-
Current Capabilities of AI: practical aspect question
Hello ye, oh the enlightened in the ways of AI 🙂 I have a question which stems from a challenge I am facing in my short production. The challenge is the location, or, put it clearer, lack thereof. I have been looking for a real location that would look like a fancy home office. With no budge the task is daunting. So I am considering filming it somewhere and later enhance it with AI. I want to know how much it is doable with the current state of AI planforms out there. They seem to do a decent job with stills and slow motion and for very short clips, like 3-5 sec long. How much it could be accomplished for a my short (5-10 min)? There is no heavy action going on, just two people conversing. Yes, at times it might get a bit jerky and emotional, but there is no fight or shooting or anything of that sort. ChatGPT warns me that it might still affect the final result. We don’t want to look ridiculous holding a banana instead of piece of paper or with distorted faces due to limitations of AI. Any observations/insights? I know the safest is still to secure a location, but I am running out of options.
-
13 Replies
-
Hi Dovlet,
I don’t want to be pessimistic, nor do I want to discourage you, but I want to share my experience and some of what I know about the subject so you know what to expect and could use that knowledge to avoid obstacles. Frankly, the problem with AI is its lack of specificity due to the vast amount of data it relies on and its inability to form an impression of what you want by reading between the lines of what you write. As humans, during text conversations, we can unconsciously form an idea about the person we’re talking to, and this becomes clearer the more we read their writings, even if they haven’t revealed anything about themselves. We all do this daily, although our skills vary from person to person. This is something AI lacks. Therefore, you need to write exactly what you want, just like someone writing in a programming language. The only difference is that the output of code now always comes in handy, and there are no limitations on what you write. Whatever you write, there will be a result, but is that result what you’re looking for? That’s the problem. If you don’t know exactly what you want, or if it’s something simple and specific by nature, like adding a background to an image, it will be easy. Otherwise, I struggled a lot with editing even still images until I lost hope and edited them myself because the images It contained a lot of details, and the modification I was looking for was very specific. The funny thing is, I once asked the editor for his opinion about me, and he told me, “You’re someone who loves excessive precision and pays too much attention to detail,” as if he were resenting me )
I hope this is helpful to you.
-
Hi, Thank you for sharing your experience. I think it boils down to a) finding the right AI, and there 100s of them around and b) master the art of writing prompts. I experimented with AI, both stills and video, the result was often unpredictable, but I managed to achieve something that worked for me. But I am still not sure if current AI video editing would allow me to compete a full, albeit short, production, or I am limited to 5-10 secs. Would I be better off shooting with a chroma key screen? That will influence my decision.
-
You’re very welcome, Dovlet! I think we need someone who has tried doing something similar. There are people on forums who say they were able to produce entire animated episodes using AI, but they are not professionals, the kind who don’t even know exactly what they want, so they are amazed by anything they get as long as they know they couldn’t have done anything without AI. But a professional who has the skills to do the required work but wants to reduce the effort or budget, as in your case, would has a clear vision of what he is looking for and the idea he wants to embody as an artist first and foremost, so he will not be satisfied to change ideas just because the algorithms threw him into such direction. So, as I stated before, I think we need to hear about an artist’s experience in doing something similar to know what we should expect.
-
Hey Dovlet,
Storyboard your scene.
Carefully plan your actors placement etc.
Shoot greenscreen – then use free-to-use blender or Unreal Engine 5 to build your background.
Or buy “office elements” off the shelve – plenty of those around.
Example:
https://cosmos.leartesstudios.com/search?q=office
If you are looking for a tutorial, this is where I got mine from (this is my refferal link!):Yes, you can use AI for some of this, but the hard work is going to be to get the basics right from the start – before you look inside the viewfinder and press record on that camera.
Hope this helps get you moving with the project.
Atb
Mads
cosmos.leartesstudios.com
Browse thousands of ready-to-use 3D models and environments. The ultimate marketplace for creators working in Unreal Engine and Unity. Start building now.
-
Hi. Thanks for suggestion. The other day I stepped into the Unreal Engine territory and it looks intimidating. I have some modest experience with 3d animation using DAZ and Cinema4D, but this whole new universe seems too grand for my project. And it looks too heavy for my computer. DAZ offers a lot of 3d assets, so to other websites, so should I go this route I should not have issues finding various props. The real work is the compositing, camera angles, perspectives, lighting etc. That is what expect AI to come to rescue. On FB I saw this guy, Blake Ridder, posts his videos produced with AI in part or in whole and charges $79/mo for teaching it. So it really makes me scratch my head. See the example. https://www.facebook.com/watch/?ref=saved&v=878216351776126
facebook.com
Discover Popular Videos | Facebook
Video is the place to enjoy videos and shows together. Watch the latest reels, discover original shows and catch up with your favorite creators.
-
One useful distinction here is replacing a location versus planning around one. Would a 20–30 second stress test—with two actors, hands and props, and a small camera move—be enough to compare a real location, green screen/3D, and AI enhancement? For those who’ve tried it, what fails first: eyelines, lighting continuity, temporal consistency, or turnaround time? I’m trying to identify the test that shows whether AI saves money or merely moves the cost into post.
-
Hey Denise,
Good point about always making the test first.
However, over one hundred years of film-making, good cinematography (DOP/Lens/Lighting/Angle) can make most ordinary scenes look amzing. Or at least give enough space for the actors to do their bit.
Two good actors, dressed accordingly to what the script wants them to be. With a couple of props like good fancy coffee cup, fountain pen and expensive paper on what looks like a solid upmarket table. Then the viewer will soon have the impression of an expensive meeting room.
With a bit of editing, like an outside shot of an “imposing” building as a setup for the scene, then you can create an fancy home office without ever having one of those.Although, back in the day when I was starting out, we did phone the town’s mayor and ask to use his office. Even got him to do a mock wedding for us (Yes, I got married without getting married, as a spotty teenager).
All you have to do, is ask.
And that is one curious courtesy that no amount of prompting an A.i. can “teach”.
What always fail, is lack of vision and planning.
After that, it is not knowing your eyelines, and how to shoot for them.Having said that, there is nothing wrong with trying out some of the plenty different digital tools to get what you want.
Would be nice to get an update from @dovletov on how he is getting on with his project?
Or hopefully get to see his masterpiece.Atb
Mads -
That’s a useful correction: first test whether framing, props, eyelines and one establishing shot already sell the space, then compare any digital method against that baseline. When you say eyelines fail first, would you test them with a simple two-shot/over-shoulder sequence before adding background work?
-
Hey Denise,
It is attractive to consider using A.I. for everything, but that won’t neccesarily solve the basic problems on a production.
Yes, you can use A.I./3D software for advanced low-cost story boards. But if you as a film-maker can not visualize the end result, how will you be able to go on set and give directions, or operate the kit?
This is where experience, and plenty of trial and error, comes in to the frame.
The first thing to go is the eyeline, particularly where you film 2 or more people in single shots, speaking to one another.
Will we see cameras that can tell you when something doesn’t match up, undoubtedly, or will it be cheaper to ask A.I. to fix it?
Meanwhile, Christopher Nolan’s Odessey was shot on IMAX 65mm film.
That is very expensive to make mistakes on, hence why a crew at that level will be amongst the best in the world.
As humans, we can only aspire to be better, although it won’t stop us using A.I. as a utility, but still with us in control.
Right?
Atb
Mads -
Yes — that keeps the order of operations honest. I’ll test the eyelines and performance first with a plain two-shot and singles, then add any digital background only after the scene already cuts correctly. If the basic coverage fails, AI would only hide the problem.
Reply to this Discussion! Login or Sign Up