> I made a new kind of AI, how do I make money from it?
Well, presumably, this new form of AI was created because it makes doing something of value easier than it would be without it, so use it to do whatever that is, and either sell the proceeds to people who are interested in the outcome or sell access to the tool to people who are interested in producing the outcome.
Kind of hard to be more specific without information on what the "new form of AI" is or at least what it is for.
It's applicable in a large variety of cases, but I originally invented it for procedural generation of 2D video game levels. But after I created it, I began to see other (better) use cases. I tried to launch the tool I originally built, but during my beta with developers I knew, everyone was unwilling to try it because the UI was a CLI based domain specific language.
Sorry, from your post could not understand, what value could create your AI?
And yes, that is part of answer - how to make money, depend on what type of value it create. Next may need (may not), to create some product, which is similar to some existing but much better because of your AI value. This need to answer to question who will be client and how sales could argue him to buy product.
Ah it's okay! I'd edited the post to reduce details since I didn't want to force people to read so much. There's a number of ways it can make money.
1. In 2024, I built a demo of my internal tool for level design and tested building a level in Unreal 4, Source (2013) and Unreal with my tool. Source took 164 minutes, Unreal (alone) took about 122 minutes, but with my tool Unreal took about 23 minutes. A level developer's average salary at that time was $112,000. I projected 40% of their time on a block-out, and a 6x reduction (based on those numbers), which is a savings of $37,333.33 per-developer.
2. The heuristic maps have let me create a level the size of an old school game (Gameboy Color) in around the span of an afternoon. But because the implementations I'm using are simple (I only had time to build them over a few days), they don't have the full fidelity depth of such a map.
3. There's one demo I was building, but needed a full custom MIDI library for, that would let you style-transfer from a single song of sheet music onto an existing song's sheet music. I'd planned to try style-transferring Moonlight Sonata. I could see this being huge for musical creation, but the investors I talked to all mentioned that the music field is extremely niche and none of them understood it.
4. The newest technique I discovered was implanting it into a finite state machine lets it record how someone reacts to certain types of events in digital environments. But to demo this, I'd need a game, game engine, virtual developer environment, or a custom shell. Walking through the math on-paper though, it should be able to replicate basic developer responses in consoles similar to mitigating simple on-call issues.
I think all of these solve problems, but I've still struggled to get any eyes or interest.
Well, I want to be honest, and from what I see, gamedev is not in best shape now, I even seen some decline in number of hires on few markets.
And as I know, most gamedev companies already use AI to generate game assets, if not all.
If you planning to sell AI tools to gamedev industry, I think, main question, would be, how good your tool integrated into their pipeline.
What I mean, most I see, used some sort of AI generators, essentially making bitmap images. Sure, also used AI code generators and narrative (text) generators.
I seen demo of Unity official AI system, which include image generators, code generators and narrative (text) generators, and it is now provided as free beta, but announced paid tiers for heavy usage (you could google freemium model).
If you have some money, you could pay for cloud servers and provide some limited time service (for example 10 minutes per day), and gather money for additional usage.
If you have not money, but have good fantasy, you could yourself make game and sell it. I agree, 6x reduction of developer time is really significant for gamedev, but from my experience, when talked to tops, they answered "we will buy your solution just now, if it will give 10x improvement, but for 6x, we will note your address and may be phone you at beginning of next fiscal year").
What I want to say - sell service is much-much harder than sell material product, because for service you must first show it and somehow give client taste of using service, and once could try to demand money for usage.
Explanation is very simple - consumers used to return material product if not satisfied (with moneyback, sure), but they just cannot imagine, how to return service.
Thank you for the explanation. That makes sense, and I've experienced first-hand getting ghosted with a provable 6x, so I think you're right there too. I guess it's back to the drawing board for what I can build with it in the next 1-2 months.
You need a sales / product cofounder to be CEO. Create a compelling demo of it doing amazing things, then put all your energy into finding a cofounder. Startups are a team sport and you need help with customer development.
I have a tool that can up-scale heuristic images based on a training sample, and a smaller input image. But I've found the reception to that to be very lukewarm, maybe because technical proc-gen is quite niche?
