AI Color matching and AI for Editing

So, I have 2 requests for a future Resolve version (hopefully the next one) because I think they will increase any working colorist and editor's productivity and efficiency, which is very important to me and I'm sure you. The less work we have to do, the more we can create.

1. AI Color Matching. The new look dev features in Resolve 19 is going the right direction. I am still using ColourLab 3 AI because it has a look design and shot color matching capabilities and it puts me in a very good start, and many time, 60-70% there. I subscribed to ShotDeck and I use their references to get inspiration and work with clients, then I can download reference shots from them and use it in C3AI to do a color match. Then after doing that, I push it back to Resolve and finish it off there. It will be a nice feature add if the new look dev feature extends with the ability to take images and shot match to its color scheme, so it can work with any reference image including those from ShotDeck.

2. AI for Editing. The thing I can envision here is that I could import some kind of markup document that's generated from the script and using generative AI figure out all the sequence of clips and takes in order of the script and put them on the timeline of the CUT or EDIT page. It doesn't have to be perfect and even just close enough to get started and save a lot of time trying to figure out where the corresponding files to the lines in the script is on the drive. I understand that this may be a bit challenging because many times things don't go as written on the script. But some curious engineer at BMD may be able to find a way.

Ellory, just for me to understand, you are asking to make a “press a button, copy this work of another artist, and make my material look close enough with no effort” function for free? (And by the way killing someone else application and work)?

I’m very aware of what Dado Valentic is working on, and I might agree or not on the specifics of it, I do make an objection on deliberately make an function to allow copying someone else work effortless.

i might be old style and you can certainly tell me to “go back to farming” or whatever expression you might have, but I make an objection on using someone else work and style without recognition.

Yes, yes, you guys can tell me to go on with the times and other similar “that is the future man” kind of objections, I wish for you that your career will be long and prosperous when you will be the victim of electronic plagiarism….

- just my 2c.

+ 1, I would like to see something like this too.
The actual color matching tool works, but it's very often a miss.

I personally would like a more intelligent "auto" white balance too.

The one on the color page works only on "perfect" footage. I work mostly with regular "non" log videos, with people often using automatic white balance on their camera (I can't do anything about it).

This gives me an idea for a feature request

AI Color Matching. The new look dev features in Resolve 19 is going the right direction. I am still using ColourLab 3 AI because it has a look design and shot color matching capabilities and it puts me in a very good start, and many time, 60-70% there.

What if the look you're trying to match to had five power windows, two layer mixers, a couple of keys, and a bunch of secondaries? The basic tools in a Matching engine are not going to get there.

Half the stuff I do these days has all kinds of windows going on, so matching the basic look doesn't finish the job. It's the "everything else" that takes all time.

I keep pointing out: "they don't even now have Full Self-Driving A.I. that can take a car down Wilshire Blvd. at 5PM on a Monday afternoon in LA. There's no way they can use A.I. for color correction or for editing or for sound. There's too much chaos involved with the process."

I don't have a problem with A.I. assisted tools, and I would welcome A.I. assist for conform, so it would automatically go out and look for all the missing files from a show. Creative jobs? Nope, not interested.

My response to Walter. I’m not advocating plagiarism at all. The movie industry is always a copy of someone else art either for better or worse but it becomes new art. So copying is not new. In fact ShotDeck lets its subscribers download images of a certain movie that is used by many to copy or try to “match” its look. For example, a dessert sense may adopt the look from Dune. That’s not copying. Now what Dado Valentic’s ColourLab AI3 lets us do is take that image that was downloaded from ShotDeck and use it to match the colors using AI. The colors may end up the same but the recipe is definitely not the same. This latter statement is now a response to Marc, with the issue of complex nodes (multi-windows, etc). I don’t think ColourLab tries to reverse engineer the color workflow; it just tries to make it look the same. And today, we do that all the time, whether it is using styling LUTS or by brute force taking a reference shot and dialing wheels and knobs to match ones’ shot with it. Dado Valentic AI just automates it and cut down a lot of work so coloring can be easier and faster. It is not replacing people. It’s just increasing productivity and allows the colorist to deliver faster, or grade more business. Just like in software development today, a big welcome to software engineers is GitHub CoPilot. It will write code for you so the engineers can deliver more value. If you don’t know what CoPilot is, Google and you’ll be amazed. So it’s kind of the same thinking… enhance the tools to automate what can be automated with color grading so colorist can do more with less time and tweaking. We might disagree and that’s cool but if Resolve holds back and Dado Valentic gets acquired by Apple or Adobe or even Avid, then they will give BMD a run for their money. What Dado has come up with is pretty neat and although AI should always be made to assist and not replace, it is inevitable that this technology will be the norm.

It is not replacing people. It’s just increasing productivity and allows the colorist to deliver faster, or grade more business.

I'm not buying that. I think the reality is that less-talented people will start to believe they can do the work of talented, experienced people by pushing a button. I'm opposed to this for a lot of reasons -- and I also think it's not that easy to replace them.

Again: I don't have a problem with machine-learning tools that track faster and better, or an A.I. tool that literally will go and look on 9 attached drives for a missing file in the background. I'm fine with that. I just draw the line at people believing that A.I. tools can replace or equal talented people doing creative work.

