Assetto Corsa PC Mods General DiscussionPC 

  • Thread starter Thread starter daan
  • 148,959 comments
  • 44,972,811 views
In my experience you can make pretty much each and any car fast with only one fastlane. If you have problems then change the ai.ini of you car (brakehint, tyre hint, aero hint). Different setups for certain cars seems to be very workintensive and unnecessary.

I constantly switch between 30s, 50s, 60s and 70s.

I have AIs for the lotus 49 that are absolutly perfect, but don't fit the wsc for example. Not even mentioning the formula 1 from the 30s that are so special that almost no "regular" AI work. And when they do, i'm in for the most boring race ever.

Another thing worth it in my opinion, some little optimisations for certain classes. Cutting the apex works well sometimes, and especially, the AI doesn't open the insides, which is crucial to my little adventures.

For the price of a few kb, it makes sense to me.
I don't want to race the AI, i want to be beaten flat dead by it.

Edit : forgot to mention as well, my capacity to delete perfectly decent fastlanes or hints while tweaking. Or even sharing an AI that sucks and that should just only be used in specific conditions. Having it secured and being able to just compare without copy pasting or multiplying files in the original folder sounds nice to me.

Edit 2 : that would work fine as well for your laggio maggore fix, you could keep a dedicated layout for GT3 and under, and make a new one for faster cars.
 
Last edited:
Mod: rss gt legend pack
Skin: the beautiful "FIA GT 2002" pack by DavidDGA, new version. ( https://www.overtake.gg/downloads/2002-fia-gt-championship-skinpack.49420/ )
I create a custom championship in CM, but unfortunately... the game crashes and this error appears.
A few days ago (and also in the past, 2 years ago) I used the old version of this skin pack, and I had no problems...😥
 

Attachments

  • IMG_20250804_210044.webp
    IMG_20250804_210044.webp
    345.4 KB · Views: 3
Chimay - v1.01

1754385206542.webp


features:
  • 24 pit/grid with AI
  • 2 layouts: historic and no-chicane
  • mostFX™
  • historic vibes

credits:
  • original track by Sissou, converted to AC by @RMi_wood
  • 3d fixes, trees, physical road mesh, crowd facing spectators by @pugsang
  • terrain geo mapping, flag fixes, more trees, misc shaders by Tyrone-Nukedrop
  • loading logo, sections file by @Fanapryde
  • beta testing by @Breathe, @lux3y and those listed above
  • track conversion suggested by @Breathe


:: DOWNLOAD ::
:: DOWNLOAD v1.01 ::
Chimay - v1.01
- update for missing surfaces.ini in the nochicane layout

:: DOWNLOAD v1.01 ::
 
Last edited:
In my experience you can make pretty much each and any car fast with only one fastlane. If you have problems then change the ai.ini of you car (brakehint, tyre hint, aero hint). Different setups for certain cars seems to be very workintensive and unnecessary.

Exactly. People who complain that AC ai is bad should usually complain that the track fastlane is badly done instead. Sure there are bad mods and bad cars like unmodded SimDream but most of the time it's just the fastlane. Or then they have set global aggression to 99 from CM. Gee... what a surprise that they drive like they do and crash to you on the corners. When you set the aggression for the drivers carefully and INDIVIDUALLY... usually never exceeding 25 for the top drivers (depends a bit on the cars too) you'll be fine.

There's almost unlimiited things you can do to improve the ai on AC. I really suck at car mods but even a dummy like me can copy the suspension from a VRC car and test how it behaves on the SimDream train wreck, etc.

We need more people like you who do these things. @Thockard doesn't get enough credit for his work on track fastlanes. 1998 track pack is pure perfection thanks to his AI and I absolutely loved every minute when I played my 1998 season.


I swear all his track fastlanes are insanely competitive and safe. There's no bad AI if things have been done properly all around.

I think I did really well with my 2004 Kunos car by merging it with the VRC 2007 car. Drives great. Not because I'm a good modder... but because I tested and experimented so much and copied other people's work. I just got lucky and VRC 2007 suspension and tyres worked great with Kunos F2004 (with some additional adjustments). Ironically vanilla Kunos F2004 drove like crap on most popular F1 tracks... and AI couldn't even stay on the track on Kunos Nurburgring. VRC 2007 tyres solved all problems.
 
Last edited:
Exactly. People who complain that AC ai is bad should usually complain that the track fastlane is badly done instead. Sure there are bad mods and bad cars like unmodded SimDream but most of the time it's just the fastlane. Or then they have set global aggression to 99 from CM. Gee... what a surprise that they drive like they do and crash to you on the corners. When you set the aggression for the drivers carefully and INDIVIDUALLY... usually never exceeding 25 for the top drivers (depends a bit on the cars too) you'll be fine.

There's almost unlimiited things you can do to improve the ai on AC. I really suck at car mods but even a dummy like me can copy the suspension from a VRC car and test how it behaves on the SimDream train wreck, etc.

We need more people like you who do these things. @Thockard doesn't get enough credit for his work on track fastlanes. 1998 track pack is pure perfection thanks to his AI and I absolutely loved every minute when I played my 1998 season.


I swear all his track fastlanes are insanely competitive and safe. There's no bad AI if things have been done properly all around.

