19 replies to this topic

[GaryInMotion]

A simple problem for probably most seasoned table developers, but one in which I was stumped. So, I asked ChatGPT, "Can you write code using the VPinballX program?"  The thoroughness and accuracy will, of course, be debated (as with most subjects), but my jaw dropped as I watched the answer form in real time.

 

 

For those of us ramping up (pun intended) in Visual Pinball coding, this is amazing!

 

garyinmotion

Edited by Pancake, 23 March 2023 - 05:27 PM.

[Gravy]

Yep, Ai is going to infiltrate many aspect of table creation I think.

I created a VR skybox a few weeks ago for use as a background image in VR.
https://www.vpforums...&showfile=17155

Ai is already being used to upscale poor quality table graphics.

Ai can easily be used to create new table graphics (not sure if any table designers are already doing this but it is very easy now).

I'd imagine at some point that that an Ai image model will be created specifically for coming up with new table layouts, trained on thousands of pinball machine layouts and based on the popularity of the entire layout or even just parts of individual table layouts.

ChatGPT is currently able to come up with unique new rule sets and scoring.

Here's a test I just did:

ChatGPT: Come up with a ruleset and scoring for a pinball table theme based around a caveman that travels to the future.
 

 

 

 

 

Sure, here's a possible ruleset and scoring system for a pinball table with a caveman traveling to the future theme:

Ruleset:

1. Launch the ball by pulling the plunger and aim for the skill shot lane to score points.
2. Hit the bumpers, rollovers, and targets to advance the bonus multiplier and increase the score.
3. Shoot the left ramp to collect letters and spell out "FUTURE" to activate the time travel mode and earn bonus points.
4. Shoot the right ramp to enter the future world and activate the multiball mode, where two or more balls are in play simultaneously.
5. Shoot the orbits to collect artifacts from the future world and increase the jackpot value.
6. Hit the drop targets to activate the caveman mode, where the player has to hit specific targets to score points and avoid dangers such as dinosaurs and lava.
7. Shoot the spinner to increase the spinner value and earn more points.
8. Complete the "Caveman Challenge" by completing a set of difficult shots, and earn a special bonus.
9. Avoid the outlanes and drain the ball as few times as possible to increase the final score.

Scoring:

1. Skill shot: 5,000 - 50,000 points, depending on the accuracy.
2. Bumpers, rollovers, and targets: 100 - 5,000 points each, depending on the location and difficulty.
3. Bonus multiplier: 2x - 10x, depending on the number of hits.
4. Time travel mode: 100,000 - 1,000,000 points, depending on the number of letters collected and the accuracy of the shots.
5. Multiball mode: 50,000 - 500,000 points, depending on the number of balls in play and the accuracy of the shots.
6. Artifacts: 10,000 - 100,000 points each, depending on the location and difficulty.
7. Caveman mode: 50,000 - 500,000 points, depending on the number of hits and the accuracy of the shots.
8. Spinner: 1,000 - 10,000 points per spin, depending on the speed and duration.
9. Caveman challenge: 1,000,000 - 10,000,000 points, depending on the difficulty and completion time.
10. Ball save: 10,000 - 100,000 points, depending on the duration.
11. Bonus: 10,000 - 100,000 points, depending on the bonus multiplier and the final score.

Of course, this is just one possible ruleset and scoring system for a pinball table with a caveman traveling to the future theme, and it can be modified and adjusted according to the designer's preferences and the players' feedback.

Edited by Gravy, 24 March 2023 - 12:37 AM.

[Tiki]

Whow. I am stunning...

[Mike DA Spike]

Whow. I am stunning...

lol

[wiesshund]

 

Yep, Ai is going to infiltrate many aspect of table creation I think.

I created a VR skybox a few weeks ago for use as a background image in VR.
https://www.vpforums...&showfile=17155

Ai is already being used to upscale poor quality table graphics.

Ai can easily be used to create new table graphics (not sure if any table designers are already doing this but it is very easy now).

I'd imagine at some point that that an Ai image model will be created specifically for coming up with new table layouts, trained on thousands of pinball machine layouts and based on the popularity of the entire layout or even just parts of individual table layouts.

ChatGPT is currently able to come up with unique new rule sets and scoring.

Here's a test I just did:

ChatGPT: Come up with a ruleset and scoring for a pinball table theme based around a caveman that travels to the future.
 

