The BRAWL² Tournament Challenge has been announced!
It starts May 12, and ends Sept 12. Let's see what you got!
https://polycount.com/discussion/237047/the-brawl²-tournament
It starts May 12, and ends Sept 12. Let's see what you got!
https://polycount.com/discussion/237047/the-brawl²-tournament
Generative AI to make tire skids decals?
sublime tool
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.
Replies
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.
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.
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)
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