LÖVE

@Jasoco, in Bandana, we used Regular Expressions to find which tile should be used at a certain position.
We wrote about it here, might be an interesting read.

In my autotiling case I just run the algorithm described on the linked site twice: once to get the cardinal sum factor, and then again to get the diagonal sum factor. This requires 31 dfferent variations of the tile for the corner checks.

However, this is not pefect (though is good enough for my purposes), and one easier method that doesn't require as much manual tile work is to separate a single tile into four parts, and each part checks the surrounding tiles independently.

I'm thinking of picking up the LOVE port of my roguelike back up! However I can't figure out how to force LOVE to redraw my map once I change the level...

MrFariator wrote:In my autotiling case I just run the algorithm described on the linked site twice: once to get the cardinal sum factor, and then again to get the diagonal sum factor. This requires 31 dfferent variations of the tile for the corner checks.

However, this is not pefect (though is good enough for my purposes), and one easier method that doesn't require as much manual tile work is to separate a single tile into four parts, and each part checks the surrounding tiles independently.

Due to the way I wanted to account for every single combination, I had to do a lot of edge case checks for certain tiles but I got something that works.



Mine has 49 different variations for tiles to account for every single corner and edge combination. (Which you can see in the screenshot above) It's probably overkill, but I want it to be flexible.

49 isn't ultimately too bad. My initial ideas for covering all possible cases went well beyond that, and it's not like I really need that kind of accuracy for my game. The auto tiling in my editor is there for making placing big solid chunks easier, while requiring fine-tuning corners and other things by hand. This is because the tile sheets I'm intending to make later down the line would be too detailed/specific (think Mega Man X on the SNES) that it would become tiresome to develop a more sophisticated auto tiling scheme when I just want 16x16 tiles.

But guess that proof of concept has to wait till I actually get around making the game prettier.

Turns out it's actually 47. I did two tiles twice. (Bottom right corner of the tile portion of the grid.) Though my posted tile sheet is also missing 8 tile variations.

I keep finding more edge cases. Ugh. I think I might have to write a new algorithm of my own.

Well I did it. I made my own algorithm that deals with micro tiles of 8x8 instead of 16x16 and it ended up taking up way less code. It basically checks the four corners of the tile and compares which neighbors are water or not and uses one of the 4 possible variations for each corner. I cut myself down from a redundantly impossible to manage tile sheet like the one on the previous page:


And replaced it with a much more manageable and smaller sprite sheet with only 12 variations:


And it works perfectly and covers every base. The algorithm also returns whether the tile has no solid neighbors so I can add details like the rock pictured above or maybe a lily pad or something.

It'll be able to make changes much easier and add variations to randomly be selected. This will work for ground-based autotiling. I'll write a different one based on the original code for wall-based autotiling which will probably still use 16x16 tiles.

Edit: Now I added some "roughness" to the tiles to make them look more natural. Kind of. I like it...



Edit 2: Experimenting.

Trying to make my own genetic algorithm Box2D car simulator. The physics engine will automatically weld nearby vertices so if you create a triangle it can degenerate into a straight line. Now it could act as a line segment but still being treated as a would-be polygon, it gets rejected. And then it also requires that resulting shape had some substantial volume and if the triangle gets really flat it throws errors as well. Picky bastard!

At least the genetic part is working and for as long as Box2D doesn't crash it functions properly.

Been trying to figure out a good lighting method and I came up with something that almost resembles Metaballs.


It's not even that difficult either. I have an image of a gradient circle that fades from solid white in the middle to fully transparent. Draw that as the light to my canvas. Run the canvas through my shader which checks the Alpha of the pixel and "snaps" it to one of a few fixed values.

My shader is simple enough, but I want to do more: Like add a dither or something. I don't know yet. But it looks really cool when moving around. I'm not going for realistic of course so I don't need shadowing or blocking the light though that'd be neat. I'll figure something out in time.

Jasoco wrote:Been trying to figure out a good lighting method and I came up with something that almost resembles Metaballs.

It looks cool, but I think it breaks the style a little - you have rectangular tiles everywhere and then this very round shadow. Maybe you could make it adjust to the tiles, i.e. a single tile can either be fully in the shadow, in the semi-shadow or not in shadow at all.
That way the shadow outline would be blocky, just like the tiles...