Gran Turismo 7 Custom Race thread

[suomi1]

OK, SPIRIT OF RACING is ready for sharing.

I don't want to post the link for the open public so DM me if you want a copy. Even @Dave A can have one

@Sir Crashalot @suomi1 anyone who is really interested in making the custom race feature come to life. I only ask that it's people with at least 1000 posts, and I am going to limit availability at that.

Most features will work without an API key form Anthropic, but if you want all the features, you will have to go buy a key. I think it's worth it. I'm loving the emails.


Like I said, DM me and I will send you a link

What is that and why would I need it ? Custom race is how PD wanted it to be and we have no way to change it.
If it's a tool to give ideas to create new custom races, I think my imagination combined to my interest in some categories of cars are far than enough to give me the fun I need when needed. Grinding to buy cars will always be more time consuming than having ideas. And if I'm not mistaken, such tools already exist.

Now, if it adds a mid race save feature, an expanded starting grid, and an option that equalizes tyres and fuel consumption between cars, why not. But I fear only PD can do that.

Using a rabbit car 100% eliminates rubber banding.

Should we have to do that? No. Does it make the racing better? Yes.

To be more precise, it holds rubber banding in a fixed position, the slowest one. So technically it's still here, and we lose one spot on an already small grid of cars. It remains indeed the most effective solution.

 

[bliprunner]

I think it's interesting that the Power Pack Sophy AI doesn't have rubberbanding, at least I have not seen any hard evidence of it.

You can sneak away in P1 and Sophy will not increase the pace to stay on your bumper.

So clearly PD think it's fine to have Sophy AI without rubberbanding, the option could come for custom races... one day... 🤞

 

[Voodoovaj]

What is that and why would I need it ?

I may as well put the whole idea here for everyone. There is no direct connection to the PS5, because there's currently no sharing of this information.

It's a way to track race results, so instead of custom races being a one off experience, this gives them direction.

You give it your custom race set up (drivers, teams, tire wear, fuel usage, and race distance). You then given it all the tracks you want to use. Then, you weight the teams best teams get 1, worst teams get 10. It then uses this information to guide each event and season.

For every round, it gives you a grid order based on the team tier, the previous race performance of the drivers, and the total team score. This change in starting order affects how each driver behaves. So, as AI drivers finish better, their rank changes, and the qualify better.

You then give it the race results, which in turn drives the grid for the next round. I also built in a relationship feature where the drivers attitude toward you changes over the season, so they become allies or rivals. This feature works without any additional AI, but it's better with it.

Then (totally optional) you can add an Anthropic AI API key. Then, for every round, your team manger gives you a goal for the race (finish P8, finish P10, finish P1, etc). Your results then drive the communication from the team manager, sponsors, manufacturers, you personal assistance (for paddock gossip), and the above mentioned drivers. How you perform changes everyone's attitude toward you. It's the career mode that many have been asking for.

I'll share more screenshots today so people can get a better idea of it.

It all runs in a single HTML file in a browser window

Edit - I'll make a website for it

 

[DS13_supergt]

I wanted to ask a thing to custom race experts...
I'd like to do a 3 hour long 24h of the Nürburgring, is it better to use normal ai or sophy? Are they decent with strategies?

 

[Voodoovaj]

I wanted to ask a thing to custom race experts...
I'd like to do a 3 hour long 24h of the Nürburgring, is it better to use normal ai or sophy? Are they decent with strategies?

Depends on the car - Sophy doesn't support some cars.

 

[DS13_supergt]

Depends on the car - Sophy doesn't support some cars.

Gr3: amg gt3 2016, r8 lms 2015, M6 2016.

 

[Voodoovaj]

Gr3: amg gt3 2016, r8 lms 2015, M6 2016.

Then yes, I would go with Sophy





On a separate note, I have posted the race manager on a public site.

https://web-page-builder--ajmarini.replit.app/ BTW, this is only live for 1 month because I am still working on a final product. If you like it, I can send you the file directly and you can keep it locally. Since it uses your browser memory, you won't lose anything, but there is an import/export function to move save data (including API key if you have it)

There is a manual when you enter the site that explains everything. If you want to use a rabbit for your races, I suggest you set them up as the best team in your series so they will qualify at the front.