I think I did really well with my 2004 Kunos car by merging it with the VRC 2007 car. Drives great. Not because I'm a good modder... but because I tested and experimented so much and copied other people's work. I just got lucky and VRC 2007 suspension and tyres worked great with Kunos F2004 (with some additional adjustments). Ironically vanilla Kunos F2004 drove like crap on most popular F1 tracks... and AI couldn't even stay on the track on Kunos Nurburgring. VRC 2007 tyres solved all problems.


Talking about improvments, there's a way to write luas for AI, has anyone here ever tried to script one? Wanted to give it a try but never found time for it. Also, there's a new "adapting AIs" box in CSP New AI behavior, out of the loop on this one as well.
 
Last edited:
Is there a list/database of complete motorsport series for AC?

When I say "complete motorsport series" I mean a complete season of any motorsport series including all cars, liveries and tracks, mods or otherwise. I know there are collections of mods out there with which you can combine to create a full season (i.e. Pyyre's track extensions combined with SuzQ's F1 car packs to recreate the 2024 Formula 1 World Championship), but is there a list showing what combinations of mods you'd need/are available?
 
Is there a list/database of complete motorsport series for AC?

When I say "complete motorsport series" I mean a complete season of any motorsport series including all cars, liveries and tracks, mods or otherwise. I know there are collections of mods out there with which you can combine to create a full season (i.e. Pyyre's track extensions combined with SuzQ's F1 car packs to recreate the 2024 Formula 1 World Championship), but is there a list showing what combinations of mods you'd need/are available?

I have complete F1 seasons 1982, 1983, 1984, 1986, 1987, 1988, 1990, 1991, 1992, 1993, 1994, 1998, 2002, 2003, 2004, 2005, 2007, 2008. All tracks, cars and period correct TV graphics. Some of the cars are pay mods but 1986, 1991, 1993, 1998, 2002, 2004, 2005, 2007 and 2008 are 100% free.

Link in my signature.
 
Last edited:
I have complete F1 seasons 1982, 1983, 1984, 1986, 1987, 1988, 1990, 1991, 1992, 1993, 1994, 1998, 2002, 2003, 2004, 2005, 2007, 2008. All tracks, cars and period correct TV graphics. Some of the cars are pay mods but 1986, 1991, 1993, 1998, 2002, 2004, 2005, 2007 and 2008 are 100% free.

Link in my signature.
Perfect, thanks.

I was wondering, has anyone here done something similar for other series?
 
There's no bad AI if things have been done properly all around.
I mean, I know you bang on and on about this, chief, but as someone who's worked for dozens of hours in the background with the AC AI, I can assure you that this is absolutely not true at all.

The AI is far more limited than you realize, in way too many ways. And is extremely fragile - as in, it can change behavior or break completely with one CSP preview update.

Take, for a very simple example, one of my half-dozen projects I'm currently working on: Belo Horizonte by Authentic Simulation. It is looking like it is quite literally impossible to get these cars through the chicanes - they are too stupid to understand there are death barriers on the left side of both of the chicanes. They cannot do it. No alternative line will help, no DANGER hint will help, nothing. They are too dumb to comprehend the absolute basics of a racing circuit.

This extends to all street circuits, really - I know you say you're not quick, but near-universally, if you do want the AI to be fast on thin lines with hazards (hills, walls, etc.) it's a huge challenge and some times damn near impossible.

(For posterity's sake, I quite like Thockard's work, but generally I find them in the "safe" range a la Rainmaker. I am thankful for Rainmaker's stuff, but I'm generally almost 2 seconds a lap quicker than his fastest, with more time to find - and while I would say I'm fast, I'm by no means an alien).


There are at least 10 more problems in my standard workflow that are 100% attributable to the AI being unbelievably stupid. And that's just building the AIs themselves - as AC drivers, everyone is familiar with their limitations - they love to ignore your presence on an overtaking line, fall into a train-like robotic rhythm after a few laps, cannot understand fuel loads, cannot qualify properly, cannot set pit strategies for endurance races, and on and on and on.

If offline AI racing is going to improve - and it needs to, desperately - it will come from CSP. The "dynamic" AI introduced in 0.2.11 is an interesting one - I have noticed slightly more mid-stint "human" mistakes from cars. Kind of like the hilarious spin-outs from NFS Shift, but at a lesser, believable degree. It's a start - and that's a good thing.

More importantly, though, it signifies to me that the AI might finally become a priority for CSP in 0.3 and beyond. Better awareness, new parameters for AI creation (like understanding hazards), more human behavior, pit stops (i.e. folding Nuzzi's app into CSP officially) are all on my wishlist.

Here's hoping something comes of it.
 
Perfect, thanks.

I was wondering, has anyone here done something similar for other series?

Not so many. Apparently not so many people are interested in full seasons. For me it's the only way to race. Sure I hotlap occasionally but I usually do that on ACC and EVO.

I'd love to do a full CART 1999 season but AC is just not built for that (no flying start and lack of tracks, etc). The 1999 CART cars drive very well.