 

 

 

 

Sure, here's a possible ruleset and scoring system for a pinball table with a caveman traveling to the future theme:

Ruleset:

1. Launch the ball by pulling the plunger and aim for the skill shot lane to score points.
2. Hit the bumpers, rollovers, and targets to advance the bonus multiplier and increase the score.
3. Shoot the left ramp to collect letters and spell out "FUTURE" to activate the time travel mode and earn bonus points.
4. Shoot the right ramp to enter the future world and activate the multiball mode, where two or more balls are in play simultaneously.
5. Shoot the orbits to collect artifacts from the future world and increase the jackpot value.
6. Hit the drop targets to activate the caveman mode, where the player has to hit specific targets to score points and avoid dangers such as dinosaurs and lava.
7. Shoot the spinner to increase the spinner value and earn more points.
8. Complete the "Caveman Challenge" by completing a set of difficult shots, and earn a special bonus.
9. Avoid the outlanes and drain the ball as few times as possible to increase the final score.

Scoring:

1. Skill shot: 5,000 - 50,000 points, depending on the accuracy.
2. Bumpers, rollovers, and targets: 100 - 5,000 points each, depending on the location and difficulty.
3. Bonus multiplier: 2x - 10x, depending on the number of hits.
4. Time travel mode: 100,000 - 1,000,000 points, depending on the number of letters collected and the accuracy of the shots.
5. Multiball mode: 50,000 - 500,000 points, depending on the number of balls in play and the accuracy of the shots.
6. Artifacts: 10,000 - 100,000 points each, depending on the location and difficulty.
7. Caveman mode: 50,000 - 500,000 points, depending on the number of hits and the accuracy of the shots.
8. Spinner: 1,000 - 10,000 points per spin, depending on the speed and duration.
9. Caveman challenge: 1,000,000 - 10,000,000 points, depending on the difficulty and completion time.
10. Ball save: 10,000 - 100,000 points, depending on the duration.
11. Bonus: 10,000 - 100,000 points, depending on the bonus multiplier and the final score.

Of course, this is just one possible ruleset and scoring system for a pinball table with a caveman traveling to the future theme, and it can be modified and adjusted according to the designer's preferences and the players' feedback.

 

 

LOL AI plagiarism.
That suspiciously looks like a reworded ruleset for another machine.

i suspect it isnt actually writing a ruleset so much as it is looking at others.

[Gravy]

 

 

Yep, Ai is going to infiltrate many aspect of table creation I think.

I created a VR skybox a few weeks ago for use as a background image in VR.
https://www.vpforums...&showfile=17155

Ai is already being used to upscale poor quality table graphics.

Ai can easily be used to create new table graphics (not sure if any table designers are already doing this but it is very easy now).

I'd imagine at some point that that an Ai image model will be created specifically for coming up with new table layouts, trained on thousands of pinball machine layouts and based on the popularity of the entire layout or even just parts of individual table layouts.

ChatGPT is currently able to come up with unique new rule sets and scoring.

Here's a test I just did:

ChatGPT: Come up with a ruleset and scoring for a pinball table theme based around a caveman that travels to the future.
 

 

 

 

 

Sure, here's a possible ruleset and scoring system for a pinball table with a caveman traveling to the future theme:

Ruleset:

1. Launch the ball by pulling the plunger and aim for the skill shot lane to score points.
2. Hit the bumpers, rollovers, and targets to advance the bonus multiplier and increase the score.
3. Shoot the left ramp to collect letters and spell out "FUTURE" to activate the time travel mode and earn bonus points.
4. Shoot the right ramp to enter the future world and activate the multiball mode, where two or more balls are in play simultaneously.
5. Shoot the orbits to collect artifacts from the future world and increase the jackpot value.
6. Hit the drop targets to activate the caveman mode, where the player has to hit specific targets to score points and avoid dangers such as dinosaurs and lava.
7. Shoot the spinner to increase the spinner value and earn more points.
8. Complete the "Caveman Challenge" by completing a set of difficult shots, and earn a special bonus.
9. Avoid the outlanes and drain the ball as few times as possible to increase the final score.

Scoring:

1. Skill shot: 5,000 - 50,000 points, depending on the accuracy.
2. Bumpers, rollovers, and targets: 100 - 5,000 points each, depending on the location and difficulty.
3. Bonus multiplier: 2x - 10x, depending on the number of hits.
4. Time travel mode: 100,000 - 1,000,000 points, depending on the number of letters collected and the accuracy of the shots.
5. Multiball mode: 50,000 - 500,000 points, depending on the number of balls in play and the accuracy of the shots.
6. Artifacts: 10,000 - 100,000 points each, depending on the location and difficulty.
7. Caveman mode: 50,000 - 500,000 points, depending on the number of hits and the accuracy of the shots.
8. Spinner: 1,000 - 10,000 points per spin, depending on the speed and duration.
9. Caveman challenge: 1,000,000 - 10,000,000 points, depending on the difficulty and completion time.
10. Ball save: 10,000 - 100,000 points, depending on the duration.
11. Bonus: 10,000 - 100,000 points, depending on the bonus multiplier and the final score.

