Generative AI to make tire skids decals?

gnoop · Sep 11 2025

Those sort of dark stripes from blocked wheels racing cars do in turns. Could somebody suggest one and what the hell should be the promt wording . Whatever I try come with something pathetic looking . Photoshop can generate something quite nicely looking in regular perspective camera , still unusable due to low res but when I try to nudge it to top camera ortho style image it does an utter crap. Midjourney and anything else I tried do something cartoon style fancy looking or wilde . Chat GPT draws me pictures of something repetitive and so weird looking always on sort of concrete blocks for some uncertain reason having not even a hint it might be of a real world.

Where is all that AI generative power when you need it ?

Those damn decals always looks like someone draw them with one dub brush in photoshop by a low dpi mouse. So I hoped that's the areas where AI would shine.

dimwalker · Sep 12 2025

That's exactly my experience - neural networks are great when you don't care about results, but if you want something specific and for it to have high enough quality for production it's just pure disappointment.

Try to generate closest thing you can with photoshop and then try to improve the quality/resolution by upscaling and inpainting with MJ or SD.

gnoop · Sep 12 2025

dimwalker said:

with MJ or SD.

Well , Photoshop at least trying . Here is what Stable Diffusion does:

Image: https://us.v-cdn.net/5021068/uploads/editor/27/tq13gkwn98my.jpg

Is my wording bad or something? looks like only things AI is comprehending are tire and racing track .

I started to do SDesigner generator with its curve nodes but it turns to be a hell of an artificial look . reseeding them to death. Feel I am falling into bottomless hole and will waste forever on this. So need a simple quick way to do naturally looking variations .

dimwalker · Sep 12 2025

gnoop said:
Well , Photoshop at least trying . Here is what Stable Diffusion does:

To clarify - I meant inpainting, not generating from scratch. As in get generation from photoshop (kind of what you want, but poor quality) and then run THAT through other NN of your choice or maybe PS again. Idea is for NN to remake existing guiding image with higher details, quality, resolution.
With SD you can use comfyUI to get more control over the process. There are installers that take care of most tedious convoluted parts and dump ~60Gb of stuff on your PC. You would still need to figure out how to work with its node system. Or if your are lucky maybe someone already made one.

Imho SD generates not as pretty as MJ, but can run locally so NDA-friendly if you care about that. You can use controlNets that allow you to guide images by depth masks, contours, style example image etc
MJ is much easier to use and does a better job, but usually not what you want.

Designer graph will probably give you best result if you are willing to put a lot of time into it. But you will have full control over every step and you only need to make it once to get guaranteed coherent results feeding it different splines as input.

Joopson · Sep 13 2025

gnoop said:
Is my wording bad or something?

It's not great. I think if you wrote that prompt to a human artist, they'd have questions and need clarification. The human perspective is what I'm coming from here, so, there's no guarantee a humanly better worded prompt will get you what you need, either.

I'd personally use "top down" instead of ortho (which is short for orthographic anyway), and isn't really a common term outside of 3D.

Maybe something like "Top down view of tire skid-marks" or, if the turning aspect is important, "Top down view of tire turning skid-marks", and tack on other important words I may have left out (racetrack, nascar, whatever)

Noren · Sep 13 2025

gnoop

Those damn decals always looks like someone draw them with one dub brush in photoshop by a low dpi mouse. So I hoped that's the areas where AI would shine.

Well, it does sound like a good case for a couple custom brushes. Brush size / resolution might be a problem, so upscaling would be the application for AI, here. Maybe use ZBrush or a different software with more controls for how the strokes are laid down.

Other than that, make sure to add terms for what you are actually expecting to see and in which context this would have been part of the training data, e.g. texture, decal, brush pack

pior · Sep 13 2025

Perhaps instead of skid marks, you could ask the plagiarism machine to generate some pictures of shit stains ?

And as an added bonus, no further prompt engineering will be needed to get some clean alphas on a white background.

Eric Chadwick · Sep 14 2025

lol perfect!

zetheros · Sep 15 2025

why would you want to use AI to generate skid marks... this sounds like a spline based decal placement job in UE

gnoop · Sep 15 2025

pior About plagiarism, human beings do it with equal if not greater taste for that. I understand the frustration but it's like fighting wind mills IMO.

zetheros . Well It's not so easy task really. Especially when you need to do many variations . Last time I do them I used Creative house expression allowing up to 10 randomly alternating non-square bitmaps + start/end caps smoothly deforming along its vector spline. The software is long dead and I lost my skid-mark brush set . There is no modern alternatives . Anything painted by both Photoshop and Painter looks instantly artificial and not true always lacking a true feel of car physics and flow I hoped a machine could recognize and distinguish. Yet Photoshop Ai makes something eerie looking with a feel of Giger's works.

Image: https://us.v-cdn.net/5021068/uploads/editor/av/q5588m2zev0z.jpg

BTW, thanks for the tip Joopson looks like top down does indeed works better yet it still try to mimic perspective camera each time .

Eric Chadwick · Sep 17 2025

This seems like it wouldn’t be too difficult to create with a randomizable Substance Designer graph.

okidoki · Sep 17 2025

Maybe simply type: 'skidmarks transparent' into a "tool" available for several years now ? An internet search engine and select the option 'images' :

Image: https://us.v-cdn.net/5021068/uploads/editor/bo/gu4w8170chm8.jpg

Eric Chadwick · Sep 17 2025

Don't try to source rando images off the web for your textures, unless you're positive the image license permits it. Your lead or AD will have a fit if they see you sampling rando web images, opening the company up to copyright infringement.

