A one hour video album made with the Save Point pipeline to test some new functionality in development. It contains 16 songs.
All visuals and music are original. Production cost was around $0.35.
Upvote
3
Ghost Forest
- Thread starter Nate North
- Start date
Beautiful stuff, Nate.
This was part of a stress test so I could measure the capabilities of the current pipeline versus the simple process of creating video music albums. The file above and these two full length movies were created in a single day using the current build of the system. That's 4 hours of finished video created and uploaded in a day.
The use case will not be to spam content, I'm just automating every non creative task, so I can continuously raise quality without pointless obstacles like spending hours each day editing cuts to a beat or sorting raw footage.
These early films are intentionally made of very easy content, just attractive enjoyable music videos. But substantial work has already been done on things like plot and character, that will in time lead to the ability to create the infinite branching film as originally promised.
Just for fun I'm releasing a new feature length film next week that will be a bit more sophisticated than these two. It's just a Terrence Malick style art film, But it's going to be a lot more impressive than these.
Last edited:
Not sure what you were working on but I thought your top effect was with the water... ripples, etc. and the least effective was the clouds/fog. Hope this helps.
Well I appreciate the feedback, but I actually haven't been working on graphical details for a while now maybe 4 months. I'm on the final stretch now just writing python code all day long every day. The visuals on those projects above are kind of medium settings for my pipeline because I can generate them five times as fast as the high settings, these are all tests for another part of the engine, which is the core.
This test for example is running on about 35,000 lines of custom python code, that does so many different things I won't even describe it here. None of it is visible on screen, but that's kind of the point. That's over four hours of edited content with every cut and fade and title perfect, Done in one day. Last year I was spending two weeks making a 3 minute video.
This test for example is running on about 35,000 lines of custom python code, that does so many different things I won't even describe it here. None of it is visible on screen, but that's kind of the point. That's over four hours of edited content with every cut and fade and title perfect, Done in one day. Last year I was spending two weeks making a 3 minute video.
Beauty, man, beauty. You said it only cost you 35 cents? That's awesome!
I'm making one right now that's at the medium high graphics level, I did already make a short one at the high graphics level currently available, Which I think you already saw
And one testing the vr output of the pipeline
I do appreciate feedback about the visuals and always will, I do put a lot of effort into making them interesting. Since version 8.6 (i'm now at 9.1) Of the pipeline I felt like the visuals were "Interesting enough to get started". I'm making a feature length art film this week as kind of canon fodder for development of the program, much like the three "Ghost Forest" films were. But this next one drops the stylization and goes for more of a photo real thing which you might like better. I tend to use heavy stylization for purposes of branding and consistency, valuing a cohesive style and recognizable look over absolute visual fidelity.
I wish I could explain more about what's so interesting about this core code and why I'm way more excited about it than any specific visual. But I've gotten to a point with the pipeline where it's doing things no one else can do yet, and I thought about it a lot, and I think it's a smart move if I just don't talk about it or explain any of my processes to any one ever, For fear of expediting the reverse engineering that will inevitably devalue my work. Nobody cares right now and I understand that, but a year from now when this thing starts suddenly making a lot of money people will come looking for clues about how to copy it.
As a metal guitarist you probably know this story.
And one testing the vr output of the pipeline
I do appreciate feedback about the visuals and always will, I do put a lot of effort into making them interesting. Since version 8.6 (i'm now at 9.1) Of the pipeline I felt like the visuals were "Interesting enough to get started". I'm making a feature length art film this week as kind of canon fodder for development of the program, much like the three "Ghost Forest" films were. But this next one drops the stylization and goes for more of a photo real thing which you might like better. I tend to use heavy stylization for purposes of branding and consistency, valuing a cohesive style and recognizable look over absolute visual fidelity.
I wish I could explain more about what's so interesting about this core code and why I'm way more excited about it than any specific visual. But I've gotten to a point with the pipeline where it's doing things no one else can do yet, and I thought about it a lot, and I think it's a smart move if I just don't talk about it or explain any of my processes to any one ever, For fear of expediting the reverse engineering that will inevitably devalue my work. Nobody cares right now and I understand that, but a year from now when this thing starts suddenly making a lot of money people will come looking for clues about how to copy it.
As a metal guitarist you probably know this story.
Last edited:
Basically I think that vying for ai visuals supremacy is a lost battle with 10,000,000 now bandwagoning on that particular track. I frequent AI communities where you see equivalent or sometimes better art than my pipeline can produce, but here's the thing, It's literally 10,000 community submissions of a 10 second clip of a pretty girl dancing.
Right now filmmaking AI research is as wide as an ocean and as deep as a puddle. I'll be able to develop or just buy unlimited high quality visuals from many sources, And it will all look great. That's not going to be where I win any significant market share though. The crowd is pretty much 100 percent smash and grab tactics, going for the fastest easiest way to look better than the next person.
