About a year and a half ago I quit my cushy high paying jobs as a big data engineer and started a YouTube channel with the hopes of producing short films. I had absolutely no filming or editing experience at all and it shows. I've been lurking here for a few weeks and learning a lot. I wish I would have found this forum a year and a half ago. Due to my utter lack of anything close to film making experience I have yet to produce anything worth watching. My terrible YouTube channel turned into a gardening channel as that is what I had to film while learning. To add to the irony of my stupidity I ended up on a low carb diet so I don't even eat the things I film growing in my garden. Now as to why I jumped off this cliff. Well, I got tired of seeing the same trash films and videos. I miss watching a good unique story. I have a great appreciation for early films produced without modern technology. The directors, actors, and crews were very resourceful and creative in bringing stories to life on the screen. Now I wouldn't say my time invested has been a failure as I have learned a great deal about editing and how YouTube promotes content. All that said, now that I have learned a few things, I would like to return to my original intentions and create some good stories. Nobody wants to watch my grass grow and I'm tired of filming it. I really like the little things that add to a film. The sound effects, lighting, and gimmicks. Years ago I played around with making 3d worlds, and my favorite part was adding the sound effects and lighting to set the mood. Due to my now dwindling budget I have been trying to tell stories using static images that I manipulate in the video editor while telling a story. While it's still not what I want to be creating for now it at least allows me to tell my stories. I really miss having critics. It seems the whole world has become afraid of passing criticism. I don't like the sugar coated world, I want to know what I do wrong so I can do it better the next time. When I started this journey I wanted to create a post apocalyptic comedy in a style reminiscent of old cartoons. Where there really isn't any dialog, but instead quirky props and slap stick situations. Think loony tunes mixed with ma and pa kettle in a destroyed world. Anyhow that's what has brought me here.
Hi, I'm an idiot engineer and proud of it.
- Thread starter shofstetter27239
- Start date
Welcome!
I tried creating some short videos on a youtube channel myself, I never got any traction.
I have about 50 subscribers after 10 years
It's rough! People expect to see beautiful A-list talent with a 200M budget, and they have a list of 110 other shows recommended by their friends that they still haven't gotten around to yet...
Very tough field.
I tried creating some short videos on a youtube channel myself, I never got any traction.
I have about 50 subscribers after 10 years
It's rough! People expect to see beautiful A-list talent with a 200M budget, and they have a list of 110 other shows recommended by their friends that they still haven't gotten around to yet...
Very tough field.
The best way I can describe YouTube is that it is an advertiser content mill. The subscribers and likes are meaningless metrics. All YouTube cares about is CTR, viewer retention, and session watch time. Those three things are how they determine the "quality" of your videos. And that is why you often see strange editing done by creators there. Things like cutting out every moment of dead time when nobody is talking to speeding up the audio so you have to rewind the video and listen to it again to understand what was said. To the dopey ways people try to get viewers to click on the end screen elements. Let's not leave out the insistence on click bait style titles and thumbnails. None of which improve the viewers experience. YouTube also does it's best to crush any hopes you may have about being creative by forcing your channel to be about only one subject. At this point I'm only using YouTube as "free" video hosting. If you want YouTube to actually show your videos to a specific target audience then you have to pay them to promote it through there advertising program. Which isn't really intended to promote films, rather it's intended to be used for commercials to sell products. I did get my channel monetized, and that in itself is a joke as it just means you get to keep 40% of the ad revenue they make from the traffic you drive to the platform the watch the content you created. The "merch" you can sell is also a big joke as you get to pick from the same garbage sold by the same vendors as all the other creators. The super chats, and stickers make me think of the whole platform as only fans for ugly people.Welcome!
I tried creating some short videos on a youtube channel myself, I never got any traction.
I have about 50 subscribers after 10 years
It's rough! People expect to see beautiful A-list talent with a 200M budget, and they have a list of 110 other shows recommended by their friends that they still haven't gotten around to yet...
Very tough field.
I see people that focus on a single subject, like photography, color grading, etc build up their audience and then try to sell that audience on a product or service. Like a photography book they wrote on lighting with a bunch of pictures, or a masterclass on color grading, etc
so even these success stories, they're trying to make the real money outside of youtube.
At this point I'm not really trying to make any money from what I create. It would be nice to have some to be able to produce better videos, but I have thought of a few gimmicks I can make use of to get the most out of the YouTube Algorithm.
This is one of my unpublished videos I created as a proof of concept for a choose your own type story. It has one end screen element linked to another unpublished video. They are accessible through direct links though:
I think this is one way I could have fun telling a story on the platform, but it's a bit difficult for me to create without actors or a set. Generating then editing images like what I used for the background isn't really a viable option when it comes to actually showing a character since I can't really regenerate the exact same image with only slight modifications. AI just outright sucks for anything other than simple one time tasks. I've thought about just going around and filming some interesting places to get some of my own stock footage, then using a green screen to add in characters. I'm not sure if that would be any easier though. I myself am a terrible actor, even on a good day. I'm not even good at narration. My brain doesn't consistently work well with my mouth and I'm very dry.
This is one of my unpublished videos I created as a proof of concept for a choose your own type story. It has one end screen element linked to another unpublished video. They are accessible through direct links though:
I think this is one way I could have fun telling a story on the platform, but it's a bit difficult for me to create without actors or a set. Generating then editing images like what I used for the background isn't really a viable option when it comes to actually showing a character since I can't really regenerate the exact same image with only slight modifications. AI just outright sucks for anything other than simple one time tasks. I've thought about just going around and filming some interesting places to get some of my own stock footage, then using a green screen to add in characters. I'm not sure if that would be any easier though. I myself am a terrible actor, even on a good day. I'm not even good at narration. My brain doesn't consistently work well with my mouth and I'm very dry.
