What's your favorite KID variation?

This is mine. It has to be the single weirdest "logical" conclusion to theory in all of chess. The KID: Advance Anti-Grunfeld, Danube Gambit, Donau Gambit. The Advance Anti-Grunfeld scores decently well, considering how bad it looks. The Donau has been played 3 times, but all 3 led to 6. Nc3, unfortunately.
This variations has:

1. A move that checks the king with a pawn
2. Two pawns reach the seventh rank
3. Three points of material advantage
4. Only two moves in the opening are NOT pawn moves from White.
5. Queen trade and a move that surrenders castling rights are the only non-pawn moves by White.
6. No attempt to develop any minor pieces from White
7. Black not taking a free pawn unless it commits the unforgivable sin of checking the king.

The best part about this? These are all the SF 15.1+ NNUE's top choices at depth 30 (two moves are second at depth 30, but close enough that it fluctuated until depth 30) once we reach the Advance Anti-Grunfeld KID. The best part is: White's winning by less than a tenth of a pawn, according to the engine.

In what universe is this a KID?

1 d4 Nf6 2. c4 g6 is a KID, and then everything after that is Anti-Grunfeld, gambits, and best play. I'm on your side, but technically this is in fact a KID.

It's nothing like a King's Indian.

The true King's Indian is 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. ...(e4/g3/) d6. With typically closed centre style play. What you showed is more like a Grünfeld in play style with more of an open centre.

Also I don't like it at all as white's position looks very ugly. In the final position, white has nothing developed, the king is on d1 and the c1 bishop is tied down to b2.

My favourite KID variation is the Bayonet Attack.

Also the Sämisch.

While positionally Black is doing better, after Bb5+ Bd7 Nc3 Rxa7, White will be up two pawns and should be able to develop quickly enough to defend against any mate threats, in which case the connected outside passed a and b pawns should prove extremely valuable. It's certainly an interesting position, but I'd actually prefer White here, despite being a Benko player myself and therefore theoretically knowing how to play this as Black.

Apparently it's equal but I still don't like this style of play for white. Also the early queen trade. The other thing is that black doesn't have to play the gambit and can carry on like a normal King's Indian amd then d5 is just a waste. It gives black the c5 and e5 squares for no reason and e6 will chip away at white's overextended pawn.

After 3... c5, White is actually just playing a normal KID (or Benoni depending on Black's play) again and scores as such. Funnily enough in the Lichess Masters database, White outscores Black (36/36/28), and there isn't a line where Black outscores White after 3. d5. Fedoseev played this in 2017 against a 2600 and won as White. I'd never adopt the Advance Anti-Grunfeld (3. f3 is much more principled and at least transposes to the Saemisch) but it is interesting that this is playable, and that the triple pawn gambit is also technically within the realms of playable.

The queen trade is absolutely necessary for White because the king is so exposed and necessary for Black for faster development to make up for the material loss.

It's really only a double pawn gambit as black can win the a-pawn back on every single move.

Then according to stockfish analysis black's piece play is so good that the other two pawns also fall to give exact equality.

It is somewhat interesting to me that white has two connected pawns he can't hang on to but it's not that uncommon or surprising to see this sort of thing in benko type positions.

Yeah, sorry, my mistake, it's only a temporary triple pawn gambit.
At lower depths, possibly, but at depth 30 SF 15.1+ NNUE White always has equality if not a slight advantage. However, this is very very playable for Black.
It really does feel like a Benko position, with the difference being White's king safety is in shambles, the queens are off, and Black's c-pawn is gone also.