Consider this a prototype of what a single player campaign could be. This could reside as a separate career path in GT. There's often a request for custom championships, so this is as close as we can get.

 

[DS13_supergt]

Then yes, I would go with Sophy





On a separate note, I have posted the race manager on a public site.

https://web-page-builder--ajmarini.replit.app/

There is a manual when you enter the site that explains everything. If you want to use a rabbit for your races, I suggest you set them up as the best team in your series so they will qualify at the front.

Consider this a prototype of what a single player campaign could be. This could reside as a separate career path in GT. There's often a request for custom championships, so this is as close as we can get.

Very nice, thank you!

 

[suomi1]

I may as well put the whole idea here for everyone. There is no direct connection to the PS5, because there's currently no sharing of this information.

It's a way to track race results, so instead of custom races being a one off experience, this gives them direction.

You give it your custom race set up (drivers, teams, tire wear, fuel usage, and race distance). You then given it all the tracks you want to use. Then, you weight the teams best teams get 1, worst teams get 10. It then uses this information to guide each event and season.

For every round, it gives you a grid order based on the team tier, the previous race performance of the drivers, and the total team score. This change in starting order affects how each driver behaves. So, as AI drivers finish better, their rank changes, and the qualify better.

You then give it the race results, which in turn drives the grid for the next round. I also built in a relationship feature where the drivers attitude toward you changes over the season, so they become allies or rivals. This feature works without any additional AI, but it's better with it.

Then (totally optional) you can add an Anthropic AI API key. Then, for every round, your team manger gives you a goal for the race (finish P8, finish P10, finish P1, etc). Your results then drive the communication from the team manager, sponsors, manufacturers, you personal assistance (for paddock gossip), and the above mentioned drivers. How you perform changes everyone's attitude toward you. It's the career mode that many have been asking for.

I'll share more screenshots today so people can get a better idea of it.

It all runs in a single HTML file in a browser window

Edit - I'll make a website for it

Then yes, I would go with Sophy





On a separate note, I have posted the race manager on a public site.

https://web-page-builder--ajmarini.replit.app/ BTW, this is only live for 1 month because I am still working on a final product. If you like it, I can send you the file directly and you can keep it locally. Since it uses your browser memory, you won't lose anything, but there is an import/export function to move save data (including API key if you have it)

There is a manual when you enter the site that explains everything. If you want to use a rabbit for your races, I suggest you set them up as the best team in your series so they will qualify at the front.

Consider this a prototype of what a single player campaign could be. This could reside as a separate career path in GT. There's often a request for custom championships, so this is as close as we can get.

Well, it probably gonna be interesting for some but not for me, sorry.
I've always kept a notebook and a pen by my side when I'm reading or playing ; I don't need yet another gadget.

This is also the same reason why, for me, adding a championship mode to custom races is not THE priority.

 

[jesse100190]

had a early race rain last night with the track getting drier as the race went on

 

[Voodoovaj]

To expand upon what I am doing, and encourage people to try the manager maybe I can expand a bit more on what it does. Remember, I don't make money off of this. It's a prototype to show how LLM's can/should be integrated into games to make the experience better for everyone. HOPEFULLY, it happens in GT8 (and there is no reason why it can't)

First up, the qualifying. We focus so much on running custom races and our own performance, that mixing the grid is often ignored. That's a mistake, because the AI does change based on current grid position. Mixing up the grid changes the experience.

In the image below, the drivers are ordered in their starting grid position. The numbers to the right is the finishing position. I happened to fill in the blanks all the way to 12th position, but tracking points scoring position is what is most important.

Look at U. Katayama!! Started last, finished 6th. M. Salo started 18th and finished 9th. Great day for team Tyrell. Focusing on ourselves missed the great stories in the race, so I track them, and then the LLM reports them back to me. BTW, the full names displayed in the article below come from the LLM. Because of the year (1996) the context (F1) and the name similarity, the LLM assumes the drivers are the same. So, I never list "Ukyo" Katayama, or "Jean" Alesi. The LLM searches for the context and delivers that info.