Oh boy do I have great news!! You came to the right place!!
You should talk to @Nate North - he's like 100 steps ahead of you, he's been writing and integrating software to accelerate this exact pipeline of development full time for the past few years.
Here is his thread about separating characters from the background
Preview of his current project
Working sample of choose your own adventure implemented on youtube
Search link for other threads
Search results
Last edited:
Welcome!
Those demos are a bit old now, here's the most recent one, if you're interested in this technology for CYOA filmmaking. Basically I just looked at some efficient models for old school animation studios, and replaced as many roles as possible with AI automation. You still have to do all the creative work, writing, directing, editing, etc. But no more tweening, or rotoscoping, or concept drawings, or hand animating. It shaves off about 97% of the work hours compared to something like a 90's animation studio. I'll be glad to help you get started, though how much help I can actually be depends on several factors, such as your GPU model, programming experience, monthly resources, etc. It's not expensive, but does take a lot of time, and is drastically slower if you don't own a fairly strong computer.
Those demos are a bit old now, here's the most recent one, if you're interested in this technology for CYOA filmmaking. Basically I just looked at some efficient models for old school animation studios, and replaced as many roles as possible with AI automation. You still have to do all the creative work, writing, directing, editing, etc. But no more tweening, or rotoscoping, or concept drawings, or hand animating. It shaves off about 97% of the work hours compared to something like a 90's animation studio. I'll be glad to help you get started, though how much help I can actually be depends on several factors, such as your GPU model, programming experience, monthly resources, etc. It's not expensive, but does take a lot of time, and is drastically slower if you don't own a fairly strong computer.
Very impressive!
As far as my abilities, I am a Big Data Engineer in my last position I was processing petabytes of data per hour in the amazon cloud.
I have 23 years of programming experience mostly python and javascript, but I have also worked with scala, C/C++, php, bash shell scripting. I've worked on a few machine learning projects, and most of my roles have involved ops automation.
My PC specs: Lenovo IdeaCentre Gaming 5 with 64gb memory intel I7-12700 x 20, NVIDIA GeForce RTX 3060. I dual boot Ubuntu 23.04 and Windows 11. I should have stuck with 22.04 as I haven't been able to get the NVIDIA sdk installed as it hasn't been ported yet. I do have CUDA setup, and the backgrounds in my video were generated on my machine using invoke.ai
I'm not too worried about the specs of my machine because if needed I could always setup a kubernetes cluster in either GCP or AWS and perform some more intensive work using custom built docker containers.
My monthly resources are a bit tight at the moment as I am just now going back to work. I got a position as a WITCH contractor in a hybrid role. So it's going to be eating into my time a bit, but I should have some decent cash flow starting next month.
Ok, good news bad news. But mostly good.
Your qualifications are pretty much ideal. Honestly, you couldn't be better positioned on the tech end at least.
You can absolutely run everything on a bunch of rented cloud GPUS, and some aspects have to be done externally, for access to A100 cards for certain aspects, mainly character animation computations.
When I asked about monthly resources, it's because I get a lot of people in very bad financial situations these days, and something like buying an external HDD, or paying 200 a month in external coprocessing is a huge deal for them. We are an artistic community, so underfunding is..... common. From your description I wouldn't think it a concern.
Essentially you should be in good shape.
I did integrate UE5 into my pipeline early on, but it's not mandatory. It does allow a path towards total control of the AI output in 3d space that wouldn't otherwise be possible though.
About Hardware. If you're looking to go all in on this path, as I did (I have a RED Epic gathering dust now), you would probably want a 3090 or 4090 locally, with the big deal being the 24gb Vram. This type of work can eat up HDD space fast, especially if you don't have huge amounts of time and great organizational skills to manage output. I'm medium organized, but have to share time between many tasks, so my data storage is now about 40 TB local access. It's in no way a minimum starting number, just where I'm at after a few years, trying to preserve the dev chain. I'm probably rather overzealous about it. You're already in a good place, by which I mean I typically have to spend time explaining to would be filmmakers that they need to have a case with fans in it to do industrial computing tasks.
Can you tell me a bit about your interests and goals in terms of CYOA filmmaking?
The bad news is that I have a migraine today, and I probably won't be of much help to you until around monday, so not a big deal.
Last edited:
I'll post a couple extra videos here.
This one shows what's probably the most significant concept in terms of how my pipeline works. It was built from the ground up for CYOA development, and I expect you'll see this and understand immediately what it means.
Here are a couple showing what I refer to as skeleton design, Motion scenes created in 3D software unreal engine in this case, That can later be reskinned by the ai layer, Collapsing large stacks of tasks into their most efficient form.
This one was an interactive test with choices at the end, Not much story but about 70 interlinked videos, Just using the base layer only.
Here's just a UE5 short we made. This one really shows how significant it is to have full independent control over cinematography and subjects.
It was the first successful test I did of the core tech working properly (sort of)
That video above shows how we can use AI to seamlessly combine live action footage and cgi sets in a single stylized output layer with total control.
Obviously you're already familiar with prompts for ai imagery, In this system these UE5 videos essentially are the prompt, Allowing unlimited communication of detail and refabrication for different purposes using text prompt modification and other bespoke control methods.
This one shows what's probably the most significant concept in terms of how my pipeline works. It was built from the ground up for CYOA development, and I expect you'll see this and understand immediately what it means.
Here are a couple showing what I refer to as skeleton design, Motion scenes created in 3D software unreal engine in this case, That can later be reskinned by the ai layer, Collapsing large stacks of tasks into their most efficient form.
This one was an interactive test with choices at the end, Not much story but about 70 interlinked videos, Just using the base layer only.
Here's just a UE5 short we made. This one really shows how significant it is to have full independent control over cinematography and subjects.