It is not replacing people. It’s just increasing productivity and allows the colorist to deliver faster, or grade more business.

I'm not buying that. I think the reality is that less-talented people will start to believe they can do the work of talented, …. I just draw the line at people believing that A.I. tools can replace or equal talented people doing creative work.

What’s wrong with having someone interested in doing color work that’s not talented or experienced yet if they are willing and able to do it and the tool can give them the ability to be productive at least up to 50-60% of the effort? I don’t see anything wrong with that. I think it is not about AI replacing people that is the issue but fear of change by those people that they will have less control. We are talking about a color grading tool that even with generative and assisted AI still requires people to use it. DaVinci 19 has a lot of AI features that has made it much easier to use. Shot color match is just going to be one of them down the road and Dado has shown how good it can be with ColorLab AI. I use ColorLab in conjunction with DVR Studio and it gets me through 60% so I can deliver half of the time faster than I used to, and my clients like it because we sip through shots on ShotDeck that comes close to what they’re looking for and try them for size. Once they get something that’s close to their liking, that gets pushed back to Resolve and additional tweaking happens - hence it is not just a one button press workflow. There’s always work to do that ends up with new creative work.

This falls under the basic discussion of AI stealing creative work of people and how to deal with it.

I'd prefer a 'simple' AI camera clone function.

Filmatic AI ColorClone does a pretty good job of matching cameras as far as I can tell and is quite affordable, maybe BMD can acquire them

What’s wrong with having someone interested in doing color work that’s not talented or experienced yet if they are willing and able to do it and the tool can give them the ability to be productive at least up to 50-60% of the effort?

I'm fine if they have the same ability as I do. But I'm not a fan of people believing they can do the work of people with ten times their experience.

Shot color match is just going to be one of them down the road and Dado has shown how good it can be with ColorLab AI. I use ColorLab in conjunction with DVR Studio and it gets me through 60% so I can deliver half of the time faster than I used to,...

I know Dado and like Dado, I've had lunch with Dado, and he's had me on as a guest on his classes before. The shot match function works in certain situations, but not all situations. If you have two shots that are wildly different, shot on different cameras, and one has lots of relighting going on... it won't work. My issue is I can already do all of it by hand, just with my fingers and my eyeballs. And I can do it in seconds, as opposed to the minutes it takes with a 3rd party program. My match is also 100% perfect, for the simple reason that I'm looking at it.

I think you confuse A.I. with intelligence, ability, and experience. It's just not true. The need to understand the tools has never been greater, and the amount of functions and plug-ins and modes in Resolve has never been greater than it is right now. I'm already on record as saying I'm a big fan of the machine-learning tools in Resolve 19 that have greatly improved tracking, stabilization, and Face Refinement, so there's always a place for tools that help colorists and editors. But there is a limit as to how far machine learning and A.I. can go.

I'm fine if they have the same ability as I do. But I'm not a fan of people believing they can do the work of people with ten times their experience.

Marc, with all due respect, I am not saying that people that are less experienced but have such AI tool will be equally as good as someone with 10 times the experience. The tool just helps them get to where they need to be. I can do everything by hand too but the value of the tool is to provide productivity when time is limited and your business needs to produce or bust. I have a very open mind and I see young people so capable, they can do the work of someone that had many years of experience. That should be encouraged and doors should not be shut because of lack of experience. Everyone, including you and I, started from somewhere. The AI tools just makes it easier and faster to get started.

The shot match function works in certain situations, but not all situations. If you have two shots that are wildly different, shot on different cameras, and one has lots of relighting going on... it won't work.

100% agree. That’s why you need to know what you’re doing and how to leverage the tool. I never said that by using such the shot match function, it’s just a one-click job and you’re done. For crying out loud, it is just a tool to get something close to a reference shot, assuming the source and target are not wildly different. Even with that, you will still need to make adjustments. Where and when it is functionally viable, it saves time for the colorist… that’s it.

My issue is I can already do all of it by hand, just with my fingers and my eyeballs. And I can do it in seconds, as opposed to the minutes it takes with a 3rd party program. My match is also 100% perfect, for the simple reason that I'm looking at it.

Well that’s good for you. The opposite may be true for others. Regardless, everyone working on color grading will have to be looking at it and subjectively make decisions what is good and 100% perfect for them. BTW, there is not such thing as 100% perfect but I know what you mean and I digressed.

I think you confuse A.I. with intelligence, ability, and experience.

I know the difference believe me. But nothing will convince you if you insist otherwise. That’s raw intelligence. I think you are conflating experience with generative smart logic that is harness from models that makes tools smarter. People grow with experience, machine learns by training and model generation. People get educated too. Machines, like dogs, get trained. There’s that saying … “You don’t train people, you educate them. You train dogs.” Again, I digress but it’s education over time that makes experience. It’s not a repetitive process that makes you experience.

... But there is a limit as to how far machine learning and A.I. can go.

Yes, for now there is. It’s up to us as to how far it can go. If it can improve humanity and complement our abilities, I’m for it.

Recently, I attended an AI conference. In one of the session, the PowerPoint listed Einstein’s IQ between 160-180. ChatGPT 4 intelligence was measured between 165-185. Not that it matters for shot matching.