You will also note the weather in the top left of the grid photo (Heavy Rain). I give it as a suggestion to the player (it's all suggestions after all) but the LLM will report on the race with that weather in mind. I also think it helps us to avoid choosing the same weather types.

What I am focusing on now is the "goals" for the races. In the version I put up for public use, the goals are pretty simple "Finish P2, P3, P4" etc, depending on where you are in the points and what should be achievable. What I am working on is the goal mechanic from Super Monaco GP, where you develop a rivalry, and your goal is to beat your rivals.

The Red angry faces below are rivals. The beginning of this is in the public version, but I am a few iterations ahead already.

This stuff, all meta game that cannot change the actual game, changes the experience significantly. Implementing this into GT8 would be trivial, but imagine how much more could be baked in if there were actual race data feeding into it. Like tracking who hit you and adding that to the experience, or the opposite case of feeding this information back to the AI to affect it's behavior.

For instance, Coulthard and Hakkinen (above) are both at the angry end of the scale. That's because I have been pushing Footwork ahead of McLaren in the constructor championship. Wouldn't it be a great feature to have those two cars racing me more than the others? Have those two truly race me to the end!! It's 100% possible, and straightforward to implement. All the necessary pieces are there, it just needs the triggers.

 

[Voodoovaj]

Here is an example of emergent gameplay.

Let me set the scene - The championship is very close. Hill was 2 points ahead of me. I needed to finish in front of Hill. A win would be great, but I needed to outscore Hill. Meanwhile, Williams needs me to get no score, and for Hill to get a win to lock up the title before the last race.

With that in mind, watch this (wait for the end)


No script, because it's impossible. GT7 knows absolutely nothing of my championship. This is an emergent occurrence when the context is applied to the way Reggie works. I managed to get back to 10th, so Hill scored 11 points more than me.

Now, going into the last race, I have to win, and make sure Hill scores no more than 1 point.

 

[05XR8]

Need more SOPHY.

 

[DS13_supergt]

So I was trying a race at the Nürburgring and when it starts raining they stick with dry tires... Why? The wet tyres are purchased.

 

[05XR8]

So I was trying a race at the Nürburgring and when it starts raining they stick with dry tires... Why? The wet tyres are purchased.

The AI will switch to wet weather tires on their own. No need to purchase them.

 

[DS13_supergt]

The AI will switch to wet weather tires on their own. No need to purchase them.

Yh but they aren't switching, they are sticking to hard tyres even if the track is fully wet...

 

[Voodoovaj]

Yh but they aren't switching, they are sticking to hard tyres even if the track is fully wet...

Racing hard? Or street hard? They'll stay on street tires because those work in the rain. The wet weather tires are for use with the racing tires.



Announcement - I have placed a download for the race manager on my website (the link is in my signature). Scroll down, click the link, and it downloads the HTML file, which then runs directly from your dervice.

This version has a distinct separation between the LLM powered stuff, and the functionality that runs locally. Now, you don't need the anthropic API key if you don't care about the messages from your crew chief, team manager, assistant, etc. If all you want is to track points, this version does that.

This version will also determine rivals for you (without the LLM), and give you rivals to beat in a particular race. "Why do I need to beat a rival?" - I am glad you asked. So, if you start last for example, as many of us do, it will set out a target driver to get past, rather than "win the race". Stretching a points gap to that rival will give you new rivals. This doesn't work if you are only running one race, because it needs race finish information to create the rivals.

I am working on making the LLM usage more efficient so, if you do choose to use it (and I recommend it), then you can get the most bang for your buck. Also, this version only hits the LLM API if you choose to check messages, which is never a hard gate requirement. I check messages after 2-3 races and in between seasons. You get offers from different teams to help you with your car decision.

For instance, I finished 2nd at the end of my 1996 F1 season and I got a sweet offer...

I have to say, when you add a proper single player to the AI, either Reggie or Sophy, the whole AI system of the game makes WAAAAAY more sense. It's like they thought about a proper single player campaign, but abandoned it.