It was the first successful test I did of the core tech working properly (sort of)
That video above shows how we can use AI to seamlessly combine live action footage and cgi sets in a single stylized output layer with total control.
Obviously you're already familiar with prompts for ai imagery, In this system these UE5 videos essentially are the prompt, Allowing unlimited communication of detail and refabrication for different purposes using text prompt modification and other bespoke control methods.
My interest isn't just in CYOA film making, but it is a part of my interest. I really enjoyed those types of books when I was a kid. I also enjoyed playing some of the older text based RPG's like Dangerous Dave. I also want to be able to create a continuous video series to be able to tell a much longer story that is developed based on viewer reaction. In order to do that I would need to be able to create an 18 minute video every week. For better or worse I've always been an avid day dreamer so I don't really have much issue coming up with a story to tell, my trouble starts when it comes to trying to take the story from my head to the screen. I've never used URE, but I downloaded and set it up yesterday to start learning. About 20 years ago I tinkered around with 3D Game Studio and made some VRML worlds. 3d modeling has come a long way since then. I still do some 3d modeling, mostly in tinkercad creating parts to print with my 3d printer. Last year I made a functional radio prop from the fallout video game series:
I think I still have a lot to learn before I will be able to create anything with URE, and I think I'll need to pickup another hard drive or two. I have a 5tb usb driver I use to archive my videos, but the data transfer rate is extremely slow. I think I still have a slot open where I can add another internal drive so I might see if I can get a 20tb to install internally. Right now I only have 2 1tb ssd's in the case, and they have filled up quick over the last year. I am interest in your project, if you could direct me somewhere to take a look at it. I tried you website, but it gave me a wix error. My website isn't in much better shape, I set it up in GCP as a static site ties to a git repo in cloud source using cloud build to deploy updates any time I push changes to the master branch. In hindsight I should have just deployed wordpress to a compute instance. But $0.25/day isn't a bad price for hosting.
Are there any of the NVIDIA industrial server cards that could be used? I figured they would handle the higher load a bit better then the typical commercial gaming cards. They also tend to have a lot more memory as even the old tesla k80 had 24gb of memory.
Indietalk probably has more information about my project than anywhere else at this point. For a few years I built out a discord server and we had about 50 people join and try to work with me on the project. I eventually just soft closed that server, simply because the vast majority of people that tried to join had some major issue or another that prevented them from being much use. A huge majority were musicians trying to find a place where their song could be heard. I have nothing negative to say about any of them and I understand their various strategies in approaching a market as chaotic as this. It's unfortunate in the current climate that there were a half dozen people who were intelligent, hard working and had a good attitude, but were simply a liability in terms of supply and demand issues. I'd spend 100 hours of time working with someone, and they would produce a full length symphonic score of the caliber that was at least on par with a mid-range television show, and it became a complete loss scenario as the value of something like that on the digital market crashed to less than a dime.
Keeping up the main website became at least a temporary liability, as it was attracting a number of people to the project who would eat up a lot of my time without helping me move the needle very much. In years 3 and 4, I've mostly just worked solo on what I thought was the most practical and useful aspect, which is the core automation pipeline. In any case, that's why you're not currently seeing a bunch of websites about this. I did at one time build out a relatively full featured website with extensive explanations of the project, its goals and methodology. Sometime this year I'll rebuild it.
Since it would take you an ungodly amount of time to cobble together a full understanding of the project through the hundreds of posts on indietalk, I'll just try to explain it as succinctly as possible right now. it's a useful exercise for me anyway.
Here are the core concepts.
Exactly as you say, to produce an entertainment product that allowed a somewhat exponentially increasing number of choices, one would need the capability of producing drastically more footage per week than any previous system was ever capable of. So from the beginning my strategy is based around extreme production speed. The other metric I feel strongly about is very simple. Given the choices available in the market, would anyone choose to watch this based on its quality moment to moment? For years I've looked at the work I was producing and said no, but this summer it's actually finally getting to the point where I'm looking at the demo material and saying, If this had a good story and enough scope, I would watch this if I saw someone else had made it. By the end of this year I believe that on a visual and production level the system will be competitive with some shows I see on streaming networks in the wild.
I'd reference the "River" demo As an example where neither the quality nor the story was good enough, But the production speed was. That was a demo where I produced 17 interactive cells fully animated in 4 K, In one seventy hour week.
The Save Point interactive fiction project is designed for many creators to work on simultaneously. It was originally intended as a modular system, Where various artists and directors can contribute to the growing tree, achieving universal coherency via a rule set and some style guides. As an example, a lot of people made Dungeons and Dragons campaigns back in the day, and you can see constant expansions of that universe that all fit together and made one gigantic product. With what will eventually become hundreds or even thousands of interlinked videos, It provides a natural synergy for all involved, consolidating advantages across all creators while allowing independent creation and monetization.
I'll spell out a practical scenario to make this easy to understand. 20 people each want to make interactive fiction film on youtube, Every one of which has to Create their own technology pipeline workflow asset library and hardware stations, Then publicize and grow the channel using their own resources. That's the old way. The new way I've Envision here would have all 20 people or groups work on the same interactive fiction story but build out different branches of it. A percentage of revenue across the entire system would be fed back into the central development pool, Which would help to supply a number of things to all the creators involved.
1. Advancement of the bespoke core technology platform, freely available in each iteration to all involved and constantly improved as time and resources are sunk into its development.
2. A shared asset library which is legally available to anyone working under the corporate umbrella. (Tons of this has already been accomplished)
To explain briefly, An enormous amount of time is spent licensing gathering organizing and curating assets of different types some of which are expensive. Once the save point project has either created these assets and owns them, Or legally licenses them for corporate use, All registered members of the team can use them freely under the terms that they are all used by one company for one project. This is an extremely valuable hack, Though for large corporations it's not considered a hack at all and is just traditional everyday business synergy.