Another full season of classic 90's GT cars would be nice. I think it was FIFA GT / BPR series or whatever we had on that GTR2 sim back in the day. You know those amazing cars... Jag XJ220, Porsche 993 GT2 (the GOAT), Ferrari F40... I'd love to play full season of that series on AC on accurate tracks. Those cars are way cooler than today's. Sure today's cars are even faster but these are way more analog, raw and menacing.

CvSdjFK.jpeg


1920px-Jaguar_XJ220_GT_at_Goodwood_2014_003.jpg


1280px-1995_McLaren_F1_GTR_%2820195044452%29.jpg
 
Not so many. Apparently not so many people are interested in full seasons. For me it's the only way to race. Sure I hotlap occasionally but I usually do that on ACC and EVO.

I'd love to do a full CART 1999 season but AC is just not built for that (no flying start and lack of tracks, etc). The 1999 CART cars drive very well.

Another full season of classic 90's GT cars would be nice. I think it was FIFA GT / BPR series or whatever we had on that GTR2 sim back in the day. You know those amazing cars... Jag XJ220, Porsche 993 GT2 (the GOAT), Ferrari F40... I'd love to play full season of that series on AC on accurate tracks. Those cars are way cooler than today's. Sure today's cars are even faster but these are way more analog, raw and menacing.

CvSdjFK.jpeg


1920px-Jaguar_XJ220_GT_at_Goodwood_2014_003.jpg


1280px-1995_McLaren_F1_GTR_%2820195044452%29.jpg
I was thinking about making a spreadsheet similar to @Breathe's track listing, cataloguing every single racing series as archived on racing years.

Each entry will document the cars, tracks and skins available, as well as highlighting any missing cars/tracks/skins. This will improve mod visibility and highlight any missing content.
 
X90
I mean, I know you bang on and on about this, chief, but as someone who's worked for dozens of hours in the background with the AC AI, I can assure you that this is absolutely not true at all.

The AI is far more limited than you realize, in way too many ways. And is extremely fragile - as in, it can change behavior or break completely with one CSP preview update.

Take, for a very simple example, one of my half-dozen projects I'm currently working on: Belo Horizonte by Authentic Simulation. It is looking like it is quite literally impossible to get these cars through the chicanes - they are too stupid to understand there are death barriers on the left side of both of the chicanes. They cannot do it. No alternative line will help, no DANGER hint will help, nothing. They are too dumb to comprehend the absolute basics of a racing circuit.

This extends to all street circuits, really - I know you say you're not quick, but near-universally, if you do want the AI to be fast on thin lines with hazards (hills, walls, etc.) it's a huge challenge and some times damn near impossible.

(For posterity's sake, I quite like Thockard's work, but generally I find them in the "safe" range a la Rainmaker. I am thankful for Rainmaker's stuff, but I'm generally almost 2 seconds a lap quicker than his fastest, with more time to find - and while I would say I'm fast, I'm by no means an alien).


There are at least 10 more problems in my standard workflow that are 100% attributable to the AI being unbelievably stupid. And that's just building the AIs themselves - as AC drivers, everyone is familiar with their limitations - they love to ignore your presence on an overtaking line, fall into a train-like robotic rhythm after a few laps, cannot understand fuel loads, cannot qualify properly, cannot set pit strategies for endurance races, and on and on and on.

If offline AI racing is going to improve - and it needs to, desperately - it will come from CSP. The "dynamic" AI introduced in 0.2.11 is an interesting one - I have noticed slightly more mid-stint "human" mistakes from cars. Kind of like the hilarious spin-outs from NFS Shift, but at a lesser, believable degree. It's a start - and that's a good thing.

More importantly, though, it signifies to me that the AI might finally become a priority for CSP in 0.3 and beyond. Better awareness, new parameters for AI creation (like understanding hazards), more human behavior, pit stops (i.e. folding Nuzzi's app into CSP officially) are all on my wishlist.

Here's hoping something comes of it.

Street circuits are tough. Rainmaker's Detroit, however, works quite well until AI faces the backmarkers. Phoenix pay track AI was trash but that was fixed. I have the fixed ai. Phoenix has its limitations and AI still creates drama there sometimes. But you can have a decent race unless you crash yourself (happens to me often).

I haven't had time to play Shin's new Long Beach 1983 but it seems ok. Dallas 1984 has issues but even that is raceable (although half of the cars retire) but then again so did half of the cars in the real 1984 Dallas race hehe. Monaco 2017 started to have issues with new CM that it never had before. Cars' racing lines changed suddenly even without updating the CSP. But sure... Monaco is always questionable. It doesn't work perfectly on the longer race... ever. Sometimes you get a flawless one but it's not guaranteed... backmarkers might ruin it badly.

Valencia and Singapore run pretty much problem free in 2008. Adelaide (both 1988 and 1992) work flawlessly, etc.

These factors may be the reason why ai in my F1 seasons work well:

1. F1 cars' much better traction that are pretty easily adjustable by trial and error. If I had an issue on certain track (like I had with Kunos F2004) I simply adjusted suspension and tyres.
2. Besides the mentioned street circuits most F1 tracks don't have many challenging things for the AC ai... you know like tight walls like Monaco
3. Every driver aggression in my system is very limited
4. Excellent cars like VRC that don't really cause drama
5. Laps limited to 20
6. I've set the user to play the top car or the second best car and the lower end cars/drivers have way smaller AI rating and almost zero aggression (easy to overtake) and my starting position is usually in top 3-4.