 

[Voodoovaj]

Update on the Spirit of Racing manager (btw, I need beta testers, so feel free to grab it and give me feedback)

The non-AI bits seem solid, so if you want to just track seasons, get a starting grid, etc, it's good to go.

I am focusing on the AI/LLM parts not. What I am shooting for is a "drive to survive" feel to the drama that occurs off track. This gives you (the player) more context around the racing. Here's some examples.

Drama/Gossip from other teams.

Drama and flavor for your own team (I am with Footwork)

Hopefully Sony decides to push PD into making an actual single player component to the game. I always argue that there isn't one, really, because all the player does it a bunch of one off races. Even the few "championships" are barely more than a bunch of one off races with no real impact on the game.

Now, it can be a never ending story. Also, I built in non-linearity. You can go from 70's LeMans racing to something brand new, and it doesn't matter. The story just takes it in stride and rolls with it. (Note, again, this requires an anthropic API key that costs $5USD)

 

[equalitus]

No offense, but you don't realize what you are asking for. You are thinking of the AI like people. No racing AI can work as you describe. NO game has ever implemented that, and there's a good reason for it, it's boring and way more annoying.

You're going to try to fit 10 million players into 4 or 5 tiers, none of which will be interesting to anyone.

Trust me (and no disrespect meant) if you are a solid B when racing with others, then the AI is slowing down for you. You want that rubber band in there. The AI is slowing down for almost all of us. No one complains about it, because few notice it. It's the rapid recovery we hate.


On another note, my team manager (from my app) told me to finish 4th. I was closing in on a podium and realized that I woudl be 1 lap short on fuel. The series leaders seemed to be dropping back from me, and I realized that they were conserving fuel, while I was burning it. Here is the last lap. I REALLY hoped the series leaders would pit, but no luck. Still, the race was on for 4th. Big mistake by Sophy at the Monza chicane made it interesting.



I'm finding that 3x Fuel and 5X tires is the sweet spot for these sprint races 10-15 laps.

You are wrong about rubber banding because GT3&GT4&GT5&GT6 had no rubber banding, or perhaps marginally rubber banding compared to GT7. While the regular AI could be too slow for the average player, after AI Sophy there's no need for marginal rubber banding whatsoever! The AI should drive faster or slower only based on power, weight, tire, downforce, not rubber banding presets which are very boring.

 

[Theufcveteran]

You are wrong about rubber banding because GT3&GT4&GT5&GT6 had no rubber banding, or perhaps marginally rubber banding compared to GT7. While the regular AI could be too slow for the average player, after AI Sophy there's no need for marginal rubber banding whatsoever! The AI should drive faster or slower only based on power, weight, tire, downforce, not rubber banding presets which are very boring.

Just wanted to mention, GT3 had crazy rubberbanding. If they AI pit, they would catch up even if you were half a lap ahead after they came out the pits. I'm not sure if they slowed down when they were ahead though.

 

[benjoi84]

The AI should drive faster or slower only based on power, weight, tire, downforce, not rubber banding presets which are very boring.

Wading into this discussion based on my zero understanding of the content, but isn't what you're asking for here just "the car with the best specs will always win"?

 

[Nebuc72]

Wading into this discussion based on my zero understanding of the content, but isn't what you're asking for here just "the car with the best specs will always win"?

There should always be variance in performance corner to corner but people just don't want that variance to be power/grip modifiers based on distance to the player to keep them artificially close. That's what the rubber banding is.

 

[Voodoovaj]

.... after AI Sophy there's no need for marginal rubber banding whatsoever! The AI should drive faster or slower only based on power, weight, tire, downforce, not rubber banding presets which are very boring.

They degree of hubris in this is astounding. 😒

Sophy can be perfect. It can out drive any human. With that in mind, how would you defeat a field of 19 of them driving at 100% in equal machinery? You can't. No one can.

Ergo, they have to monitor your pace and stay more or less at your pace (what a rubber band actually is). I feel I need to emphasize PACE - It should maintain a pace. It can only maintain this pace by having knowledge of your pace and adjusting to it.