So in practice I go on to the unreal engine marketplace, And on behalf of save point I purchase a $200 kit with several hundred models for building a metropolitan city. Those become part of the save point asset library, And can be freely lent to anyone working on the project. We only purchase assets that have licenses allowing any number of people to use the asset conditional upon its use in a single product. The Magic happens with the CYOA structure. We can have those 20 people come in and everyone can create their own storylines, But legally it is all viewed as a single product which is what most licenses specify as an allowed use case. This means that the number of creators attached to the project serves as a force multiplier for the assets purchased.
I'll give you something concrete here. Through licensed purchase contribution composition or AI curation, We've already amassed close to 3000 professional grade scoring tracks. Some are better than others but almost every one of those 3000 tracks meets production standards for broadcast. When a creator in the system wants to build a "cell", Which is what we call a single video with a choice, They have easy access to that entire music library, All of which is legally cleared with no effort or expense required. We've been doing the same thing for Foley.
I really wanted to get that point across, Because from a top down strategic perspective it becomes extremely significant.
3. A synergistic channel for advertising and revenue.
A project like this is difficult to make viral. It's a niche interest, But has real potential for the few million people who share that interest (viewers). I worked with large corporations on advertising campaigns in several roles over more than a decade and I've seen first hand how effective standard advertising is in comparison to Facebook posts word of mouth and many of the other strategies YouTube content creators tend to employ. It's not two or three times as effective, It's sometimes 10,000 times as effective. In example you might spend hundreds of hours of your time trying to evangelize about this thing you made, And the same result could be achieved with $5000 in advertising. If you attach any value at all to your time, and most smart people do, Then you can see where spending years of your life trying to equivocate 5 grand in Adwords spin doesn't really make any sense. By pooling Many CYOA plot lines into a single channel, It allows a sort of amplification of ad spend, With every subscriber gained becoming a subscriber of all creators in the system via the seamlessly earned videos that at least on the surface appear to tell one continuous story. For people who are very interested in finding out what's next, Finding a single video in the chain and following its links could conceivably lead them through the works of a dozen creators.
Because of the nature of choose your own adventure it also gives us a unique opportunity. The idea is to frequently create choices which steer viewers into story topics that they would find most interesting. Theoretically this should automatically sort viewers coming into the meta chain into plot lines that they would have the best experience with.
It's not a good example but I'm trying to do this quickly so I'll give you an ultra simplistic version of what that might look like. Let's say in the plot there is a time when you are acting as a spy and need to collect information on some faction. You are given a choice of several ways to infiltrate that faction. One is to get placed inside a corporate business environment within that faction, And another is to try a more direct route by befriending the ceos degenerate street racer son for direct access into the home of the ceo. On the surface that looks like a perfectly normal choice, But what we're really asking the audience is do you want to spend the next episode hobnobbing with underground street racers, Or performing corporate espionage. This way we can direct viewers towards scenarios they would find more interesting. Theoretically it should create a type of meritocracy for individual creators, Wherein designing plot lines that appeal to a broader audience would get them more individual revenue
4. Revenue sharing
The plan is for this to be done per video, Tasking AI to do the accounting work. With zero boots on the ground experience this is currently just set at a 5050 share of net revenue between the system and the creator. The 50% that goes to the system works out like this, 10% is our business revenue. 20% is Reinvestment into the shared assets and development pool, or global synergy pool, And the final 20% is always advertising to pull in new viewers to the entire chain. This ensures a steady stream of new assets, tech development, and new viewers for all involved.
I've done the math on how this all works out, And the viewership numbers to provide in example a middle class full time income for a creator will be very difficult to achieve. That has almost nothing to do with my system but rather the state of the digital world. In almost every case however, Individual profits will probably be much higher than they would be for people who had to handle all of these aspects themselves. In addition the majority of people are far more interested in telling a fun story than they are in building massive infrastructure. Personally I like building infrastructure, but that's just me.
5. Save Point
Save point is the name of the company and also its flagship product. It is intended to be a massive CYOA story that eventually spans hundreds of creators. In short it is the story of one man's life.
The metaplot we have created works like this. A man named Max lives in the current day. Events around him begin to take place that challenge his understanding of what the world around him actually is, Which is the central mystery of the entire plot. The end goal is to unravel the mystery of what the universe and existence is. The story is divided into three main acts.
Act 1. Breaking free.
Max lives a normal life in our world. At the beginning of the story He comes into possession of a strange blue crystal, The save point. This item allows him to travel a brief distance back in time allowing an unlimited number of opportunities to make different decisions at various junctures. You'll immediately recognize this as the core allure of interactive fiction. " What if I had........."
In any one of 100 ways, Max accomplishes his first goal which is essentially to achieve freedom here in the normal world via financial gain or other achievements that will eventually lead to the same conclusion.
Act 2. Stepping into a larger world
Once the player reaches an economic level through whatever path they took, He discovers that an extremely small group of the Ultra Rich have been aware of and traveling between other planets and galaxies for years, And it's at this point that the adventure expands onto a galactic scale. I'll stop here and point out that it doesn't ever actually have to, And that a player interested in living out Max's life as a terrestrial business tycoon can do so.
In this section of the story a grand variety of sci-fi adventures are possible, Broadly guided by a heroic struggle against a monopolistic galactic tyrant. It's intentionally generic to allow a broad canvas for individual creatives to paint their own stories within a genre.
Regardless of what track a player might take the fight seems impossible, Think Luke Skywalker versus the Empire. David and Goliath.