Sure... any Lotus98T based F1 car has the dreaded issue that they might start "crawling" on the track after they get hit and don't retire. That is a major issue but it happens only on Rainmaker Detroit and Oesterreichring (80s seasons).

But I have played full years of 1998, 2005, 2007, 2002 and now 2004... etc... and never had one single issue that you mention. Sure overtaking must be very carefully planned and same with the backmarkers that don't act like backmarkers... but they actually race you. That is unforgivable. And of course you're not overtaking them on the Spa bus stop. The risk of crash is way too high. As you said they don't care you lead the race... or are even there. You're right. But you can overtake them after Eau Rouge straight for sure. I'm not denying there aren't issues but you can adapt to them.

But sure... I see your point... some street circuits are tough for the ai. And I can see there might be many more issues if you drive standard cars with less traction and more slide on the corners. And especially if you like longer races. I pretty much race against AI only the F1 cars (and only hotlap the street cars) so my view is limited.

And if you feel the player controlled car is too fast (or you are too fast against ai) simply go to car data and add some CD gain in aero and reduce tyre traction from DX REF for your team. I guarantee it's very easy to make your car slower... or AI faster. Of course this doesn't work on skin based seasons (when you have only one or two base cars). But in what other sim you can do that? NONE.

However my seasons work very well for me and I don't have any major AI issues that would ruin any races. And yes, it might be a blessing to be just an average sim racer. The main idea is to have a challenging and enjoyable difficulty FOR YOU... not to win every race. At least I wouldn't want to drive a superior car and win every race. Where's the fun and challenge in that? That's why I like to drive the challenger (like Montoya in 2002 or Alesi in 1994).

About the other sims... I haven't played full seasons that much... I did iRacing Indycar offline season back in the day and it was OK and AMS2 might have more developed AI... I played full 1988 season... and they have done so much great classic F1 stuff... but for me the driving is uhh... not so good. I played RF1 back in the day a lot... also EA 2002 F1 and GP4. Trust me, they all had their issues, way worse than AC.

F1 2020 Codemasters actually had pretty decent AI and it was the only Codemasters F1 game worth playing. Way better than its successors. FFB was crap though... too bad you can't purchase it for PC anymore. I had it on PS4 then. If anyone knows where I can still get it for PC let me know. I'd like to revisit it only to compare... and if my memory even serves me correctly. Or maybe I have to drag my PS4 from the attic.

Cheers.

I was thinking about making a spreadsheet similar to @Breathe's track listing, cataloguing every single racing series as archived on racing years.

Each entry will document the cars, tracks and skins available, as well as highlighting any missing cars/tracks/skins. This will improve mod visibility and highlight any missing content.

I don't know if it's necessary as there really aren't many. And you have to have paid version of the CM to even play Championships.Not sure how many people have even purchased the full CM version for that.

F1 is pretty much covered. Of course not all tracks have the banners at 100% correct place... I use 1988 track pack for all 80's years... but it's in pretty good shape. One major missing thing from 1980s-today is the 80's Zandvoort. It's crazy that no one has done that circuit. 1989/1990 Phoenix layout is only available as a pay track. 1982 Long Beach is missing, proper 80's Brands Hatch not done, either. But I use 90's Brands with 80's ads (made them mysef). It's ok as beggars can't be choosers.

There are some missing tracks from the 70's seasons though.
 
Last edited:
I was thinking about making a spreadsheet similar to @Breathe's track listing, cataloguing every single racing series as archived on racing years.

Each entry will document the cars, tracks and skins available, as well as highlighting any missing cars/tracks/skins. This will improve mod visibility and highlight any missing content.
UPDATE: This is going to take the rest of my life to complete...
 
Exactly. People who complain that AC ai is bad should usually complain that the track fastlane is badly done instead. Sure there are bad mods and bad cars like unmodded SimDream but most of the time it's just the fastlane. Or then they have set global aggression to 99 from CM. Gee... what a surprise that they drive like they do and crash to you on the corners. When you set the aggression for the drivers carefully and INDIVIDUALLY... usually never exceeding 25 for the top drivers (depends a bit on the cars too) you'll be fine.

There's almost unlimiited things you can do to improve the ai on AC. I really suck at car mods but even a dummy like me can copy the suspension from a VRC car and test how it behaves on the SimDream train wreck, etc.

We need more people like you who do these things. @Thockard doesn't get enough credit for his work on track fastlanes. 1998 track pack is pure perfection thanks to his AI and I absolutely loved every minute when I played my 1998 season.


I swear all his track fastlanes are insanely competitive and safe. There's no bad AI if things have been done properly all around.

I think I did really well with my 2004 Kunos car by merging it with the VRC 2007 car. Drives great. Not because I'm a good modder... but because I tested and experimented so much and copied other people's work. I just got lucky and VRC 2007 suspension and tyres worked great with Kunos F2004 (with some additional adjustments). Ironically vanilla Kunos F2004 drove like crap on most popular F1 tracks... and AI couldn't even stay on the track on Kunos Nurburgring. VRC 2007 tyres solved all problems.
I've always been curious about the AI aggression settings... I've seen some folks say "never go above 50%" and you've said you don't go above 25%. This seems to be a common thing, but for me if I follow these guidelines the AI never pass each other. If the leaders reach lapped traffic, unless they are considerably faster they just fall in line and ride around which immediately breaks immersion for me.