While I totally understand why they would do that, and why it really works sometimes, I think that surface level is going to be so severely diluted as to be valueless, outside of services that provide compute power rental. I did spend the first three years of research trying to get visuals up to a quality level and down to an expense that would allow me to publish things people would actually watch. But now I'm in the end game, with all of those type resources I need readily available, and the only goal is to do something unprecedented, since that's where the real value is.
But I think a trained observer could already see something different here. At first the above post just looks like another mid to low grade youtube video, and of course that's exactly what it is. But how many times in the last 30 years have you seen an indie filmmaker publish 4 1/2 hours of any type of finished edited film in a day? Adobe doesn't have any tools to do that. Apple doesn't have any tools to do that. You can't do that in Davinci resolve. And this is the part of the pipeline's capabilities that I'm not particularly excited about, just a foundation for what's to come.
Right now filmmaking AI research is as wide as an ocean and as deep as a puddle. I'll be able to develop or just buy unlimited high quality visuals from many sources, And it will all look great. That's not going to be where I win any significant market share though. The crowd is pretty much 100 percent smash and grab tactics, going for the fastest easiest way to look better than the next person.
While I totally understand why they would do that, and why it really works sometimes, I think that surface level is going to be so severely diluted as to be valueless, outside of services that provide compute power rental. I did spend the first three years of research trying to get visuals up to a quality level and down to an expense that would allow me to publish things people would actually watch. But now I'm in the end game, with all of those type resources I need readily available, and the only goal is to do something unprecedented, since that's where the real value is.
But I think a trained observer could already see something different here. At first the above post just looks like another mid to low grade youtube video, and of course that's exactly what it is. But how many times in the last 30 years have you seen an indie filmmaker publish 4 1/2 hours of any type of finished edited film in a day? Adobe doesn't have any tools to do that. Apple doesn't have any tools to do that. You can't do that in Davinci resolve. And this is the part of the pipeline's capabilities that I'm not particularly excited about, just a foundation for what's to come.
Right now development speed is moving so fast due to more powerful tools becoming available, That I'm producing one submodule or another for my main program that does some tasks like you're talking about, about every two or three days. I've thought about breaking off some of these modules and licensing them but the problem is taking on the workload of trying to market and manage something small that wouldn't necessarily make a ton of money, While simultaneously trying to pursue a much larger goal that's worth exponentially more.
In example yesterday I spent 13 1/2 hours building a task specific module that simply finds and eliminates near duplicate video files in very large folders. It has its own built in custom media player that allows the client to specifically review and delete or move any video file located in the clients selected directory and fitting a set of criteria set by the client. It's extremely useful to someone like me who is dealing with thousands of files per week and copying them around a different directories where minor duplication accidents happen all the time, But who knows who else that might be useful to. Locating my customers would cost more than my customers would pay me for that exact module.
It's now the best tool in the world for that niche task, But it would probably cost me more to market it properly than it would make. One of the now 38 sub modules can watch any size folder of video clips and sort them into directories automatically by Anything you might search for. This footage contains dog that footage is at night etc. Two months after I built it adobe announced something similar as part of creative suite. Even if I had gone full in on marketing that module as a standalone program it would have become valueless before I got my money back. It's not that I didn't design it well, it's not that it didn't work perfectly, it's not that it wasn't useful, they weren't smarter than me, they weren't faster than me, But they have one million times the funding. That's just going to keep happening unless I do something so far beyond what they're thinking about that they can't catch up to by spending three decades of my salary in one afternoon. I'm literally finishing one of those essential tasks every couple of days like clockwork, but it's the larger machine the orchestrates all those small parts synergistically that I believe will actually provide me with a win.
One day this week I spent about 12 hours building an ai file renamer, Which is a lot harder than it sounds. You type a prompt into the gooey such as " Sci-fi blues song names" And point it at a directory and it will give every file a unique song name around that theme. This kind of thing saves a lot of time for someone producing 2000 music tracks a month, But I couldn't see myself getting rich off of it.
Unfortunately the modules that are really truly valuable are the ones that I need to keep secret. I did think about exactly what you're suggesting for a while, Selling off each small advancement piecemeal, But I think ultimately I'd be shooting myself in the foot because a lot of these unavailable components would kind of suggest my hidden goals once they were out in the wild and enough people started thinking about the possibilities. People will have a lot harder time putting 2 and 2 together if I don't box up 2s and sell them to everyone.
Lastly some sections of this pipeline would be kind of dangerous in the wrong hands, kind of what Mara and Celtic are talking about. I could probably make quite a bit of money right now by putting some of the main modules on the market, But the insane flood of content that would occur if I did that would bury me and everyone else. It's a very tricky situation moving forward, and just like with agi, everyone is aware that they should slow down and be careful, but none of us can, because in every case it's the least careful person who gets the product to market fastest, who wins all the financial power to determine what's happening going forward. Honestly this whole thing is a recipe for disaster. My current strategy is to land one big hit that sets me up to retire and then bow out of the circus before things get out of hand, which I guarantee they will.