A successful run through this section eventually culminates in the player finding the magic bullet. Later in the travels of act 2 a character is encountered, Which character or situation depends on the plot line they are involved in, And this character tells Max about "a place where time ends"
Similar to the first act this circumstance only occurs after the player has achieved a particular plateau. Act one money act two power act three knowledge.
Act 3. Stars end
In the final leg of the fiction max has become an incredibly powerful figure, Still outgunned 1000 to 1, but in possession of unique knowledge that he believes will allow him to topple the evil empire that has enslaved the galaxy.
Like every section this can go many different ways, But the ultimate finale revolves around the reveal that life and the universe was actually much different than the way humanity perceived it. Ultimately this revelation provides Max with the Power he needs to defeat his opponent. Sort of.
For spoiler reasons I can't really go into how the entire story ends.
As a side note some people that I've talked to over the years get the impression from this that this is a CYOA story that only has one ending. That's far from the case, I'm only describing the one optimal ending that completes the meta plot line. If you made certain choices Max's story could end up being about a guy who got famous for wrestling alligators, Never found out that he could leave Earth at all, And ended when he got eaten by an alligator. It could be about Max following his aspirations as a business tyrant, And the story could end as he achieved market dominance. It's intended to be up to the viewer. In addition the scope of this project is so large that I can't really say in these early days what it might become. I only know that it's significant to the coherence of the overall product that I provide this direction and scaffolding early on so that it's not entirely aimless.
Lastly I'll briefly introduce myself and my qualifications. I was an AI developer starting in 1999, I transitioned into film and video work around 2007. I've produced several feature films, and over 900 videos for clients such as Hyundai national and IBM. I also provided hundreds of content pieces for the big digital signs along the vegas strip during conventions.
Feel free to ask any questions you may have.
Last edited:
You have an interesting starting point, but I think you are limiting yourself and potential viewer/creator engagement. You could for instance start off with a variety of characters encapsulating most archetypes not just max. Have them all be different, but connected in some way. This gives the viewer the ability to pick a char they most identify with. That by itself should increase the appeal to a much broader audience. When the user get's the blue crystal they could then choose the type of universe they want to play in. I also wouldn't define a predetermined sequence of events. Allow the various creator to determine what happens in their universe and with collaboration you could even have universe crossovers. In this way you could have a CYOA that doesn't have an overall end, just character endings. Certain character endings could also lead to the beginnings of new chars sort of picking up the torch for a new adventure with a new char without having to start over from the beginning. This allows for an ever expanding universe and a never ending adventure. It also would allow you to see which story types resonate the most with viewers leading to higher profitability. Additionally it makes it easy to pivot stories and plots to be more aligned with social changes over time. I believe society goes through popularity cycles based on current events and generational changes. Especially when it comes to science fiction. Scifi Plots typically expand on how current scientific discoveries could play out in the future which is why scifi films sometimes appear to have predicted the future when viewed years later. The news of today are the inspiration for the films of tomorrow.
I agree that word of mouth doesn't work today. It's much more effective paying for promotion, but I also learned a few tricks to grow a channel without much cost. In 2024 subscribers no longer matter. Only 3 things really make a difference.
1. CTR (percentage of viewers that click to watch your video vs swipe away).
2. Video retention time (how many viewers watch a video to the end).
3. Viewer session time (multiple video views via end screen element clicks).
CYOA series fit those criteria perfectly. An ad campaign on the start video would be all that is needed to kick off success one you have some videos ready to watch.
I'm curious of the type of ai you were working on back in 1999. The only things I remember about ai at the time was the psychotherapist found in the Linux Emacs editor among other chat bots, NPC programming, and the human genome project which utilized the google toolbar to create a large distributed computing system.
I got started as a web developer in 2001, and moved from there into smart phone app development and web 2.0 in 2012, and finally into big data processing in 2015. It was during the web 2.0 kick that I got into animation using the html5 canvas with javascript. I made a few small games and converted a couple from other systems. Most notably I made a conversion of the Circus Linux c/sdl game to a html5 canvas/ javascript chrome app. I was a bit disappointed when google removed the apps from the chrome browser and made the chromeOS only. Some of the apps I made were the only ones that actually worked offline. Most were just a frame around an existing website.
I agree that word of mouth doesn't work today. It's much more effective paying for promotion, but I also learned a few tricks to grow a channel without much cost. In 2024 subscribers no longer matter. Only 3 things really make a difference.
1. CTR (percentage of viewers that click to watch your video vs swipe away).
2. Video retention time (how many viewers watch a video to the end).
3. Viewer session time (multiple video views via end screen element clicks).
CYOA series fit those criteria perfectly. An ad campaign on the start video would be all that is needed to kick off success one you have some videos ready to watch.
I'm curious of the type of ai you were working on back in 1999. The only things I remember about ai at the time was the psychotherapist found in the Linux Emacs editor among other chat bots, NPC programming, and the human genome project which utilized the google toolbar to create a large distributed computing system.
I got started as a web developer in 2001, and moved from there into smart phone app development and web 2.0 in 2012, and finally into big data processing in 2015. It was during the web 2.0 kick that I got into animation using the html5 canvas with javascript. I made a few small games and converted a couple from other systems. Most notably I made a conversion of the Circus Linux c/sdl game to a html5 canvas/ javascript chrome app. I was a bit disappointed when google removed the apps from the chrome browser and made the chromeOS only. Some of the apps I made were the only ones that actually worked offline. Most were just a frame around an existing website.
Well, I do have answers to all of this.
in reverse order.