Of course, this is just one possible ruleset and scoring system for a pinball table with a caveman traveling to the future theme, and it can be modified and adjusted according to the designer's preferences and the players' feedback.

 

 

LOL AI plagiarism.
That suspiciously looks like a reworded ruleset for another machine.

i suspect it isnt actually writing a ruleset so much as it is looking at others.

 

It's generally pretty good at coming up with unique stuff, but obviously has to be trained on existing data so no doubt there will be similarities with existing table rules.

[toxie]

Also, always consider that we're at the very beginning of actually easily usable AI. Fast forward 2 or 3 years, and who knows, you may actually be able to generate a functional script from that ruleset above automatically. So with a bit of wiring work and some fixes here and there, everybody might be able to design a working ruleset.

 

Did somebody try to point ChatGPT to that github site with all the modded vbs scripts, so that it gets more context of VPX/VPM?

[wiesshund]

Also, always consider that we're at the very beginning of actually easily usable AI. Fast forward 2 or 3 years, and who knows, you may actually be able to generate a functional script from that ruleset above automatically. So with a bit of wiring work and some fixes here and there, everybody might be able to design a working ruleset.

 

Did somebody try to point ChatGPT to that github site with all the modded vbs scripts, so that it gets more context of VPX/VPM?

 

Dont let it pilot your space craft yet.

 

I tried to have to calculate an orbital escape trajectory from a certain planet (not earth cause it could just look it up from NASA)
The good news is the result was spectacular.

The bad news is it was spectacularly bad, LOL
Total fatality, among other things, it failed to factor in the binary star.

 

It will write you a book or research paper, but be careful, it does learn things but it doesnt quite grasp the concept of plagiarizing.

especially if you hint it too heavily down a particular path.

 

It has a long way to go before Skynet or the Red Queen.

 

On the other hand, i think if you gave it some rules and some context and content
And set it loose in a virtual world or some sort, with the ability to control and manipulate the world, it would make a formidable game master
though i am not sure that People would be up for such a game yet, not a game where there are no guarantees and you cant watch a video on how to accomplish anything
because the game master is learning, and it's primary function is to defeat you within the ruleset, but it is bound by no script.

It would take one ass kicking parallel server farm to run it most likely though.
Something that would make blizzard's WOW servers or Revant's EQ servers look like tinker toys.

[toxie]

Absolutely. I don't think it can or should be used for any reliable or scientific stuff (yet). But writing a VPX script.. .. Well, nobody will be harmed if something is wrong there. But it could be a great base to go from there and have the bot sort out all the 'boring' stuff initially.

[allknowing2012]

Toxie -- ChatGPT likes to pull in outside libraries. Any concerns on performance for the tables if use/abuse this? (pulling in lots of libraries)

[wiesshund]

Absolutely. I don't think it can or should be used for any reliable or scientific stuff (yet). But writing a VPX script.. .. Well, nobody will be harmed if something is wrong there. But it could be a great base to go from there and have the bot sort out all the 'boring' stuff initially.

 

Heh i would be curious to see what it comes up with, given enough data for it to tinker with.

It is kind of terrible at abstract concepts though (i mean it is AI after all)

And a full table is a very abstract thing.

But, perhaps if you limited it to just small pieces at one time.
Asked it to do specific interactions, and then have the human assemble them as a whole?

 

If you look at some of the real world tables, that were accomplished with the bare minimum of assets though, i think it will be a very long time before any AI is remotely capable of mimicking such imagination and ingenuity?
Silicon Computers can only see 1 or 0, the human computer can see infinite fractions of both in full analog

 

Still, i'd like to see what it came up with.
Given enough time, chance dictates it would have to make at least one table that is at least somewhat interesting?

[Hawyer]

oof

[GaryInMotion]

oof

Did AI help you with this comment?

[Gravy]

Thought this thread wouldn't be complete without a mention of Firepower Vs A.I. edition https://www.vpforums...&showfile=14051



 

Edited by Gravy, 27 March 2023 - 05:49 AM.

