the Node Tree?
Aways at the beginning
I still think that pulling a parallel off the input and then having the exposure on the top is strange. The exposure and balance etc. seems to me that they should be set before feeding anything - so the bottom row there could have some crazy balance feeding your nodes after just pulling off the input. I haven't tested your structure there because I have had my own node tree. Maybe I should play with yours a bit?
If you want a clean key - you can pull a node off the input and then feed the luma line anywhere in the node tree and your qualification will be stable.
I saw nothing wrong with the way you used to do it with serial nodes.
Parallel in my opinion is only needed when you require to feed two or more the same image - which then gets treated and blended.
Great that someone caught this and you are able to analyse and fix it.
What is funny is that I commented about this the first time you showed the workflow.
Yes play with it, please. It would be very nice if the different approaches were tested by professionals and I'm sure that something good will emerge for everyone in the end.
I struggle to wrap my head around this also, I think I need to try it.
Where can we go to look at some of your work?
Exactly what I think. Also, given many balance adjustments completely change hue and saturation relationships, wouldn't it be easier to pull keys out only then?
I ended up using Walter Volpatto's node tree structure.
Thank you for the great video!
I have an unrelated question:
Why am I getting different results by adjusting my exposure using offset in log space vs. using the camera RAW tab?
Because Offset, even in Log with Resolve Color Management, is not Exposure adjustment. We have the HDR Pallete Global wheel to reproduce that behavior. In ACES, though, Offset behaves much more like exposure.
Just think physically. If anytime we double the time that the sensor gets exposed to light we say we've increased the image in 1 stop, then to be able to completely behave like exposure and raise 1 stop we need to input 2x the amount of light that were before. In a Color Managed pipeline, changing the Node Gamma to Linear (to mimick how the light hits the sensor) and just increasing the Gain wheel to 2.00 in Y, R, G and B makes it behave exactly like exposure would. Similarly, you decrease in 1 stop by typing 0.5 in Gain. And 2 stops would be 4.00 or 0.25, just to keep the double or half ratio. But obviously you get limited to how much the sensor has captured in the low range above the noise floor, because you can't make things to appear if they didn't get captured.
@@GabrielPassarelliG Wow! thank you for the detailed explanation. However, I somehow thought Cullen always just use the offset wheel and say that in log space, that's mathematically equivalent to exposure adjustment... is it because it's a close-enough estimation for small adjustments?🤔🤔
@@GabrielPassarelliG For context, the reason Offset in a log color space kinda works as an exposure adjustment is this: suppose you have a linear code value X, then raising by a stop would require you to compute 2X. If you instead had LOG(X), then one stop brighter is LOG(2X) which is equal to LOG(X)+LOG(2). As a result, adding an offset to a LOG code value is usually ballpark kinda close to an exposure adjustment. However, this usually fails in the shadows because almost every log curve in practice is a piecewise function that's actually linear in the bottom end, and usually the format is more like A*LOG(BX+C)+D in the highlights, so it's not really perfect there either. TLDR, the math suggests that offset will do an okay but imperfect job of matching exposures in the highlights, and a poor job in the shadows. Gain in Linear is the better way, as you said.
does not apply to matte/qualifier blurring i guess, right?
Right
Hey, Cullen, do you offer a course?
“Yes! I have a course called Colorist Career Accelerator I offer a couple times a year. If you grab one of my freebies in a video’s description it’ll get you on my email list where I announce it.”
there are many professionals making 'mistakes' every day at work. We just don't see. Very often they are not mistakes, just less than ideal ways to do something. Nothing is perfect. The 'best' Hollywood films contain mistakes, and they also often are crap 6/10 films anyway (the whole film could be called a mistake). But we would love to work on one of these professional projects given the chance. Some of the 'best' films ever, are technically not complicated and may contain many 'mistakes'. Perfection is a myth. What is not a myth is good people, who are skilled and honest and pursue their art seriously
Just in time, I was contemplating where to use noise reduction in my node tree.
At the end
@@DannyGan no, NR is always frist. You do a single node of NR then you aply the paralell node structure and then in the end you put FX nodes.
First node
While many colorists prefer it at the start of their node tree, I put mine at the very end. IMO, noise reduction ultimately is a type of damage to the footage, and may as well do as little of that as possible as late as possible instead of damaging the imagery first thing.
@@CullenKelly Yes, I agree with you. I believe some informations will lose when using NR.
Hey Cullen, is it okay to use noise reduction down stream?
I'm wondering this too. Just in general where Cullen would place NR in this node graph scenario.
I usually do it upstream.
How does this impact your choice to put NR in a parallel node?
Had the same thought. Should we put NR in the first serial node before the parallell node structure to have the cleanest signal?
Put the noise reduction or added grain on a serial node before the parallel.
@@mrshaheedmalik Really, grain before the parallell structure? Not after?
I learn a lot from you, but one thing that I keep wondering about every time I watch one of your videos; is your skin really that yellow? And are there so few color gradients on your skin?
I think if my skin would look like yours in your videos, people would think I had a disease..
I follow multiple colorists on UA-cam. One of which is Darren Mostyn. In his videos, colors look real, whereas yours look all yellow. Is this on purpose?
It’s hard for me to imagine a scenario where the person that brought this to your attention wasn’t a huge d*ldo about this, though I doubt that’s the kind of feedback you get in regards to your best practices. You’re a colorist ninja, cullen. Keep hitting them with your 5d chess.
ua-cam.com/video/CufoCJiNcy4/v-deo.html PARALLEL NODES ALL YOU NEED TO KNOW