I did a race awhile back where the leader was about 20+ seconds in front of me and I was gaining about 0.4 seconds per lap. I was trying to do the math in my head 'can I get to him before the race ends??'. Well, needless to say I was able to catch him the next lap as the leader was just chilling behind a backmarker. I easily passed them both and by the end of the race I was 40 seconds ahead. He never passed the lapped car. And my aggression was set to like 66% or something.

I guess what I'm getting at is: How do you make those numbers work? The biggest issue I have with AC AI is the same I had with NASCAR Racing 2003: They are blind to danger (blindly pile into crashes) and they won't pass slower cars.

I don't do any AI modding, AI.ini is a foreign language to me. I will download the amazing updated fastline files that are posted but aside from that I don't have the knowledge or the time to tweak every single mod/track/etc.
 
I've always been curious about the AI aggression settings... I've seen some folks say "never go above 50%" and you've said you don't go above 25%. This seems to be a common thing, but for me if I follow these guidelines the AI never pass each other. If the leaders reach lapped traffic, unless they are considerably faster they just fall in line and ride around which immediately breaks immersion for me.

I did a race awhile back where the leader was about 20+ seconds in front of me and I was gaining about 0.4 seconds per lap. I was trying to do the math in my head 'can I get to him before the race ends??'. Well, needless to say I was able to catch him the next lap as the leader was just chilling behind a backmarker. I easily passed them both and by the end of the race I was 40 seconds ahead. He never passed the lapped car. And my aggression was set to like 66% or something.

I guess what I'm getting at is: How do you make those numbers work? The biggest issue I have with AC AI is the same I had with NASCAR Racing 2003: They are blind to danger (blindly pile into crashes) and they won't pass slower cars.

I don't do any AI modding, AI.ini is a foreign language to me. I will download the amazing updated fastline files that are posted but aside from that I don't have the knowledge or the time to tweak every single mod/track/etc.

Good question. I can't say for sure that "don't go over 25%" works all the time. It depends on the car. For SD 1982-1984 cars even 25 is overkill.

But even if you increase the aggression and they catch the traffic they can't always pass the backmarkers any easier. The backmarker problem is going to be there no matter what. In that sense the ai is broken no doubt. But that problem is 100% worse if the slower backmarkers have high aggression.

The way I build them:

So... let's take 1998 which has two base cars. McLaren and Ferrari and all other team drivers are just skins. There was only Schumacher and Hakkinen fighting for the title. They're ai 99-100, aggression there in 25%'ish. Then you have the second drivers in the top teams Coulthard and Irvine. Hill was driving pretty mundane Jordan so he'd be like 93 or so, etc.

Then you have low ranked teams with slow drivers (in relative terms, all F1 drivers are quick) like Takagi, Diniz, Tuero. They should be ranked at 89-90, with aggression level 1%. There's that 10 digit difference in ai skill which is plenty which helps the higher ranked drivers to overtake them when they become backmarkers.

Then there's the difference between McLaren and Ferrari base car performance. You'll find out when you let the AI qualify what driver values you have to adjust to achieve optimal, realistic QF result. Take one solid track like Melbourne 1998 and see how they perform. Repeat a few times and see if your settings are right. If say, Villeneuve is behind Tuero (because Tuero base car is McLaren which has better performance) increase/decrease the individual driver AI skills as soon as you achieve realistic looking results. You only have to change 2-3 digits usually and that does the trick. Just make sure the top 2 drivers are on the different level. Like 99-100 and make a significant (like 6 digit difference) to the "next class" (Coulthard, Irvine).

And the point is... when you race 20 laps and you still have the differences between the drivers at moderate level (89-100) you will not have many backmarkers in that short of a race and when you only rank the top drivers' aggression higher, the backmarker issue is smaller. The lower ranked cars/drivers won't really affect the race results in any way as most F1 seasons have always been two, max three driver battles.

1988-1990 it was all Senna and Prost, 1998-2001 Hakkinen vs Schumi and even today it's just McLaren's fighting for the title etc. So it's max four drivers that should be winning races, every season. Sure 2012 was an exception but generally.

Check this video I posted yesterday. Clean overtake... Alonso ai level 97, aggression at 19%. If I have set his aggression to 60% there is a big chance that he would have knocked me out on that corner. Now you can pretty much always make a clean overtake when the aggression is lower as long as you keep your line. AI won't challenge you on the turn if you're clearly past him.

From 2:00



Sure you (as a driver) have to adapt to this AI and many times you have to be sure that the place where you attempt the overtake is good and safe. I suspect many don't have the patience. For example in Magny Cours there's only one place to do the overtake safely and that's at the end of Golf straight before Adelaide chicane. If you try to overtake anywhere else you'll probably crash or at least have a hit. Sure that's unrealistic as in real races overtakes happen in more places but that's the thing you have to adapt to.

And when I've used higher aggression numbers for the AI, I've seen them crash off the track way more often and ruin their qualifying by ending up on the gravel. When they do and return to the pit, they are not trying another QF lap after that (for whatever reason). And as I said they also tend to knock you out in the overtake situations way more eagerly and challenge you even on the turn.

And you know the AI problem where you simply brake in front of a AI car behind you it might just stay there and won't even attempt an overtake if you're still in the racing line. Increasing the aggression won't take this problem away. However these things won't happen if you race normally, unless you have that "crawling car problem" which happens with Lotus98T based cars. Those can ruin the races for sure and I don't know what causes it.

This is another video I made of AC ai. 2002 cars (skins) at Kunos Monza. Schumacher overtook me immediately at Parabolica when he had a chance. Aggression only at 25. I put more to the video description.



That race was amazing.

I'm not saying I've discovered anything new or that I have some "special secret" but I've tested these for years and personally I have very enjoyable races with almost zero ai issues. Sure I may have "adapted" to the AI after driving these same cars and tracks so much but I have no problems in my seasons at all. Usually when I crash it's my own fault and I tried the overtake on wrong place.

If someone could improve these known issues, it'd be amazing. Nuzzi has a blue flag app but I have never tested it as I don't have a problem with the backmarkers with my system (except for street circuits sometimes).

Good discussion.
 
Last edited:
Good question. I can't say for sure that "don't go over 25%" works all the time. It depends on the car. For SD 1982-1984 cars even 25 is overkill.

But even if you increase the aggression and they catch the traffic they can't always pass the backmarkers any easier. The backmarker problem is going to be there no matter what. In that sense the ai is broken no doubt. But that problem is 100% worse if the slower backmarkers have high aggression.

The way I build them:

So... let's take 1998 which has two base cars. McLaren and Ferrari and all other team drivers are just skins. There was only Schumacher and Hakkinen fighting for the title. They're ai 99-100, aggression there in 25%'ish. Then you have the second drivers in the top teams Coulthard and Irvine. Hill was driving pretty mundane Jordan so he'd be like 93 or so, etc.

Then you have low ranked teams with slow drivers (in relative terms, all F1 drivers are quick) like Takagi, Diniz, Tuero. They should be ranked at 89-90, with aggression level 1%. There's that 10 digit difference in ai skill which is plenty which helps the higher ranked drivers to overtake them when they become backmarkers.

Then there's the difference between McLaren and Ferrari base car performance. You'll find out when you let the AI qualify what driver values you have to adjust to achieve optimal, realistic QF result. Take one solid track like Melbourne 1998 and see how they perform. Repeat a few times and see if your settings are right. If say, Villeneuve is behind Tuero (because Tuero base car is McLaren which has better performance) increase/decrease the AI skill as soon as you achieve realistic looking results. You only have to change 2-3 digits usually and that does the trick.

And the point is... when you race 20 laps and you still have the differences between the drivers at moderate level (89-100) you will not have many backmarkers in that short of a race and when you only rank the top drivers' aggression higher, the backmarker issue is smaller. The lower ranked cars/drivers won't really affect the race results in any way as most F1 seasons have always been two, max three driver battles.

1988-1990 it was all Senna and Prost, 1998-2001 Hakkinen vs Schumi and even today it's just McLaren's fighting for the title etc. So it's max four drivers that should be winning races, every season. Sure 2012 was an exception but generally.

Check this video I posted yesterday. Clean overtake... Alonso ai level 97, aggression at 19%. If I have set his aggression to 60% there is a big chance that he would have knocked me out on that corner. Now you can pretty much always make a clean overtake when the aggression is lower as long as you keep your line. AI won't challenge you on the turn if you're clearly past him.

From 2:00



Sure you have to adapt to this AI and many times you have to be sure that the place where you attempt the overtake is good and safe. I suspect many don't have the patience. For example in Magny Cours there's only one place to do the overtake safely and that's at the end of Golf straight before Adelaide chicane. If you try to overtake anywhere else you'll probably crash or at least have a hit. Sure that's unrealistic as in real races overtakes happen in more places but that's the thing you have to adapt to.

And when I've used higher aggression numbers for the AI, I've seen them crash off the track way more often and ruin their qualifying by ending up on the gravel. When they do and return to the pit, they are not trying another QF lap after that (for whatever reason). And as I said they also tend to knock you out in the overtake situations way more eagerly and challenge you even on the turn.

And you know the AI problem where you simply brake in front of a AI car behind you it might just stay there and won't even attempt an overtake if you're still in the racing line. Increasing the aggression won't take this problem away. However these things won't happen if you race normally, unless you have that "crawling car problem" which happens with Lotus98T based cars. Those can ruin the races for sure and I don't know what causes it.

This is another video I made of AC ai. 2002 cars (skins) at Kunos Monza. Schumacher overtook me immediately at Parabolica when he had a chance. Aggression only at 25. I put more to the video description.



That race was amazing.

I'm not saying I've discovered anything new or that I have some "special secret" but I've tested these for years and personally I have very enjoyable races with almost zero ai issues. Sure I may have "adapted" to the AI after driving these same cars and tracks so much but I have no problems in my seasons at all. Usually when I crash it's my own fault and I tried the overtake on wrong place.

If someone could improve these known issues, it'd be amazing. Nuzzi has a blue flag app but I have never tested it as I don't have a problem with the backmarkers with my system (except for street circuits sometimes).

Good discussion.

Thank you for that thoughtful and detailed response. You have raised a valid point in the sense that I have been running higher aggression levels to encourage passing (esp. for lapped traffic) but that doesn't seem to actually make a difference. Furthermore, with my settings the backmarkers ALSO have high aggression so they are likely fighting with the leader rather than letting them pass. I also see from your videos that your lower aggression settings still allow some passing, which is surprising to me. I will do some experimentation with lower aggression values, esp at the back of the field and see if that helps. Thank you very much for your explanation.
 
I've always been curious about the AI aggression settings... I've seen some folks say "never go above 50%" and you've said you don't go above 25%. This seems to be a common thing, but for me if I follow these guidelines the AI never pass each other. If the leaders reach lapped traffic, unless they are considerably faster they just fall in line and ride around which immediately breaks immersion for me.

I did a race awhile back where the leader was about 20+ seconds in front of me and I was gaining about 0.4 seconds per lap. I was trying to do the math in my head 'can I get to him before the race ends??'. Well, needless to say I was able to catch him the next lap as the leader was just chilling behind a backmarker. I easily passed them both and by the end of the race I was 40 seconds ahead. He never passed the lapped car. And my aggression was set to like 66% or something.

I guess what I'm getting at is: How do you make those numbers work? The biggest issue I have with AC AI is the same I had with NASCAR Racing 2003: They are blind to danger (blindly pile into crashes) and they won't pass slower cars.

I don't do any AI modding, AI.ini is a foreign language to me. I will download the amazing updated fastline files that are posted but aside from that I don't have the knowledge or the time to tweak every single mod/track/etc.
For me I keep the ai aggressetion around 70-79 percent it's a good combination for ai speed and passing especially for CSP version above 0.1.79. But alot of ai aggression is based on tracks and the individual cars and mods them selves. For cars with aggressive Trail and braking hints in the ai.ini file. Sometimes it's best to play with aggression settings in game or modify the trail and braking hints in the ai.ini file. The same is true for the opposite scenario.
 
For me I keep the ai aggressetion around 70-79 percent it's a good combination for ai speed and passing especially for CSP version above 0.1.79. But alot of ai aggression is based on tracks and the individual cars and mods them selves. For cars with aggressive Trail and braking hints in the ai.ini file. Sometimes it's best to play with aggression settings in game or modify the trail and braking hints in the ai.ini file. The same is true for the opposite scenario.

Sure. My experience is very limited as I only adjust and drive classic F1 cars. Stuff that I do work with my seasons and lower aggression numbers definitely reduce the problems. And surely there are cars that can probably handle bigger aggression numbers but I think you should not use general settings but set them to every driver individually. Especially in F1 seasons it's very important to make the two, three top drivers more aggressive and max their AI level (compared to the rest of the grid). There should be significant differences between Lance Stroll and Max Verstappen. In the old days (80's) 4-5 second gaps in qualifying were normal. Now they're all within 1 second of eachother which probably would make setting up a 2025 season way more difficult.

When I started tuning these years ago I only did stuff of CM... you know ballast, restrictor and they never worked for me. I got idiotic qualifying results like Luis Perez-Sala outqualifying Senna and Prost in the 1988 season.

I use the championship JSON files to define things and never, ever touch anything on CM. And even if I did I think those JSON files are always the priority and ignore the CM settings anyway. However I get ultra realistic QF results. For example in my 1988 season Senna and Prost took 15 poles out of 16. Berger took the one that was left. And amazingly, that's how it was in real 1988 season too. In my 2007 season Alonso, Raikkonen and Hammy ended up within 6 points to eachother.

But yeah... if you have cars that can handle higher aggression and don't have those issues (and don't have the time or patience) to make individual driver settings general 75 could work fine. Cheers.
 
Sure. My experience is very limited as I only adjust and drive classic F1 cars. Stuff that I do work with my seasons and lower aggression numbers definitely reduce the problems. And surely there are cars that can probably handle bigger aggression numbers but I think you should not use general settings but set them to every driver individually. Especially in F1 seasons it's very important to make the two, three top drivers more aggressive and max their AI level (compared to the rest of the grid). There should be significant differences between Lance Stroll and Max Verstappen. In the old days (80's) 4-5 second gaps in qualifying were normal. Now they're all within 1 second of eachother which probably would make setting up a 2025 season way more difficult.

When I started tuning these years ago I only did stuff of CM... you know ballast, restrictor and they never worked for me. I got idiotic qualifying results like Luis Perez-Sala outqualifying Senna and Prost in the 1988 season.

I use the championship JSON files to define things and never, ever touch anything on CM. And even if I did I think those JSON files are always the priority and ignore the CM settings anyway. However I get ultra realistic QF results. For example in my 1988 season Senna and Prost took 15 poles out of 16. Berger took the one that was left. And amazingly, that's how it was in real 1988 season too. In my 2007 season Alonso, Raikkonen and Hammy ended up within 6 points to eachother.

But yeah... if you have cars that can handle higher aggression and don't have those issues (and don't have the time or patience) to make individual driver settings general 75 could work fine. Cheers.

I suppose high downforce and grippy cars allow for a better AI behaviour in AC. I haven't tried your championship, will def do (i'm still stuck in the early 70s).

From my experience, most issues i have with AI are with low grip cars. The AI just... sucks, and i'm just 10 secs ahead whatever the track.

I'd be pleased to be proven wrong with a vintage car/track combo that would leave me in the dirt, but i haven't found one yet, and i don't really think current hardcoded AI limitations allows for more room on the matter.

And yet, with broken car app or the AI app, you have some minor access. In my dreams, i hope for basic control : add clutch at start, bring defensive coordinates for fastlane when opponent ‹ 5m behind, use alt power.ini, etc. Just "tricks" to simulate a racing behavior, instead of this brainish minionry where you just hold a tear each time the AI randomly makes something interresting, usually before ending its own life on the next corner's exit because you are on the fastlane, and it's 2m away from it, and life is not worth it that far from the fastlane.
 
Last edited:
I suppose high downforce and grippy cars allow for a better AI behaviour in AC. I haven't tried your championship, will def do (i'm still stuck in the early 70s).

From my experience, most issues i have with AI are with low grip cars. The AI just... sucks, and i'm just 10 secs ahead whatever the track.

I'd be pleased to be proven wrong with a vintage car/track combo that would leave me in the dirt, but i haven't found one yet, and i don't really think current hardcoded AI limitations allows for more room on the matter.

I think you're 100% spot on here.

Some people complained to me that cars in my 1982, 1983 and 1984 (and 1988 in some extent) seasons have unrealistically high corner speeds. And yes they're right. Because increasing DX ref (traction of the tyres and using the qualifying tyre model only) was the ONLY WAY to make these SD cars to race safely and still keep them fast and competitive. Had I been better car modder I'm sure there could have been other ways to do that but that was the only way that worked for me after rigorous testing.

I outqualified the vanilla SD cars by 3-4 seconds before the adjustments and they constantly ran out of tracks and caused havoc. Increasing the grip solved all the problems. It was a mess and I can see why people hate SDD. They probably don't even test anything before releasing. I mean I've seen F1 releases that don't even turn at Monaco Loews hairpin because the wheels don't turn enough.

I compensated the excessive grip by reducing their high speed in aero.ini settings. Now the top cars only reach about 315 km/h with lower rear wing which is about 10-20 km/h too low (esp. on Monza, Spa. Paul Ricard Mistral straight etc). Now the lap times are extremely accurate in 80% of the races but the cars are 1-2 seconds too slow on some faster tracks.

When you have that grip it makes the cars easier to drive and AI is way quicker and safer because you eliminate the randomness that comes from the understeer and slide in the corners. Sure it reduces the realism for these older cars but I believe that would be the only way to make the 70's cars competitive.

One quick option (if you have individual cars) is simply to make your car (and team) slower with CD GAIN aero setting and increase DX REF grip only for the ai cars. Compensate the slower car by increasing your teammates' AI level.

I have some real life karting experience from my youth and I know how much grip even those cars have. You can't drift open wheelers. You have the grip or not... it's almost on/off. Of course nobody of us has ever driven a F1 car but you can see from the footage that 2000's F1 cars have ENORMOUS grip. So driving them is "easy" in that sense. Look at the cockpit view footage of any race and those guys can drive the cars with incredible consistency and have racing lines that are centimeter-like accurate for 70 laps. Sure they're ultra talented but for them they're relatively easy to drive on the limit.

When Keke Rosberg retired from F1 and went to DTM and the press asked him "how do these cars feel after F1"? He answered "It feels like being a cab driver". The grip and brake power difference to GT cars is ridiculous.

Creating F1 cars like on AMS2 (where they constantly oversteer and have controllable slides) is extremely unrealistic. Some people think that more difficult automatically means more realistic which is certainly not the case in sim racing. The only thing more realistic on AMS2 is that you have to drive those old F1 cars "more actively".

As you said AC ai is at its best with these 2000's F1 cars. VRC and CIM 2008 cars drive so well. I think not so much without the TC which decreases the grip to very unrealistic levels. I adjusted 1998 VS Ferrari rear wheel grip and it made a world of difference.It's super enjoyable to drive now when the excessive oversteer has been eliminated.

Your 70's cars probably have somewhat realistic behaviour and traction... the lap times were like 10-15 seconds slower back then. That's where AC ai fails miserably. Decreased grip brings a lot of AI issues and more randomness and errors. Fastlanes on the 70's tracks might be bad too... I don't have much experience of cars and tracks that old.

PS: Try the 2004 Championship that I created. It's all free and based on Kunos F2004. If you can take poles at Sepang and Melbourne with my settings, I'm genuinely amazed and probably realize how bad sim racer I am. I was 1.2 seonds off Schumacher's pace at Melbourne and almost 2 seconds off at Sepang. It's a small download (skin based package).

 
Last edited:

Anyone have the 2025 150 Black Falcon Bilstein for the GUERILLA Mods BMW M4 GT4 G82 ? The 2024 is available on Overtake can't seem to find the 2025.​

 
Sharing a mod? Host it on GTPlanet Downloads. Free, public hosting for files up to 10GB in size.
Back