First of all, come play this thing. Rate it. Give us attention and validation!

I don't think two years ago when @alpacalypse and I started doing this that we foresaw being where we are right now. Six jams our of six where we manage to produce something reasonably complete and three out of three where we consciously aimed for (and got) clean launches without major bugs. We only had two bugs this time around (one of which has been fixed) and I'm not sure that anyone noticed either before we did. Hopefully, we continue the trend of producing something better and better each time.

I'm the one who does the art for our games but I'm not really a pixel artist. I mostly don't do art outside of LD and whenever the jams come up, I learn a bit more and then just try to muddle through by referencing whatever turns up on google and mimicking what actual artists do as best I can. Traditionally, graphics is one of our team's weakest categories. (Innovation was traditionally our strongest but we swung that so far out of our newbie park with 1001N that we've been actively avoiding it, but that's another story.)

Over the last two games, we identified that the weakness was a lack of art direction. Even though I managed decent art here and there, the different parts never really came together. 1001N (LD50) just had pieces of art placed around to fill the screen resolution and we were using totally default sans serif text and filler bar. The icons were designed to be readable but had no stylistic direction. In WC (LD51), I managed to have a tileset but it only covered the walls and some blood splatters while the background floor was just some Paint.NET filter. The item icons I think were okay but some had black outlines while others didn't. The mobs were just whatever came to mind that could be animated quickly and rudimentarily with Piskelapp and had no real theme and the heroine was 62 pixels tall and didn't mesh with either the mobs or the tiles. I'd given her a glowing ring to indicate her hitbox but the fact she wasn't the same size as her surroundings was still one of the most noticeable criticsms.

For LD53, I made it a mission to improve the graphics as much as I could.

After the theme arrived, we rapidly settled on the idea of Kiki's Delivery Service with a bit of social commentary set either in a single tower block or a street with multiple buildings. Because the gameplay would be so simple and the setting so small (in fact, we think we underscoped this time around, which is not a complaint many LD teams have I'm sure) the game would have to lean hard on visual appeal, which was exciting but intimidating. About 40 minutes into the discussion I said "Let's just take it on" and from then the ball got rolling.

WitchDeliver would become our first game with "proper" sprites and "art direction". The art direction wasn't much at all. Initially I only gave myself two rules: -I'll use the default palette of Paint.NET and stick to those colours. I don't know why they are what they are but maybe someone smarter than me picked them for a reason. -Pink and Cyan are pretty 80s, punky and cool.

I looked up how one might pixel a witch and found good examples at 16, 32, 48 and 64 pixel height. I eventually settled on 24. The biggest factor was that in order to have a unified art direction, all the sprites should have the same level of fidelity and 24 was the most I felt I could handle and still produce a reasonable number of assets. That was the next rule - people are approximately 24 pixels high and everything else scaled from that. Hence each floor of the buildings became 48 pixels high and so forth. In order to make procedural generation of pedestrians work with the palette, I imported the whole thing into a bunch of arrays by hue and parts with highlights were separated into two parts, one of which would have a colour two darker than that of the other. This all turned out to be a bit of a trick since apparently Paint.NET and Godot don't quite speak the same language as far as colours are concerned. Thanks to @alpacalypse taking over music and the overall mechanical parts being so light, I was in full Horizontal Mode by 36 hours. One window with no variation or open state became four windows with four variations, each with an open state. Two stores and two doors (from which pedestrians spawn or despawn) became ten and ten, which was enough for us to enforce uniqueness during generation. And barren roofs and alleys came to life with multiple layers of doodads. All in all nearly 300 frames of sprites plus procedural generation of dogs, pedestrians and the street, so we managed quite a bit of variety. Most importantly, having some direction to the art seems to have worked well if the comments are anything to go by. Crucially, some of the most frequently criticised assets were the delivery markers and the arrows, which is great because those were hastily implemented prototype assets that we somehow never got around to replacing (with three hours to go, we decided we needed dogs instead). The fact that people caught onto those flaws is pretty affirming.

Okay, I'm done. Go play and rate! https://ldjam.com/events/ludum-dare/53/witchdelivr