3. I built one of the first generative music AI programs back in 2000. It was relatively simple by today's standards. It was essentially an expert system that was given an understanding of how music theory and mixing worked, and then we created thousands of loops, such as a few bars of a drum beat, and tagged them with a variety of information. What tempo, what key, emotional intensity, style, etc. I think there were about 40 tags we made, and 30 or so of them were optional. The system would take some user input "make me a 90bpm blues song in G" and add a controllable amount of RNG noise, not dissimilar to the process for visual AIs today. It would take it a few hours to produce an album, and it would fool people when I played CDs for them in a car. The key to my design then, just as now, was intelligent hybriding. Knowing which tasks the computer should handle, and which it should not. Guitar licks had feeling, because a human guitarist played them. Think of it almost like an automated DAW.
2. Maybe I'm not communicating the story aspect well. It's completely open ended, and allows for almost everything you mentioned. There are multiple aspects to this that have little to do with old school traditional CYOA work, that require design parameters that might not make sense from a physical book perspective. Primary amongst these is generating cash flow. It's not my main goal, but experienced creatives know that going broke will destroy any scalable project, or prevent it from scaling. For the user to have a larger number of choices, it's significant that creatives not become homeless during the scaling process. That information is key to the final answer.
3. The very first concept for SP was to have a male and female main character. As we spent thousands of hours developing the project, one of the steps that was taken was to do the math, by creating simulations of users using the system, and estimating cash flow. I'll start by saying that even though, as you astutely noted, this design is ideal for the youtube algorithm, finances will basically never be good, unless it became unexpectedly popular. Expectations are really no more than 1 million hardcore OCD users at peak, and a few more million coming in and out at various times. That alone will be challenging to accomplish, but is likely possible. I call it a fail point when a person working full time on the system cannot earn a living wage. Let's call that 40k a year.
Here's how the multi character idea pans out in the simulations. The first character added reduces income across the board by half. The second reduces it to a third, and so on. We have a somewhat fixed number of people that will ever find this interesting, and instantly dividing them up means dividing the hits on each cell. I might make the same amount of money, or even 10% more, but everyone else on the chain would get screwed. That 40k annual income becomes 20k after I split into two main forks.
We've developed a modular solution to that. It's not 100% efficient, because there's no way that can happen, but it is more efficient than any other available path. It has to do with "episodic" content that takes place within the lifespan of the character. Unfortunately, the whole thing breaks down when the central character gets a haircut or a scar, much less a full identity shift. A single consistent main character is the primary supporting beam that keeps the entire system from collapsing. It means that high profile episodes can be reused in timelines where they have not occurred yet, as opposed to being skipped completely based on a single choice the user makes. It's a modular system that uses AI to plot what experiences are available during what acts, and in what sequence.
Obviously, I'd love to make it a free for all, which could offer millions of choices instead of thousands, but it's simply not practical to do pure exponential math, with human work hours being spent on many aspects of each cell. I also don't want to fully automate it, because I feel that without the human element on the creative end, the work will end up degenerating into a pastiche of formulaic scenarios. Anyway, it's not hard to work out how impossible this would be without some finessing of the 1980s formula. Just begin with a 2 and start doing exponents..... We did it, and I think without the system I created, it was about 2 hours into the story when every man woman and child on earth would need to work full time for the story to continue, and that's with heavy automation and an infinite resource supply. Now start your exponent chain with a 4 for two characters, and a 6 for 3.
I'll give a quick example of a module. It's kind of like a bottle episode, but wider in scope. In some cell you make a decision to take a boat out into the ocean. Could be to flee law enforcement, could be to go sailing, all depends on what plot zone you are coming from. During your sailing, the ship is caught by a storm and shipwrecked on an island. This island would form a somewhat self contained episode, perhaps a robinson crusoe type plot. It has a choke point at the beginning and end, which can be dynamically linked to the main plot or another episodic plot. Much like what is seen in Star Trek TNG. Same ship, same captain, but for this hour everything is about a conflict in the neutral zone. So for an hour of choices you survive, and try to escape the island. Might happen one way, might happen another, or you might die. Regardless of how you escape the island, or the prison, or the snowed in resort, etc, there is always a near universal adapter point where the client can be routed forwards into the larger story.
Popular storylines can be expanded indefinitely, so if we see that many people are trying to live out their lives as a stock broker or music star, we can expand them as makes sense based on analytics data. It would not make sense to create this type of system today and try to do it without using analytics to steer construction. We also have options to put ads on plotlines that are relevant to them, which is a major advantage. Steer your life towards music, and get ads for music gear. Should provide a strong boost to CTR.
In short, this looks and sounds, and even plays like the CYOA stories of old, but under the hood, what's going on is dramatically more complex than any previous attempt. Infocom stuff would be closer, but still not even close. This is something new, that looks deceptively familiar on its surface.
Last edited:
Your AI music generation sounds like it was pretty cool back in the day.
I guess I misunderstood your system.
However you have given me some insight into another potential path of development. In my case I wanted to be able to create both a series and CYOA videos. I think for myself a hybrid approach would be best. These are online videos not books after all.
So I'm thinking of creating a series where individuals don't get to make the choice in the CYOA, but instead at the end of each episode they are presented with the possible choices and they click to 'vote' on the next episode at the end always leaving out a hidden mystery choice.. This in the end creates a never ending, but linear story where the audience feels like they have some control.
Now one thing I have learned is to never give the user everything they want.
So that vote is really just for sentiment to see what most users want and how the audience is divided.
This gives the ability to use user emotion to increase engagement. remember the mystery option the users don't get to vote on.
If there seems to be either a close vote or an extremely one sided vote the mystery option is what the users get. In all other cases the majority wins on the next episode. This type of system reduces the overall work load on creators while increasing the likely hood of subscription and return visits. It also gives people something to talk "complain" about. It creates viewer competition and loyalty. Additional revenue can then be generated through fan products and side content such as live stream recaps and predictions. This also makes having several characters possible without having to deal with the butterfly effect. They don't all have to be the main lead, but could be the main focus in different episodes.