I'd love to give people here on indietalk this system and just see what they came up with using it. But then one person gives it to a couple of friends, Then they publish it to the torrent sites, And six months later there are 500 billion hours of new video on Youtube and the entire market collapses. These modules that I can't talk about, They're as much of a problem as they are a solution.
What I really need is an investor that understands the need for secrecy. But you can see how it's hard to get investment when you're trying to make sure no one gets wind of how your technology works.
You can patent your technology if it is as unique as you say. NDAs to invertors as well.
Filing a real patent application cost me $25,000 last time I did it. You can file for a patent pending pretty much for free but the patent system has been badly overloaded by patent trolls and corporation legal teams basically grid marching against every possible word combination to lock out new patents. Basically that system is totally screwed now. I'll be able to get a patent once I have millions in the bank and can hire my own legal team and patent experts to help me navigate the impossible waters.
Vcs typically refuse to sign ndas. Basically 99.9% of people that go into venture capitalists are idiots and they have an idea like "what if a beach ball was **** purple". So vc's used to sign ndas and let these guys in, of course they wouldn't bite on the proposal that this guy had invented a Purple Beach ball, But then for the rest of their lives, Because of that nda, Any other company they funded or founded That involved any kind of beach ball or any kind of purple item would result in an immediate lawsuit by that guy who came into pitch one day and had them sign an nda. So basically they won't sign them anymore. Someone in my position has to find an angel investor, And most of the time those types of investors will sign in nda's because they're not running a business mill that such paperwork would threaten.
Maybe there is a better strategy than what I currently have in mind, i'm not sure. In my current situation the best option seems to be to simply not give any of this tech to even one person, and use it to maintain an incredibly strong competitive edge in the media market moving forward. I do need to figure out how to monetize all this possible output to at least a degree so that I can start filing for patents and have a legal team available, Because long term I won't get far without that. it's millions of dollars though, just to get off the ground with the real deal finished product, So I'll have to figure something out.
I think I'll probably be able to pull it off, Within a year I should be able to generate something like a season of low end TV for $100 and sell it to small streaming networks For a few million a pop. The way I'm building this, on the day that I can make one season of television at level X, I can make unlimited seasons of television at that level. I could probably nail some easy target like a fishing show and have it enter indifferentiable from the real thing 6 months from now.
Anyway that's what you saw in this post in those three examples near the top. The system reached a level where it could produce television of type and quality X, And now it's infinite, on tap.
If you're wondering for these posts why can't capitalize anything correctly it's because of this stupid dictation thing from Microsoft screwing up the capitalization all over the place.
Vcs typically refuse to sign ndas. Basically 99.9% of people that go into venture capitalists are idiots and they have an idea like "what if a beach ball was **** purple". So vc's used to sign ndas and let these guys in, of course they wouldn't bite on the proposal that this guy had invented a Purple Beach ball, But then for the rest of their lives, Because of that nda, Any other company they funded or founded That involved any kind of beach ball or any kind of purple item would result in an immediate lawsuit by that guy who came into pitch one day and had them sign an nda. So basically they won't sign them anymore. Someone in my position has to find an angel investor, And most of the time those types of investors will sign in nda's because they're not running a business mill that such paperwork would threaten.
Maybe there is a better strategy than what I currently have in mind, i'm not sure. In my current situation the best option seems to be to simply not give any of this tech to even one person, and use it to maintain an incredibly strong competitive edge in the media market moving forward. I do need to figure out how to monetize all this possible output to at least a degree so that I can start filing for patents and have a legal team available, Because long term I won't get far without that. it's millions of dollars though, just to get off the ground with the real deal finished product, So I'll have to figure something out.
I think I'll probably be able to pull it off, Within a year I should be able to generate something like a season of low end TV for $100 and sell it to small streaming networks For a few million a pop. The way I'm building this, on the day that I can make one season of television at level X, I can make unlimited seasons of television at that level. I could probably nail some easy target like a fishing show and have it enter indifferentiable from the real thing 6 months from now.
Anyway that's what you saw in this post in those three examples near the top. The system reached a level where it could produce television of type and quality X, And now it's infinite, on tap.
If you're wondering for these posts why can't capitalize anything correctly it's because of this stupid dictation thing from Microsoft screwing up the capitalization all over the place.
Last edited:
Wow. I get that you're excited about this and I get why. Last year: "two weeks making a 3 minute video," today, "4 hours of finished video created and uploaded in a day." Astonishing.
