A good guide how to use AI ?

[kirill_7256]

Since Adobe introduced  generative AI  I have never been able to do anything helpful for me   with it.   For instance I need an image  of tire skids on concrete slabs  I could later extract from concrete  background  and use as a decal in a video game .      Covering like  20 meters stripe along  a racing track turn .       Firstly its almost impossible  to force generative AI   to do it in sort of ortho down looking view without  perspective distortion . Then it's always something fancy , never looking right  and realistic .        Whats good it doesn't try to force that recognizable illustrative style other Ai system do   but still  its always a bit blurry  and artificial looking   in always wrong scale    whatever exact scale I am telling in the prompt .

Wonder if AI is not there yet  for such things  or I am missing some  right wording in my promts.    Adobe Sampler does it even worse .   Whatever scale I tell there in ignores it?   Is there way to use some  vector guides maybe  to tell AI how I would like things shaped ?   or some magic word to force ortho  flat view ?    Can I use some condition structure in my prompts   like   a.  image cover 15x15 meters     b. top ortho view     c.  tire skids  from F1 vehicle  folow a circle of 30m    radius or  using  document path# as a guide.    or something like that?    

[Rene Andritsch]

All of the things you mentioned are ways to explore to get the wanted result. A good prompt is the key to a good result. Firefly and other generative AI models are just new tools and they only can get you a result based on their training data. That is why more often than not we as designers and illustrators need the skills to work with various tools inside of Photoshop to composite certain elements together to get the desired result. This might also include to look for (in your case) tire skids on Adobe Stock or other sources and apply it to your image through different techniques like using masks, distortion meshes or other techniques. There is no one tool fits all purposes. That is also – in my opinion – the fun thing of working with creative apps like Photoshop. There is always so much to explore and learn. So maybe in your case try to get a really good stock image of tire skids, extract them and apply them to your image with different image manipulation techniques as opposed to use Firefly to generate it.

[kirill_7256]

Thanks Rene.   Few good points for sure.   

[Trevor.Dennis]

It's still useful to know how we did things before Ai and Generative Fill became a thing.

One idea is to use a brush preset.

Set spacing so each stamp of the brush aligns with the last

Set Angle Jitter to Direction so the brush orintates to the direction of the brush stroke, and to keep it nice, Stroke a work path.  You'll need to mask out the start of the stroke to make it tidy.

This not going to work so well with curved lines, because there will be gaps on the outside of the curve, and obvious overlap on the inside, so best to make them straight, and use Free Transform warp to do the curve.

Now we need to give it some perspective, which we do with Fade (make each stamp of the brush smaller than the last.

Because our spacing is so large, we need very low value for Fade.

We are now left with two problems. 1) there is an obvious step with each new stamp, and 2) the stroke has gaps toward its smaller end.  Remember, the overlap shows if made too big.

For the first we need a new preset with some perspective.

For the second, we have to compromise on spacing, and make the stroke longer than we need, and mask off the excess. That gets us to here.

Place on your road.  Free Transform > Warp > Flag or whatever suits.

Set to Multiply and reduce opacity.

Grop the two layers and use a layer mask to align the ends.

I have used the same method with brush presets to create rope, chain, bike chain presets.  They all work well, but will not go through tight curves because of the gaps, so try to keep the strokes straight, and FT > Warp to fit the scene.

[kirill_7256]

Thank you Trevor  for this detailed explanation.   That not supporting  proper deforming along curves is a big issue actually . Thats why I prefer vector tools .   Loved  decades old Expression  where we could have 10  short tire dabs alternating randomly along a vector path while being perfectly curve deforming and sitting one after another  .  Too bad we don't really have it neighter in Substance Painter nor Designer .  Well , we have too some extent but it so much a pain to use there.    But this is beyond the point.     I am still trying to figure out how modern AI technology could speed up materials creation for 3d graphics  and not do something blurry and unrelated.

[Trevor.Dennis]

Stroking a path is what turns lesser mortals into skilled artists.  Even Bert Monroy uses that trick.

You can place beautifully flowing curves exactly where you want them.  There is no way in this world or the next, that I could draw this tapered line freehand.

I can do my signature squiggle with help from Lazy Nezumi, but it's hit and miss as to where it goes.

[kirill_7256]

yeah, if only it would stay vector and non-destructive 

[Glenn 8675309]

I have no issue with using the generative fill.   It sounds to me as if you are using just the rectangular selection box.    You can use any number of selection options to follow the curves of a road for example.    

Now you just cant type "skid marks" for the prompt: Adobe thinks you are talking about underwear skid marks and wont allow it.










I had to change the layer opacity on this a bit- but it's hard to tell that the tire marks are not part of the original image.


You asked about a guide:  follow the AI links:

https://www.google.com/search?q=ai+guide+to+adobe+photoshop&oq=ai+guide+to+adobe+photoshop&gs_lcrp=E...

[kirill_7256]

Thanks Glenn,  its amazing what you are managed to get from generative AI  but useless for my purpose since I need to do a game decal.  An ortho view from a top  without perspective .   Here what I  am getting

 

Or sometimes something looking perfectly same as my background .     Looks like  AI just doesn't work in  top ortho or aeriel  view or something?   And it's same for everything , not skid marks only.  I am almost gave up with AI  by now. 

[Glenn 8675309]

I understand your problem, and there are big issues.

1. Your original image is very pixellated.  I'm guessing it's due to what looks like you are using a google earth view of a racetrack.

2. You are typing in just "Skid marks"- you can't do that- you need to add "race car skid marks" or "truck skid marks"  just using "skid marks" as the prompt is a "no no"- it's not allowed because the adobe AI assumes you mean skid marks on underwear, and that accounts for your guidlines message.  A better prompt to try "Overhead view of car skid marks".

AI has difficulty with just 2D images--- overhead views often lack contexual clues (such as shadows and perspectives), that's why the images I posted earlier look so good.   Visually, and overhead view lacks many contextual clues it needs to work. 

Now as far as a game decal:   reduce the size of the image- as you make the image smaller- by at least a factor of 3, the pixellation will go away- not just zoom out, but make the acutal image smaller  AI is confused with pixellated images.   Make your image smaller, much smaller (for a decal) then run the generative fill. 



 

[Glenn 8675309]

Given the nature of your orignal image I would have to say that using skidmark brushes would produce better results.

https://www.brusheezy.com/free/tire-skid#google_vignette

[kirill_7256]

Yeah, thanks Glenn.   it's the best I could get .  Still nothing like in your original  "perspective camera " shots.    I mean super blurry and inspecific .   So my guess  I better go  for some vector soft   for the subject .  Hoped Ai SuperZoom  could improve it  a bit   but  looks like it can't.