So the process would work sort of like this:
Release and episode with some public and 1 private choice.
viewers get to vote for 2 days before the vote is closed.
next episode is decided and created
next episode is release, end screen voting elements are replaced on previous episode with link to current episode.
repeat.
While not a true CYOA you don't have to deal with the butterfly effect because there isn't any real branching.
The story still unfolds based on user input, but there are guard rails and leavers that can be pulled to increase engagement if needed.
There is a session time trade off on this, but optimizing for YouTube content suggestions where a viewer would be presented with a random video in the middle of a non linear series doesn't make sense anyways. Where as if the suggested video was a full episode in a series it would be more likely to get a viewer hooked instead of confusing them.
Either way it doesn't make much difference for me as I can't yet create even a simple video with actors yet.
URE5 is a lot more complicated than many of the software I have used in the past, and I think it is going to take me a long time to get to where I can create a video worth watching with it. Let alone an episode each week. I'm hoping once I get it figured out I will be able to create a world to film in, and setup the actors a players so I don't have to do keyframed animation for every actor. Only for specific close up scenes and individual dialog. Being able to 'record' to keyframes and playback a player controlled char might be helpful.
I guess I misunderstood your system.
However you have given me some insight into another potential path of development. In my case I wanted to be able to create both a series and CYOA videos. I think for myself a hybrid approach would be best. These are online videos not books after all.
So I'm thinking of creating a series where individuals don't get to make the choice in the CYOA, but instead at the end of each episode they are presented with the possible choices and they click to 'vote' on the next episode at the end always leaving out a hidden mystery choice.. This in the end creates a never ending, but linear story where the audience feels like they have some control.
Now one thing I have learned is to never give the user everything they want.
So that vote is really just for sentiment to see what most users want and how the audience is divided.
This gives the ability to use user emotion to increase engagement. remember the mystery option the users don't get to vote on.
If there seems to be either a close vote or an extremely one sided vote the mystery option is what the users get. In all other cases the majority wins on the next episode. This type of system reduces the overall work load on creators while increasing the likely hood of subscription and return visits. It also gives people something to talk "complain" about. It creates viewer competition and loyalty. Additional revenue can then be generated through fan products and side content such as live stream recaps and predictions. This also makes having several characters possible without having to deal with the butterfly effect. They don't all have to be the main lead, but could be the main focus in different episodes.
So the process would work sort of like this:
Release and episode with some public and 1 private choice.
viewers get to vote for 2 days before the vote is closed.
next episode is decided and created
next episode is release, end screen voting elements are replaced on previous episode with link to current episode.
repeat.
While not a true CYOA you don't have to deal with the butterfly effect because there isn't any real branching.
The story still unfolds based on user input, but there are guard rails and leavers that can be pulled to increase engagement if needed.
There is a session time trade off on this, but optimizing for YouTube content suggestions where a viewer would be presented with a random video in the middle of a non linear series doesn't make sense anyways. Where as if the suggested video was a full episode in a series it would be more likely to get a viewer hooked instead of confusing them.
Either way it doesn't make much difference for me as I can't yet create even a simple video with actors yet.
URE5 is a lot more complicated than many of the software I have used in the past, and I think it is going to take me a long time to get to where I can create a video worth watching with it. Let alone an episode each week. I'm hoping once I get it figured out I will be able to create a world to film in, and setup the actors a players so I don't have to do keyframed animation for every actor. Only for specific close up scenes and individual dialog. Being able to 'record' to keyframes and playback a player controlled char might be helpful.
It's an interesting idea. The only major hole I could poke in it is that it feels like it would be a slow rollout. Depending on what you want to happen, it might be perfect, and I understand some of your strategy there.
My design is more geared towards people binging videos to follow a plot to conclusion. Get the hooks in and drag them through 60 videos in one night so they can find out who the murderer is, what's at the bottom of the caverns, whether or not they won the court case, etc. I'm thinking the money doesn't come from the casual viewer, but rather from the die hard people that are playing to win. I could see your concept working as well, it's just a different strategy.
UE5 isn't exactly easy, but it's possible. The big issue for filmmakers is that it's not built for filmmaking, though that's changing a lot here in V5, and a lot of studios use it now. The Mandalorian uses UE5 now, and that was 20 mil an episode. I started from scratch a few years ago, and made great progress, just watching an hour of tutorials a day and spending the rest of the time putting what I learned each day into practice.
Where it will really bog you down is in character animation, but if you persevere, anything is possible. Where the CYOA stuff becomes practical has to do with that video "The Key", which shows how the AI layer can be used to transform a basic template scene such as two humans walking down a hallway, into ANY scene of ANY two humans walking down ANY hallway, with relatively little additional work. I'm still working out the last few kinks, but I can tell you now that this is absolutely possible, even though it's unprecedented in film. Learn enough UE5 to show a courtroom a doorstep, and a police station, with a few metahumans, and then add my AI layer, and you can make a season of Law and Order for 10 grand. That's what it will be when everything works flawlessly.
My design is more geared towards people binging videos to follow a plot to conclusion. Get the hooks in and drag them through 60 videos in one night so they can find out who the murderer is, what's at the bottom of the caverns, whether or not they won the court case, etc. I'm thinking the money doesn't come from the casual viewer, but rather from the die hard people that are playing to win. I could see your concept working as well, it's just a different strategy.
UE5 isn't exactly easy, but it's possible. The big issue for filmmakers is that it's not built for filmmaking, though that's changing a lot here in V5, and a lot of studios use it now. The Mandalorian uses UE5 now, and that was 20 mil an episode. I started from scratch a few years ago, and made great progress, just watching an hour of tutorials a day and spending the rest of the time putting what I learned each day into practice.