I kind of feel something like this, when GPT writes a paragraph of expository prose, instantaneously, that hits me, should hit anyone with an ear for writing, as elegant. A quick example. I was brainstorming a little, with GPT, some back and forth ideas, and it wrote this:
Anyway. Some beautiful work, even if still in testing phase.
Edit: I mean "Luddite," of course, affectionately
I kind of feel something like this, when GPT writes a paragraph of expository prose, instantaneously, that hits me, should hit anyone with an ear for writing, as elegant. A quick example. I was brainstorming a little, with GPT, some back and forth ideas, and it wrote this:
Anyway. You say that every cut and fade and title are perfect. I'll take your word for it, but the choices: Long shot, Medium Shot, Close up; pan right, pan left, pan up, pan down; zoom in, zoom out, static; hard cut, dissolve, superimpose; etc. etc.-- it's a lot of stuff, all part of the whole story-telling vocabulary of film, and, I imagine, all ultimately part of an individual, human vision. It can't be otherwise. So, Luddite friends, dont worry!
Anyway. Some beautiful work, even if still in testing phase.
Edit: I mean "Luddite," of course, affectionately
Last edited:
There's plenty of art to be found in the exact articulation of cinematography and editing, But that's just one side of the coin, On the other side......
It was about 2010 and I was gathering up a lot of professional source material, And during that time which is when the very first spark of this whole idea kind of hit me, I came across an old series of studio manuals for filming television dramas. I think maybe it was made in the early 90s. They had an exact reference for every single camera shot framing and movement for each type of scene. This is how we move the cameras for a two person conversation, This is how we move the shots at a boardroom table, Here's exactly how often to intercut between three characters. You're establishing shot should be two .7 seconds, You cut from this angle shot to that angle shot if these people are in seat 7A and those people are in seat 3B, And so on. I was thinking about it exactly like you're describing before I read that manual. They've had all that stuff locked down to a science for maybe four or five decades. The guy who wrote the manual was giving credit to his professor from his youth for teaching him exactly how to film every single scenario perfectly without even having to think about it.
Like I said there's still plenty of artistic interpretation and artistic vision available for a unique artist to explore. But I remember watching Boston legal back then after I read that manual and realizing that every shot for 100 episodes was within a few percent of that manual's description. I think perhaps it's art when you're on the frontier, Discovering new ways to do things for the first time, But maybe after 10,000 people have been doing something for 40 years, There are aspects where you end up wasting a lot of time trying to figure out how to make the wheel rounder.
I think that's why I love frontier art (Exploring the ways in which cutting edge technology can do things in ways that have never been done before), Because it genuinely gives you an opportunity to create new things. These tests above are pretty run of the mill artworks, I wouldn't say they are super creative, And right now there is only a minimal amount of intelligence that goes into the actual camera movements. However this whole thing keeps getting about one percent better every couple of days, And that's going to add up to something phenomenal in time.
I think, and I could be wrong about this that ai will always be bad at genuine creativity and that's the place where humans can still thrive, I'm just saying that determining how long a camera should linger on an actor's face as they react to stimulus X, Is probably more science than art at this point.
It was about 2010 and I was gathering up a lot of professional source material, And during that time which is when the very first spark of this whole idea kind of hit me, I came across an old series of studio manuals for filming television dramas. I think maybe it was made in the early 90s. They had an exact reference for every single camera shot framing and movement for each type of scene. This is how we move the cameras for a two person conversation, This is how we move the shots at a boardroom table, Here's exactly how often to intercut between three characters. You're establishing shot should be two .7 seconds, You cut from this angle shot to that angle shot if these people are in seat 7A and those people are in seat 3B, And so on. I was thinking about it exactly like you're describing before I read that manual. They've had all that stuff locked down to a science for maybe four or five decades. The guy who wrote the manual was giving credit to his professor from his youth for teaching him exactly how to film every single scenario perfectly without even having to think about it.
Like I said there's still plenty of artistic interpretation and artistic vision available for a unique artist to explore. But I remember watching Boston legal back then after I read that manual and realizing that every shot for 100 episodes was within a few percent of that manual's description. I think perhaps it's art when you're on the frontier, Discovering new ways to do things for the first time, But maybe after 10,000 people have been doing something for 40 years, There are aspects where you end up wasting a lot of time trying to figure out how to make the wheel rounder.
I think that's why I love frontier art (Exploring the ways in which cutting edge technology can do things in ways that have never been done before), Because it genuinely gives you an opportunity to create new things. These tests above are pretty run of the mill artworks, I wouldn't say they are super creative, And right now there is only a minimal amount of intelligence that goes into the actual camera movements. However this whole thing keeps getting about one percent better every couple of days, And that's going to add up to something phenomenal in time.
I think, and I could be wrong about this that ai will always be bad at genuine creativity and that's the place where humans can still thrive, I'm just saying that determining how long a camera should linger on an actor's face as they react to stimulus X, Is probably more science than art at this point.
Last edited: