MX Simulator

Using ChatGPT to Create the “Perfect Dirt/Erode Preset” in MX Simulator

Hello!

I wanted to share a really unique experience I’ve had recently while working on my track in MX Simulator. As most of you know, dialing in tileinfo can be a pretty frustrating process. You have dozens of settings to play with, and it’s not always obvious what numbers will give you the feel you’re after. For me, friction and erode are the two biggest pieces that make or break how fun a track is to ride.

I decided to take a different approach: I brought ChatGPT (AI) into the process to help me design and refine a completely new traction/erode preset. After a lot of back-and-forth — and a lot of testing — I ended up with what I’m calling my “Perfect Dirt/Erode Preset.” It feels completely unique compared to the usual stuff, and honestly, I don’t think I could have gotten here without the help of AI.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

1) Teaching ChatGPT About Tileinfo

The first thing I did was feed ChatGPT the forum post that explains tileinfo in detail. I wanted to make sure it fully understood how friction and erode actually work in MX Sim.

Once it was clear that it could explain back to me what each setting does, I started describing what I wanted the dirt to feel like — not just numbers, but actual rider feedback:

“I want the dirt to feel tacky and supportive, like you can just lean the bike and trust it.”
“I don’t want it slick anywhere, but I don’t want it so grippy that you’re wheelying everywhere.”
“I want the ruts to form skinny and deep, but flowy — not choppy or bumpy.”
“When I rail a corner, I want the berm to actually build up taller so I can lean more into it.”

The cool part was that ChatGPT was able to take those plain-language descriptions and convert them into usable tileinfo values.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

2) Iterating and Refining the Feel

The first few friction curves it gave me were close, but not quite what I was looking for.
I’d test them in-game and then give feedback like:

“This one feels too slick mid-corner.”
“I want the dirt to be more consistent at different speeds.”
“It feels good wide open, but I’m still sliding out when I roll on the throttle slow.”

Each time, ChatGPT adjusted the numbers just slightly — tightening the friction spread, making the traction more supportive, smoothing how the grip transitioned across slip speeds.

The same went for erosion. At first, it was forming ruts too wide or too shallow. I told it: “I want the rut to be skinny, deeper, and the berm to build taller on the outside.” It tweaked erode_depth, erode_displacement, and erode_lateral_scale until the results matched what I was describing.

This back-and-forth process was honestly the most fun I’ve had working on traction. Instead of blindly guessing numbers, I was basically “coaching” the AI with how the dirt should feel, and it gave me the numbers to make that happen.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

3) The Final Result: "Perfect Dirt/Erode Preset"

After refining and testing, I ended up with a traction/erode combo that I absolutely love. It’s unique, supportive, tacky, but not over-grippy. The erosion makes ruts build naturally and berms form with flow, instead of making the track unrideable.

Here’s the final preset if you want to try it yourself:

name Perfect Dirt/Erode Preset
friction 1.20 0.0
friction 1.19 0.5
friction 1.18 1.0
friction 1.17 2.0
friction 1.17 3.5
friction 1.16 5.0
friction 1.15 7.0
friction 1.14 10.0
roll_resist 52.0 77.0 37.0
roll_resist 62.0 87.0 47.0
erode_depth 5.50
erode_displacement 3.17
erode_radius 3.17
erode_inner_depth 3.17
erode_inner_radius 2.5
erode_lower 35.0
erode_upper 125.0
erode_off_speed 8.0
erode_on_speed 14.0
erode_slip_base_scale 1.0
erode_slip_scale 1.08
erode_slip_lower 0.0
erode_slip_upper 4.0
erode_min_slope -0.4
erode_max_slope 0.5
erode_project_to_ground 1.0
erode_downhill_scale 0.0
erode_front_scale 0.80
erode_lateral_scale 2.8
erode_limit 0.0 1.0 1.0
erode_limit 0.4 0.95 0.95
erode_limit 0.75 0.75 0.75
erode_limit 1.0 0.55 0.55

Copy/paste that into your tileinfo and give it a try.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

4) Why This Is Exciting

What makes this so exciting to me is that it shows how AI can help evolve MX Sim in ways we haven’t explored yet. Normally, getting traction right is hours (or days) of guesswork. But by using ChatGPT, I was able to shortcut a lot of that — not by skipping testing, but by making each test meaningful.

I’d ride, describe what I felt, and get back new numbers that directly addressed that feedback. The end result wasn’t just faster — it was better than what I could have done alone.

And honestly? This “Perfect Dirt/Erode Preset” feels like nothing else I’ve ridden in the game before. It’s unique, fun, and makes riding aggressive lines feel incredible.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

5) Final Thoughts

If you’ve been hesitant to mess with tileinfo, I really recommend giving it a shot. And if you’re open to experimenting, try involving AI in your workflow. For me, it turned what’s normally a frustrating trial-and-error process into a creative collaboration.

I think this is just the start. Imagine using AI not only for traction, but also for designing realistic erode profiles, tuning suspension to match surfaces, or even helping brainstorm track layouts. It’s a tool that can help push MX Sim forward in really exciting ways.

So yeah — that’s my story. AI helped me create a traction setup that I’m genuinely stoked on, and I think this kind of experimentation could open up a lot of possibilities for the community.

Would love to hear your thoughts if you give the preset a try.

– Cody B

While I strongly disagree with the use of AI in the track building process I do think things like this are a good, non intrusive use of it. You have essentially cut out the most boring and monotonous part of building tracks and given yourself a short cut straight to doing the actual fun part which is designing and building. As long as it's used for coding exclusively you will never see me complain but if it starts imposing on actual creative design like model and texture creation it's not something I can get behind.