I'd recommend putting together some side-by-side comparisons of your tool with a set of diverse input images against other popular upscalers like ESRGAN [1], ControlNet Tile approaches, SUPIR, DAT [2], etc.
> I have a tool that can up-scale heuristic images based on a training sample, and a smaller input image.
What, in this context, is a "heuristic image" as distinct from any other image and how is this different (in utility, not in mechanism) from the vast array of AI image upscaling tools that already exist? What does this do that you cannot do (or cannot do as well, cheaply, or easily) without it?
Explaining—and demonstrating—this would probably help gain traction.
In this context, an image where there's some data represented by each unique RGBA value, and less of the "smooth" complexities of normal images like from a camera. So ones with small color palettes. Depending on the output size, it can run very fast and I've used it to up-scale a simple 2-color map for an island in water off one 512x512 training image.
I tried to demo this to HN back in July, but my post back then was buried.
For that particular application, I think the key part is going to be explaining/demonstrating the advantage over non-AI approaches, especially given that the non-AI approaches are not patent-encumbered.
You are skipping a bunch of critical steps, like peer review, publishing your work, etc.
Ask yourself what sounds more plausible to your audience - that one individual, on their own, invented something that surpasses the entire professional and academic communities? Or that the same individual is having delusions of grandeur?
Now add in the fact that this person does not produce any demos, says that it would take months to integrate into something to show, and their "time is running low", so they are asking for people to "secure contracts" with them, without even being willing to show their work. Do such statements increase or decrease the believability of their claims?
I truly cannot say what you do or do not have. But I can say that it is unlikely anyone will pay attention to this unless you actually show your work.
I actually tried to show my work previously through my company website. I'm not an academic, so I can't even publish to arxiv right now. I have plenty of mathematics materials to prove what I've been working on though. Do you have any suggestions given that?
> I made a new kind of AI, how do I make money from it?
Well, presumably, this new form of AI was created because it makes doing something of value easier than it would be without it, so use it to do whatever that is, and either sell the proceeds to people who are interested in the outcome or sell access to the tool to people who are interested in producing the outcome.
Kind of hard to be more specific without information on what the "new form of AI" is or at least what it is for.
It's applicable in a large variety of cases, but I originally invented it for procedural generation of 2D video game levels. But after I created it, I began to see other (better) use cases. I tried to launch the tool I originally built, but during my beta with developers I knew, everyone was unwilling to try it because the UI was a CLI based domain specific language.
Sorry, from your post could not understand, what value could create your AI?
And yes, that is part of answer - how to make money, depend on what type of value it create. Next may need (may not), to create some product, which is similar to some existing but much better because of your AI value. This need to answer to question who will be client and how sales could argue him to buy product.
Ah it's okay! I'd edited the post to reduce details since I didn't want to force people to read so much. There's a number of ways it can make money.
1. In 2024, I built a demo of my internal tool for level design and tested building a level in Unreal 4, Source (2013) and Unreal with my tool. Source took 164 minutes, Unreal (alone) took about 122 minutes, but with my tool Unreal took about 23 minutes. A level developer's average salary at that time was $112,000. I projected 40% of their time on a block-out, and a 6x reduction (based on those numbers), which is a savings of $37,333.33 per-developer.
2. The heuristic maps have let me create a level the size of an old school game (Gameboy Color) in around the span of an afternoon. But because the implementations I'm using are simple (I only had time to build them over a few days), they don't have the full fidelity depth of such a map.
3. There's one demo I was building, but needed a full custom MIDI library for, that would let you style-transfer from a single song of sheet music onto an existing song's sheet music. I'd planned to try style-transferring Moonlight Sonata. I could see this being huge for musical creation, but the investors I talked to all mentioned that the music field is extremely niche and none of them understood it.
4. The newest technique I discovered was implanting it into a finite state machine lets it record how someone reacts to certain types of events in digital environments. But to demo this, I'd need a game, game engine, virtual developer environment, or a custom shell. Walking through the math on-paper though, it should be able to replicate basic developer responses in consoles similar to mitigating simple on-call issues.