[toxie]

Very very hypothetically one might be even able to allow ChatGPT to drive/control the VPX editor now: Introduction - OpenAI API

(missing piece: if there would be an external VPX API to drive the internal UI/API  )

Toxie -- ChatGPT likes to pull in outside libraries. Any concerns on performance for the tables if use/abuse this? (pulling in lots of libraries)

This is one of the things that are in constant flux. Some outside/windows stuff is considered suspicious by the antivirus/malware scanner, thus it will be considerably slower, but this seems to change all the time.

Then, maybe even more important, the less external stuff is being used, the more likely it will be that a table can actually run in the upcoming standalone VPX player, as that one will NOT use the windows script execution.

[Gravy]

So I think I have asked this before, but are there many table designers using AI in their image creation yet aside from obvious stuff like upscaling?

I was just experimenting with playgroundai.com and their new inpainting/outpainting canvas, pretty impressive stuff.

I had a go at removing the score windows from this backglass below as a test. It's a bit like the "content aware fill" function in more recent Photoshop I guess. I was impressed that it picked up on the window reflection in the ball as I did not even know what it was meant to be behind the score window until I inpainted it. It's a little bit tedious at present as there is no undo, so if you want to redo something you have to mask the area out again, I'm sure this will change soon though.

Would be pretty good for visually repairing parts of a damaged playfield where image has peeled off.

 silverball-backglass-ai-edit.jpg   172.96KB   2 downloads

Edited by Gravy, 30 March 2023 - 02:45 AM.

[Manbearpig]

So I think I have asked this before, but are there many table designers using AI in their image creation yet aside from obvious stuff like upscaling?

I was just experimenting with playgroundai.com and their new inpainting/outpainting canvas, pretty impressive stuff.

I had a go at removing the score windows from this backglass below as a test. It's a bit like the "content aware fill" function in more recent Photoshop I guess. I was impressed that it picked up on the window reflection in the ball as I did not even know what it was meant to be behind the score window until I inpainted it. It's a little bit tedious at present as there is no undo, so if you want to redo something you have to mask the area out again, I'm sure this will change soon though.

Would be pretty good for visually repairing parts of a damaged playfield where image has peeled off.

silverball-backglass-ai-edit.jpg

This is awesome! I haven't tried doing anything with AI yet, but I would like to dive in. My PS skills are self taught and I think the AI stuff is a game changer. 
On this site do you whats the process like? Did you upload the image and select the score reel area to be removed?

[GaryInMotion]

So I think I have asked this before, but are there many table designers using AI in their image creation yet aside from obvious stuff like upscaling?

I was just experimenting with playgroundai.com and their new inpainting/outpainting canvas, pretty impressive stuff.

I had a go at removing the score windows from this backglass below as a test. It's a bit like the "content aware fill" function in more recent Photoshop I guess. I was impressed that it picked up on the window reflection in the ball as I did not even know what it was meant to be behind the score window until I inpainted it. It's a little bit tedious at present as there is no undo, so if you want to redo something you have to mask the area out again, I'm sure this will change soon though.

Would be pretty good for visually repairing parts of a damaged playfield where image has peeled off.

silverball-backglass-ai-edit.jpg

Like anything else, there is a point of diminishing returns. I've been using AI to restore a lot of images (not only for pinball) for a few years now. There's a threshold where everything starts to look "strange" and not organic.  That being said, I love the tools that are available and the level of range that allows for fine tuning. BTW, nice work on the Silverball Mania.  I did something similar for 1979 Bally KISS.

[Gravy]

 

So I think I have asked this before, but are there many table designers using AI in their image creation yet aside from obvious stuff like upscaling?

I was just experimenting with playgroundai.com and their new inpainting/outpainting canvas, pretty impressive stuff.

I had a go at removing the score windows from this backglass below as a test. It's a bit like the "content aware fill" function in more recent Photoshop I guess. I was impressed that it picked up on the window reflection in the ball as I did not even know what it was meant to be behind the score window until I inpainted it. It's a little bit tedious at present as there is no undo, so if you want to redo something you have to mask the area out again, I'm sure this will change soon though.

Would be pretty good for visually repairing parts of a damaged playfield where image has peeled off.

silverball-backglass-ai-edit.jpg

This is awesome! I haven't tried doing anything with AI yet, but I would like to dive in. My PS skills are self taught and I think the AI stuff is a game changer. 
On this site do you whats the process like? Did you upload the image and select the score reel area to be removed?

 

If you haven't really looked at much Ai art generation yet you are probably going to be a bit gobsmacked by this instructional video for playgroundai canvas feature (and what the site can produce in general). Outpainting and inpainting are pretty amazing. Playgroundai.com is a pretty good free alternative to midjourney at the moment.