IMO for anything that isn't pavement or blue groove hardpack you want to use slip_function 2 and add some negative slip values where the friction falls off. The description I used for slip_function 2 in the tileinfo info post was probably too technical. In practical terms, the positive values control the grip braking and turning while the negative values make it fall off for wheelspin.

It is impressive how much the LLMs know though!

sethypeety wrote: ↑Sun Aug 17, 2025 10:22 pm While I strongly disagree with the use of AI in the track building process I do think things like this are a good, non intrusive use of it. You have essentially cut out the most boring and monotonous part of building tracks and given yourself a short cut straight to doing the actual fun part which is designing and building. As long as it's used for coding exclusively you will never see me complain but if it starts imposing on actual creative design like model and texture creation it's not something I can get behind.

I've been playing around with the stable diffusion depthmap script and it's actually pretty impressive! Could definitely come in handy for making normal maps. Not good enough to make a track depthmap from aerial photos though.

jlv wrote: ↑Mon Aug 18, 2025 1:37 am

sethypeety wrote: ↑Sun Aug 17, 2025 10:22 pm While I strongly disagree with the use of AI in the track building process I do think things like this are a good, non intrusive use of it. You have essentially cut out the most boring and monotonous part of building tracks and given yourself a short cut straight to doing the actual fun part which is designing and building. As long as it's used for coding exclusively you will never see me complain but if it starts imposing on actual creative design like model and texture creation it's not something I can get behind.

I've been playing around with the stable diffusion depthmap script and it's actually pretty impressive! Could definitely come in handy for making normal maps. Not good enough to make a track depthmap from aerial photos though.

I've used Shadermap 4 (or whatever it's called) in the past and it was very good for making normal and spec maps. I would however be interested in giving this method a try and seeing if it's a more flexible solution since in the past the norms I've exported have been overly bumpy.

jlv wrote: ↑Mon Aug 18, 2025 1:32 am IMO for anything that isn't pavement or blue groove hardpack you want to use slip_function 2 and add some negative slip values where the friction falls off. The description I used for slip_function 2 in the tileinfo info post was probably too technical. In practical terms, the positive values control the grip braking and turning while the negative values make it fall off for wheelspin.

It is impressive how much the LLMs know though!

Wow, thank you for that! Adding the "slip_function 2" to the tileinfo really made a huge difference; feels way better.

Pasted below this is another tileinfo/friction+erode setting for a really nice feeling traction;
this is based on the one I had in my original post above, but altered and refined even further from ChatGPT.

______________________________
name AdaptiveEvolvedDirt
slip_function 2
friction 1.19 0.0
friction 1.18 0.5
friction 1.17 1.0
friction 1.165 2.0
friction 1.16 3.5
friction 1.15 5.0
friction 1.145 7.0
friction 1.14 10.0
roll_resist 54.0 79.0 36.0
roll_resist 65.0 90.0 46.0
erode_depth 6.5
erode_displacement 3.6
erode_radius 3.5
erode_inner_depth 3.0
erode_inner_radius 2.4
erode_lower 32.0
erode_upper 130.0
erode_off_speed 8.0
erode_on_speed 15.0
erode_slip_base_scale 1.0
erode_slip_scale 1.25
erode_slip_lower 0.0
erode_slip_upper 5.0
erode_min_slope -0.45
erode_max_slope 0.55
erode_project_to_ground 1.0
erode_downhill_scale 0.0
erode_front_scale 0.85
erode_lateral_scale 2.2
erode_limit 0.0 1.0 1.0
erode_limit 0.4 0.92 0.92
erode_limit 0.75 0.72 0.72
erode_limit 1.0 0.50 0.50
______________________________

I think this one feels really good. You can obviously refine it endlessly and have ChatGPT create you frictions that are almost identical but slightly different so that the frictions react differently in the separate areas you paint it. You could also have the GPT refine your erode settings and keep telling it different prompts for the ways you want it to erode in each section of your track. The opportunities/options are INFINITE.

I really think ChatGPT is and can be a great tool for organization; and allowing it to keep a memory of things you tell it to, so you can refer back to it in the future when needed. I use it a lot in my day-to-day life outside of MXS, and it honestly has been a huge tool for just quick information, easy organization, self-improvement, life hacks, creating fun cartoons, chatting about philosophies, having it dive into my mind; I really just love what ChatGPT offers and I am an advocate for the professional, creative, organizational and fun use of "Ai" and "ChatGPT" or "LLM's".

To anyone reading this; Keep learning! Keep growing! Keep living! ♡

For slip_function 2 you want to specify the friction for negative slip speeds under wheelspin in addition to the positive slip speeds for braking/turning friction. Since the values you're using fall off at higher slip speeds you'd probably want to use those for the negative side and then make the positive side ramp up. Something like this:

friction 1.14 -10.0
friction 1.15 -5.0
friction 1.165 -2.0
friction 1.17 -1.0
friction 1.19 0.0
friction 1.19 1.0
friction 1.3 2.0

Try doing some burn outs. If it's slipping too much or it's too grippy, adjust the negative side. Then test braking and turning and adjust the positive side.

This is for loose dirt. For soil that's so hard you don't dig in, the traction won't really be directional so slip_function 1 and only positive slip speeds would be more accurate.

317 wrote: ↑Mon Aug 18, 2025 3:49 amI really think ChatGPT is and can be a great tool for organization

What an amazing read! Great job man. This is a good example of responsible, ethical use of AI. Otherwise, I have extremely Orwellian, dystopian, preloaded fear of what I've seen already. Like everything dazzling, it gets pushed into use way too quickly by questionable people with questionable intentions.