Where it will really bog you down is in character animation, but if you persevere, anything is possible. Where the CYOA stuff becomes practical has to do with that video "The Key", which shows how the AI layer can be used to transform a basic template scene such as two humans walking down a hallway, into ANY scene of ANY two humans walking down ANY hallway, with relatively little additional work. I'm still working out the last few kinks, but I can tell you now that this is absolutely possible, even though it's unprecedented in film. Learn enough UE5 to show a courtroom a doorstep, and a police station, with a few metahumans, and then add my AI layer, and you can make a season of Law and Order for 10 grand. That's what it will be when everything works flawlessly.
I was trying to figure out if I could stage cameras in the level to record "game play" and use multiple players to act out the scene rather than staging all the animations. For detailed shots like conversations I might be able to use something like those camera tracking avatars and overlay the footage over recordings from ure. Blending human acting with the game scene. I think that would be the optimal route to capturing some semblance of realism. I don't mind acting in front of a green screen if I could be done to control a realistic avatar. That would really cut down on the production time for my concept. It would still take a team of people, but not nearly as many.
I am curious of how you achieved that effect with the hybrid cat. Is it the same as what you did with the key?
I was thinking about making the videos about 18 minutes long with a little bit of previous episode highlights in the beginning so new viewers would note that it's a series and got to the channel page to find episode 1 and binge from there. With every video ending on a cliff hanger do to the voting process I think people would binge until they got current then subscribe and vote. The regular upload schedule should cut down on the advertising cost to acquire new viewers as the YouTube algorithm would be able to figure out the target audience after just a few videos, and show new episodes to new potential viewers that have similar interest as past viewers.
I'll admit pushing out an 18 minute video every week will be a monumental challenge. Especially if I tally the votes for 2 full days, but I'm hoping if it took off I could get enough views and votes in the first few hours after release to know which way to proceed. I think I would want to have the first 6 episodes ready at launch which could be used for basic character introduction and world building before introducing the voting. There are a lot of great shows that have been produced on smalls sets with just a few actors. I think that is where I will have to start in order to be able to put out the episodes quickly enough. However as time goes on I could buildup virtual film locations.
What do you think is the best way to reskin a green screen actor? I think I saw one of your videos where you did something similar matching facial expression and lipsyncing the model to the audio. It looks like it worked pretty well, but the model was a bit twitchy like it was jerking back to a default expression when you weren't speaking. I tried some motion tracking webcam avatar software a while back, but the quality just wan't good enough to be usable for anything. The tracking kept getting messed up and the model would do all kinds of twitchy things. It also didn't like my glasses, it did a little better when I didn't wear them, but I also couldn't see anything.
I am curious of how you achieved that effect with the hybrid cat. Is it the same as what you did with the key?
I was thinking about making the videos about 18 minutes long with a little bit of previous episode highlights in the beginning so new viewers would note that it's a series and got to the channel page to find episode 1 and binge from there. With every video ending on a cliff hanger do to the voting process I think people would binge until they got current then subscribe and vote. The regular upload schedule should cut down on the advertising cost to acquire new viewers as the YouTube algorithm would be able to figure out the target audience after just a few videos, and show new episodes to new potential viewers that have similar interest as past viewers.
I'll admit pushing out an 18 minute video every week will be a monumental challenge. Especially if I tally the votes for 2 full days, but I'm hoping if it took off I could get enough views and votes in the first few hours after release to know which way to proceed. I think I would want to have the first 6 episodes ready at launch which could be used for basic character introduction and world building before introducing the voting. There are a lot of great shows that have been produced on smalls sets with just a few actors. I think that is where I will have to start in order to be able to put out the episodes quickly enough. However as time goes on I could buildup virtual film locations.
What do you think is the best way to reskin a green screen actor? I think I saw one of your videos where you did something similar matching facial expression and lipsyncing the model to the audio. It looks like it worked pretty well, but the model was a bit twitchy like it was jerking back to a default expression when you weren't speaking. I tried some motion tracking webcam avatar software a while back, but the quality just wan't good enough to be usable for anything. The tracking kept getting messed up and the model would do all kinds of twitchy things. It also didn't like my glasses, it did a little better when I didn't wear them, but I also couldn't see anything.
I think I'm getting a bit closer.
I tried out freemocap today with a single camera: https://freemocap.org/
I used this blender plugin to export the skeleton animation to be able to import into URE5: https://github.com/ajc27-git/freemocap_tools/releases/tag/FreeMoCap_Adapter_v1.5.0
I was then able to follow a tutorial for retargeting the animation onto a meta human in ure.
My results weren't spectacular, but I think if I were to do a multicam recording the results would have been better.
I think this will work well once I get a bit more experience with it as it will greatly reduce the time it takes to animate complex movements.
Being able to reskin myself to play multiple roles will also be handy.
I think if I talk during the initial freemocap recording I should be able to extract the audio and sync it with the exported video from ure in davinci resolve and not have too many lip sync or expression issues.
I can't wait to be able to create my first video using this, but I still have a lot to learn in URE.
I tried out freemocap today with a single camera: https://freemocap.org/
I used this blender plugin to export the skeleton animation to be able to import into URE5: https://github.com/ajc27-git/freemocap_tools/releases/tag/FreeMoCap_Adapter_v1.5.0
I was then able to follow a tutorial for retargeting the animation onto a meta human in ure.
My results weren't spectacular, but I think if I were to do a multicam recording the results would have been better.
I think this will work well once I get a bit more experience with it as it will greatly reduce the time it takes to animate complex movements.
Being able to reskin myself to play multiple roles will also be handy.
I think if I talk during the initial freemocap recording I should be able to extract the audio and sync it with the exported video from ure in davinci resolve and not have too many lip sync or expression issues.
I can't wait to be able to create my first video using this, but I still have a lot to learn in URE.