I think all of these solve problems, but I've still struggled to get any eyes or interest.
Well, I want to be honest, and from what I see, gamedev is not in best shape now, I even seen some decline in number of hires on few markets.
And as I know, most gamedev companies already use AI to generate game assets, if not all.
If you planning to sell AI tools to gamedev industry, I think, main question, would be, how good your tool integrated into their pipeline.
What I mean, most I see, used some sort of AI generators, essentially making bitmap images. Sure, also used AI code generators and narrative (text) generators.
I seen demo of Unity official AI system, which include image generators, code generators and narrative (text) generators, and it is now provided as free beta, but announced paid tiers for heavy usage (you could google freemium model).
If you have some money, you could pay for cloud servers and provide some limited time service (for example 10 minutes per day), and gather money for additional usage.
If you have not money, but have good fantasy, you could yourself make game and sell it. I agree, 6x reduction of developer time is really significant for gamedev, but from my experience, when talked to tops, they answered "we will buy your solution just now, if it will give 10x improvement, but for 6x, we will note your address and may be phone you at beginning of next fiscal year").
What I want to say - sell service is much-much harder than sell material product, because for service you must first show it and somehow give client taste of using service, and once could try to demand money for usage.
Explanation is very simple - consumers used to return material product if not satisfied (with moneyback, sure), but they just cannot imagine, how to return service.
Thank you for the explanation. That makes sense, and I've experienced first-hand getting ghosted with a provable 6x, so I think you're right there too. I guess it's back to the drawing board for what I can build with it in the next 1-2 months.
You need a sales / product cofounder to be CEO. Create a compelling demo of it doing amazing things, then put all your energy into finding a cofounder. Startups are a team sport and you need help with customer development.
One way to get people to look at something is to show it to them rather than just describing it. What do you have that you could demo today?
I have a tool that can up-scale heuristic images based on a training sample, and a smaller input image. But I've found the reception to that to be very lukewarm, maybe because technical proc-gen is quite niche?
I'd recommend putting together some side-by-side comparisons of your tool with a set of diverse input images against other popular upscalers like ESRGAN [1], ControlNet Tile approaches, SUPIR, DAT [2], etc.
- [1] https://arxiv.org/abs/1809.00219
- [2] http://arxiv.org/abs/2308.03364
Thank you for the advice! Mine's mainly for limited palette upscaling, so that might be Apples to Oranges, but I'll give it some thought.
> I have a tool that can up-scale heuristic images based on a training sample, and a smaller input image.
What, in this context, is a "heuristic image" as distinct from any other image and how is this different (in utility, not in mechanism) from the vast array of AI image upscaling tools that already exist? What does this do that you cannot do (or cannot do as well, cheaply, or easily) without it?
Explaining—and demonstrating—this would probably help gain traction.
In this context, an image where there's some data represented by each unique RGBA value, and less of the "smooth" complexities of normal images like from a camera. So ones with small color palettes. Depending on the output size, it can run very fast and I've used it to up-scale a simple 2-color map for an island in water off one 512x512 training image.
I tried to demo this to HN back in July, but my post back then was buried.
For that particular application, I think the key part is going to be explaining/demonstrating the advantage over non-AI approaches, especially given that the non-AI approaches are not patent-encumbered.
That's a very good point. Thank you!
You are skipping a bunch of critical steps, like peer review, publishing your work, etc.
Ask yourself what sounds more plausible to your audience - that one individual, on their own, invented something that surpasses the entire professional and academic communities? Or that the same individual is having delusions of grandeur?
Now add in the fact that this person does not produce any demos, says that it would take months to integrate into something to show, and their "time is running low", so they are asking for people to "secure contracts" with them, without even being willing to show their work. Do such statements increase or decrease the believability of their claims?
I truly cannot say what you do or do not have. But I can say that it is unlikely anyone will pay attention to this unless you actually show your work.
I actually tried to show my work previously through my company website. I'm not an academic, so I can't even publish to arxiv right now. I have plenty of mathematics materials to prove what I've been working on though. Do you have any suggestions given that?