What you guys are calling a rubber band is NPC cheating. Assigning more capacity to the NPC than is available to the player in order to to reduce the time required for the NPC to close in on the player if the player gets too far ahead. No one likes NPC cheating. No one is supporting NPC cheating. It is an inelegant solution to making single player interesting.

 

[Bojador]

I think that regardless of what we call it, what we want, at least I do, is to be less obvious. To be believable. I want the AI to have the same constraints I have. I don't want their car to be able to get an extra 20kmph of their car top speed just because they are falling behind. I don't want them to have impossible grip.

I don't have an issue with AI being fast or slower to make for a good race. Even irl you see drivers adjusting their pace to not take unnecessary risks if they are leading or to take more risks if they need to catch up. What you don't see is them breaking the laws of physics.

 

[benjoi84]

There should always be variance in performance corner to corner but people just don't want that variance to be power/grip modifiers based on distance to the player to keep them artificially close. That's what the rubber banding is.

I understand that is the premise here, but what amounts to "variance" in motor sports is essentially the number of mistakes made by the driver, however small.

So isn't the alternative to force the CPU make a greater number than zero mistakes, in order to mimic human limitations? Do we accept the possibility they sometimes won't make any?

I'm genuinely curious about this discussion.

 

[05XR8]

I think that regardless of what we call it, what we want, at least I do, is to be less obvious. To be believable. I want the AI to have the same constraints I have. I don't want their car to be able to get an extra 20kmph of their car top speed just because they are falling behind. I don't want them to have impossible grip.

I don't have an issue with AI being fast or slower to make for a good race. Even irl you see drivers adjusting their pace to not take unnecessary risks if they are leading or to take more risks if they need to catch up. What you don't see is them breaking the laws of physics.

For sure. Makes me laugh when the AI do like Herbie and go bananas to catch up.
Plus, a big thing(yes even though this is a game) some players forget about real life racing, there are indeed back markers. From grassroots to F1!
There are even drivers that do pull out lengths over an entire field and lap cars.
Again, yes this is a game, but in trying to simulate real life racing, it'll depend how much of that experience PD are trying to balance.

I have races where it is like touring cars and NASCAR, where fields are bunched up. I'm doing my Gran Turismo 1 grids with some minor modifications to the cars. SOPHY keeps things tight and Reggie has its moments of goodness.

For four plus years many of us have been saying we shouldn't have to do a lot of work to get the race to where many of us would like. I guess that's the best balance PD can do. Give us the tools and we play how we play.

 

[Bojador]

I am happy with the challenge that Sophy is. I want to be able to choose tyres and not having to setup a rabbit, and they could do a better job with pitstops, but it is fun and believable.

 

[Aedum]

I am happy with the challenge that Sophy is. I want to be able to choose tyres and not having to setup a rabbit, and they could do a better job with pitstops, but it is fun and believable.

Yeah it is nice to have something to work with thanks to the contributors of this forum.

 

[pritchyTim]

I understand that is the premise here, but what amounts to "variance" in motor sports is essentially the number of mistakes made by the driver, however small.

So isn't the alternative to force the CPU make a greater number than zero mistakes, in order to mimic human limitations? Do we accept the possibility they sometimes won't make any?

I'm genuinely curious about this discussion.

Maybe a ‘risk’ calculation where the AI will be able to use more or less track-width/kerb (instead of artificially inflating NPC capabilities), or adding marginal percentages to throttle application on corner exit, brake points, to cause the more natural pace adjustment we’re looking for, inclusive of a risk of spinning out/locking up when the “rubberband” logic pulls too hard.

There’s a whole ‘nother world of possibility within the realm of AI pit strategy. We know it’s effectively “nil” at present, so both plenty of room to improve in addition to it being brand new territory for PD’s code. Yet a leader who starts lifting early to better align “empty gas tank” to a future “lap finish” would see a safer, slower pace, versus a car whose risk calculation puts a tire in the dirt and tips them over into “two-stop” territory, logically. Could be cool.

 

[Lunard]

Hello guys.My Super GT 2015 and later grid, with some DTM intruders.