Neox · Sep 17 2025

Eric Chadwick said:

Don't try to source rando images off the web for your textures, unless you're positive the image license permits it. Your lead or AD will have a fit if they see you sampling rando web images, opening the company up to copyright infringement.

sure, but not knowing which source was ripped off and jumbled up is better?

okidoki · Sep 17 2025

Ohh.. i implied to not only use some license selection option on the search engine but also double check this on the site.

Also.. found this almost the same way:

https://www.youtube.com/watch?v=6EEvSMXzSv8

https://www.youtube.com/watch?v=mTN5a-coJCA

https://www.gameassetdeals.com/asset/82419/ms-skid-marks-system

https://dev.epicgames.com/community/learning/tutorials/lMxR/unreal-engine-car-skid-marks-tutorial

Mostly it's about "painting" the tire profile on the asphalt and in curves maybe make this a bit more blurry..

gnoop · Sep 18 2025

Thanks okidoki I need something like in your last google image search example. 2-s decals laying next to each other. Using particle paint shader is a nice idea actually but I am afraid it may require to recreate the whole game physics in Blender with dozens of cars . Probably would be easier to ask our programmers . So far I think generative AI is totally useless for the task .

It's just we got an encouragement to use AI and I look for something it could may improve in art side of things. So far I couldn''t . Much to my surprise.

okidoki · Sep 18 2025

This isn't so much a surprise.. in fact i first used some other search words before i thought.. hmmm maybe add transparent because decals wasn't very sucessfull.. also it often gives some "markings in underwear".. so even before search engines used AI it was (and still is) sometimes not easy to find some suitable results.. often the results also suggest exactly that what you explicitely want to omit ( usinf - before the search word).

So if there is not enough data which provides good context maybe simply because you do not know some other term for it or it wasn't search before and there is no previous build search context you will not find some good results. This may have been changed some days or even week later.

Also AI is a product which is presented in the best light.. remember when scanning your disk with scandisk on windows.. 30 years ago ??

> The scan did not find any errors..

..while your disk is screwed ?? It's the same with AI.. it will never tell you:

> I only got this bad results because i may not completely understand your request.

In fact you can see this a bit when you ask the AI to do it a bit better and when it start answering with some apology. So it "understand" that it has answered badly ?? And now apologize ??? It's some psychological trick to not make go away and continue to use it.

Ever thought about when the time comes and everyone have to pay for some AI work.. of course after noone can do this thing what differs us humans form the animals... thinking. (At least we think so )

This is a good example to tell those managers (?) who told you to use AI: "Well i can but then it does cost you more of my time and if i'm not wrong about this you initially did hire me because i can do this and this is not any kind of insult but you.. not ? But.. who know maybe someone will have the idea to replace the managers by some AI.."

But.. of course this all depends on the circumstances.

Two Listen · Sep 19 2025

Decals of this nature seems like ~30 minutes of fiddling in PS with a couple custom brushes (that you could probably make from scratch pretty quickly), that will net you some brushes you can use forever into the future, and easy high-res source files to tweak as needed.

AI is good at what it's seen a lot of. Examples collected at enormous scale, made better with various LoRA that are almost always tailored to more concept art/marketing/lofi waifu shit. It struggles with concepts, things that either don't exist or are uncommon, and particularly with perspectives and assets we might seek for development. You want a nice looking piece of environmental concept art showing a cave entrance visible through some brush in the foreground, you can probably get it. You want a nice looking cave wall - from the perspective of someone inside the cave just staring at the wall, extending across the canvas so you can use it as a base for a tile? You're going to be in for a shock when every generation features the entrance to a cave, because probability dictates that keyword "cave" means you probably want...a hole in some rocks, as it shows in most of the (probably stolen) images it's been trained on featuring the entrance to a cave with that token.

You're going to struggle to get the specific angle of skid marks/tire tracks you're seeking for the same reason AI can easily generate a mobile game illustration of a "hammer" and will have absolutely no idea what to give you for a "faceting machine".

I've had to make sure all of my artists understand - AI can VERY easily waste more time than it saves. If it can take 6 hours of grunt work you really don't care about and turn it into 10 minutes - go for it, the company's paying for licenses and we're expected to use it. If it's going to take you an hour struggling with prompts, tweaks, refinement, further upscale/prompt edit iterations all for a rasterized piece of shit, but you could make it yourself in 1hr 20 and have clean source that's easy to work with? Just do it yourself.

If I were looking for images of this myself I wouldn't look for skid marks, I would look for a car actively skidding around a racetrack from a top-down view, because it's more likely that AI will understand the scene of a car doing zoomies than it is to understand the concept of the skidmarks left behind, and I would expect to have to further upscale/manually mask out and clean up the skid marks after the fact. And acknowledging that effort, I would probably opt to just go make them myself instead.

gnoop · Sep 19 2025

Two Listen thanks for your comment . it's really that way . Mostly a waste of time and nowhere close to replace people here.

As of 30 min Photoshop work it's sort of shows itself in many car games . Manually painted skid-marks , totally out of physics context when wheels gets different pressure, diffrent side of contact surface doing that skid and diffrent path bending and sadden of track zigzags. Often very artificial and hand painted looking.

It's a task not any artist capable to do , requires lots of subtlety feel and taste to look natural . Photoshop is very bad for that IMO . I use vector soft .. Did gazillion versions and use really just a few where it clicks well .

Generative AI to make tire skids decals?

Replies