what is this i don't even

[Originally written July 17 2020]

While I’ve worked out how to determine from a creature’s genetics what its actual color values will be, and I have a good understanding of how RGB itself works, I’m still missing a piece of the puzzle: how to predict exactly what the visuals will look like from those numbers. As demonstrated in this image, color rotation in Creatures 3 clearly does not correspond perfectly to that in Photoshop; the second Norn from the left has a rotation value of 191, which is exactly 1/6th of the way around the color wheel. The same amount of rotation in photoshop produces the third Norn, who is much brighter than the one in-game, although her hues are correct; the fourth image is the third one with the saturation cranked way down. So how exactly do the colors get calculated?

I decided to do some experimentation to find out. Rather than hatching edited creatures though, I simply played around with the TARG HOTS TINT command and a beach ball. Specifically, I recolored it multiple times and took screenshots, then used photoshop to find the RGB values of four specific pixels – one in each major colored section of the ball. 

This massive pile of data turned out to be less helpful than I’d hoped. For one thing, it showed that the rotation was not complete – in theory, at a value of 255, the blue channel value should equal the original value of the red channel, and so on. My data showed that it usually ended up being slightly less than the expected value – except sometimes it wasn’t. And while it usually did end up being less, I couldn’t find any pattern to how much less – not by straight value, not by percentage of starting value, not by percentage of theoretical target value. For another thing, the graphs were… weird. Wobbly. I couldn’t make heads or tails of them no matter how I tried.

I decided to examine the blue channel from the point in the red section of the ball in more detail. Unable to comprehend a pattern to it, I decided to gather more data for a finer-grained look at the 176-184 section of the graph. I repeated my recoloring and plotting procedure for every possible value in that section. It only confused me more.

There is a definite stair-stepping pattern here. A channel will remain at a particular number for several rotation values, then suddenly jump to a new value, and remain there. But the channels do not all jump at the same time. I clearly needed to increase the resolution of the whole graph, and who knew how many times I would need to do so before I could get a truly accurate picture of what was going on… so I bit the bullet and started graphing every single value until a pattern emerged.

Red Point				Yellow Point				Green Point				Blue Point
Rotation	R	G	B	Rotation	R	G	B	Rotation	R	G	B	Rotation	R	G	B
128	198	48	24	128	255	251	41	128	16	65	16	128	33	24	82
129	189	44	24	129	255	247	41	129	16	60	8	129	24	24	74
130	189	44	24	130	255	243	41	130	16	60	8	130	24	24	74
131	189	44	24	131	255	243	41	131	16	60	8	131	24	24	74
132	189	44	24	132	255	239	41	132	16	60	8	132	24	24	74
133	181	44	24	133	255	239	41	133	16	60	8	133	24	24	74
134	181	44	24	134	255	239	49	134	16	60	8	134	24	24	74
135	181	44	33	135	255	235	49	135	16	60	8	135	24	24	74
136	181	44	33	136	255	235	49	136	16	60	8	136	24	28	74
137	181	44	33	137	255	231	49	137	16	60	8	137	24	28	74
138	181	44	33	138	255	231	49	138	16	56	8	138	24	28	74
139	181	44	33	139	255	231	57	139	16	56	8	139	24	28	74
140	173	44	33	140	255	227	57	140	16	56	8	140	24	28	74
141	173	44	33	141	255	227	57	141	16	56	8	141	24	28	74
142	173	44	41	142	255	223	57	142	16	56	8	142	24	28	74
143	173	44	41	143	255	223	57	143	16	56	8	143	24	28	74
144	173	44	41	144	255	223	66	144	16	56	8	144	24	28	74
145	173	44	41	145	255	219	66	145	16	56	8	145	24	32	74
146	173	44	41	146	255	219	66	146	16	56	8	146	24	32	66

One thing that popped out immediately is that this is a stair step function – it stays at a value, then simply jumps to another one without a smooth transition, then stays at that value, and so on; I marked the points where the value changed with a bold number in the table. I was starting to see something of a pattern emerge, so I color coded the numbers by how much the value changed. I soon picked up that the red and blue channels changed according to a pattern: three changes by 8, then a change by 9. After additional graphing of the yellow point (which had the most green channel changes) I was able to nail down that pattern as well: fifteen changes by 4, then a change by 5.

This sort of pattern in computing is generally indicative of computing integers (which slice off any decimal points rather than rounding) with math involving non-whole numbers. In this case, the pattern of change for the green channel is consistent with that generated by Floor(4.0625*n) and the blue and red channels consistent with Floor(8.25*n). But how to predict when the jumps happen? And why does the blue channel in the green point, which should not change, go from 16 to 8 and then back to 16? Pseudocode for how the colors work was available on the official website but it’s obviously incomplete and probably completely wrong. Perhaps I’ll come back to this another time.