It's biggest issue with the new canvas feature is lack of undo which means when you mask out areas you need to remask again if you dont like the results.

Basically for the silverball backglass edit I simply used the eraser tool to delete the area I want to edit (the score windows) and then make a reasonably large outer border around it so it has plenty of material to work with. Then I simply entered "backglass" in the prompt window as you cant currently use no prompt. I doubt if the prompt is making much difference in this case as it's likely just using the surrounding features to work out what to infill but I had to put something in there. It took a few goes before I was reasonably happy with each section that I edited and still not perfect. I should have started with a better scan with less reflections but is was only an experiment.



 

 

So I think I have asked this before, but are there many table designers using AI in their image creation yet aside from obvious stuff like upscaling?

I was just experimenting with playgroundai.com and their new inpainting/outpainting canvas, pretty impressive stuff.

I had a go at removing the score windows from this backglass below as a test. It's a bit like the "content aware fill" function in more recent Photoshop I guess. I was impressed that it picked up on the window reflection in the ball as I did not even know what it was meant to be behind the score window until I inpainted it. It's a little bit tedious at present as there is no undo, so if you want to redo something you have to mask the area out again, I'm sure this will change soon though.

Would be pretty good for visually repairing parts of a damaged playfield where image has peeled off.

silverball-backglass-ai-edit.jpg

Like anything else, there is a point of diminishing returns. I've been using AI to restore a lot of images (not only for pinball) for a few years now. There's a threshold where everything starts to look "strange" and not organic.  That being said, I love the tools that are available and the level of range that allows for fine tuning. BTW, nice work on the Silverball Mania.  I did something similar for 1979 Bally KISS.

 

It has certainly come leaps in bounds in the last year, outpainting and inpainting features make Photoshop's content aware fill seem like last century stuff now, and the image generation from scratch or from a sample photo, sketch, pose etc are like something ported in from the future, now things are moving into video generation and 3D asset generation. Check out Runway Gen2 if you havent already https://research.runwayml.com/gen2

 

[Manbearpig]

 

 

So I think I have asked this before, but are there many table designers using AI in their image creation yet aside from obvious stuff like upscaling?

I was just experimenting with playgroundai.com and their new inpainting/outpainting canvas, pretty impressive stuff.

I had a go at removing the score windows from this backglass below as a test. It's a bit like the "content aware fill" function in more recent Photoshop I guess. I was impressed that it picked up on the window reflection in the ball as I did not even know what it was meant to be behind the score window until I inpainted it. It's a little bit tedious at present as there is no undo, so if you want to redo something you have to mask the area out again, I'm sure this will change soon though.

Would be pretty good for visually repairing parts of a damaged playfield where image has peeled off.

silverball-backglass-ai-edit.jpg

This is awesome! I haven't tried doing anything with AI yet, but I would like to dive in. My PS skills are self taught and I think the AI stuff is a game changer. 
On this site do you whats the process like? Did you upload the image and select the score reel area to be removed?

 

If you haven't really looked at much Ai art generation yet you are probably going to be a bit gobsmacked by this instructional video for playgroundai canvas feature (and what the site can produce in general). Outpainting and inpainting are pretty amazing. Playgroundai.com is a pretty good free alternative to midjourney at the moment.

It's biggest issue with the new canvas feature is lack of undo which means when you mask out areas you need to remask again if you dont like the results.

Basically for the silverball backglass edit I simply used the eraser tool to delete the area I want to edit (the score windows) and then make a reasonably large outer border around it so it has plenty of material to work with. Then I simply entered "backglass" in the prompt window as you cant currently use no prompt. I doubt if the prompt is making much difference in this case as it's likely just using the surrounding features to work out what to infill but I had to put something in there. It took a few goes before I was reasonably happy with each section that I edited and still not perfect. I should have started with a better scan with less reflections but is was only an experiment.



 

 

 

Thanks!  I've watched a bunch of stuff on Stable Diffusion, but I don't know if my main PC can handle it. Funny how my VPin cab is more powerful than my daily driver. Also the output is 512x512, but can be scaled with more add ons. I have seen a few sites and tried some sample stuff. I'll watch this video and see if i can learn a little about the playground. I love making the full DMD 3rd scree like hauntfreaks. He adds quality art and thats also what I try to do. But self taught PS is pretty weak. I say your mention of "content aware fill" and i have no clue on what that feature even is. 

Thanks for the reply. off to research!