Braid – Director Commentary with Jonathan Blow

This year we’re beside ourselves with excitement to announce that Jonathan Blow will be at GameCity 5 in person to deliver a personal account of making Braid. This will be a complete play-through of the game, with Blow commenting on the design decisions he took during its gestation and of course responding to questions from the audience. Like much of the things that happen at GameCity, this hasn’t been seen before – and this is your chance to be there when it happens.

Jonathan Blow.

My Thoughts

Jonathan gives a blow by blow (ho ho! I kill myself!) account of many puzzles in the game; and goes a little into the “story” of the game (but not much; best guesses on a postcard!), including clearing up what the cloud in the last screen is.

Yes, it has persuaded me more that I should like the game a little more and give it further credit (since the failings are more that I don’t particularly enjoy single-solution puzzles with no freedom); certainly seeing the prototype he made in 10 days in action was neat too.


Using an early prototype to show what changed, version 0.04 – 8 or 9 days of work. Had a mario-esk guy, a square person, important for gameplay reasons. Easy to miss jumps too, but there is a full experience.

You die in this version and it restarts you at the beginning. The time travel is in there, in world 2 – you had to get through a normal world 1.

The rewind is in world 2, with some hard jumping puzzles for using the rewind to get rid of mistakes. Goes on to introduce clouds (which loop), keys and doors, canons with enemies, and levels which do get reused – canon levels, falling pits – leap of faiths, and boss level.

Some design things; old keys were not special in any way so “World 3 keys don’t travel in time” worked because it was a short game. Another thing was some levels were made smaller to fit on one 16:9 screen for the puzzle. The essence of the game was there in 8 or 9 days of work.

Another difference was the time forward/back mechanic was done by going up/down, instead of left/right (so jumping moved time). It was cool, but it didn’t feel right to keep it in the game. It didn’t add enough. Also platformers normally go left to right so it’d have to have lots of vertical space, which is not interesting.

Rewind was also added in the first level. The main difference between the prototype and the finished game is 3 years of development; most of it was similar though.

The Game

It was important to have the game start immediately, with the Braid title and the music is sad, and hopeful, different to many videogame music pieces. The opening level is also really complicated; there are lots of hidden elements, particle effects and changes.

When people looked at what the game means people seemed to focus disproportionally on the text of the game rather then the game design. I’m going to focus on the game design.

Compared to the prototype, the text is broken up and also not made needing a key to skip. You’re starting immediately with interaction which sticks with people through the entire game; so don’t want players to fight it. Is why the intro level was worked on so hard.

The books before opening the level was the closest that could be got to having the story in the game.

The opening level with rewind only opens like many classical platformers – a blue sky and childhood feeling, but with an adult theme where things are not that simple. Making things less clear cut. Nothing to click through on the tutorial level, just need to follow the game. The game is intrinsically motivating – didn’t know the term then but was making the game interesting to play in itself.

Was important to show jumping up below platforms, that canons work, and so forth. Some things I don’t like about the level. The first appearance of the puzzle board; you create a platform inside it. However the frame of the frame looks like it can be jumped on. There are lots of mechanical issues that mean the art style is the way it is. Wanted to avoid confusion like in many other games; but there is confusion when solving puzzles. Need to have no way of getting confused from external things; the background and foreground must contrast, lots of requirements for the artist. Foreground has higher frequency detail where you can touch, the background or places you can’t get to are drawn to lower fidelity. This is consistent throughout the game. A philosophical way of showing clearly what can be touched.

The character art; some people don’t like it. He is 2 and a half heads tall, the art is that way mechanically. If a real human proportions it is harder to move that shape, since the feet are so small. It works well which is why people keep doing it that way.

Another issue with the puzzle board you move to do a platform is that you couldn’t get the second puzzle piece before going through more of the game. Wanted to have hints or other puzzles later remind you of earlier ones to go back to. Problem was when playtesting the game, a significant number of players, half, would be hardcore and not progress in a game if they can’t solve a puzzle. Had to force people to not solve the puzzle. Not a big thing, but now the issue was people didn’t come back. It was a late decision and waited too long; so the lesson is don’t playtests your game. Didn’t do it for much of it!

The first puzzle had you not able to complete it if you kill every enemy as you do in platformers. There is an impossible jump carefully designed. The gap was made very wide. There is a missing information thing. You have to leave and come back or remember you have a guy there you killed. I like that because it is subtle. It also says “Hunt” with 6 monster icons. These are individual icons, and individual x’s, which used to instantly change to X’s – but had to be changed to fading in and out, for later time puzzles.

Canon levels had 2 in 1 puzzles; the canon shoots something you have to jump on. Rewind destroys rhythm that most platformers have, but don’t give you clues like the fuse.

Whole of world 2 (the one after the title screen level) is all a tutorial. Would be hard to put the complex ideas in later without the baseline.

Originally gameplay functionality reasons had the backgrounds distinctly different. There are exceptions; the doors need to be walked through not into, but need to be distinct since they are usable. Found that was driven to make things simpler, which makes you understand something fundamental. Nothing obfuscated so a maximum hit. If the level is full of complex stuff, just to get to the “get the key out of the pit” then it is less fair, rather then just having it there right in front of your face to work out.

With Braid people would get stuck with hard puzzles, they’d go away, eat, have a shower, and then come back and know how to solve that puzzle. There is something about the simplicity of the puzzles, the foreground and background contrast, allowing your brain to store it easily.

Improvements on the original design include not just keys being magical; but clouds and doors can also be magical. So a set of clouds can be phased back. It adds some interesting behaviour for the simple rules like the penalty for missing a jump off magical clouds makes it harder to get back to take the jump. There were a lot of interesting consequences from combining rules of time travel. The more practical application of rules of objects also helps; no looping clouds, but instead they go poof when they hit an object. Don’t need to say “Oh, they wrap and it’s the same one and so I need to do…”. Clouds didn’t usually fire from cannons – a post prototype decision.

A very minimalist drive in Braid but there are exceptions, the level with the rabbits – when they’re first introduced – is full of monsters. It went against other levels being minimalist. The aim was to investigate how time behaves, not introduce key making. There are twists in each world, even when things don’t make sense like platforms appearing from no where for a single puzzle.

In between levels, the non-verbal doors involve a number for the puzzle pieces left and an icon describing the contents of the level.

Leap of faith is a key part of platformers. Early platforms were unfair in certain ways; to see what was down a pit you had to possibly die to see what was down the pit. The one in Braid is called Leap of Faith because with time rewind you can kill the player a lot, and in fact people die a lot.

There is an irreversible puzzle – one you can fail. It is even called The Irreversible, as a clue, but also to do with the artists comic before working on Braid. The puzzle involves not rewinding, just waiting; for the large block to come out and stop the smaller one. The aim being that puzzles, if you know how to solve them should be easy to solve.

It’s not true that World 2 is the only one with irreversible puzzles, where it seems odd because rewind is meant to allow you to never fail a puzzle. The rest of Braid is exploring the consequences of the rule changes. Not arbitrary ones, but exploring ones fully. It is interesting having the irreversible consequence. It would be an incomplete game without this level. Compared to lots of people there is not focus testing; the level is not that fun and it is adding completeness and integrity to the work, and some depth.

The 3 door key puzzle involves trying to realise that you’ve done the wrong thing, even though you can rewind them entirely. You pattern match it with the first 2 door puzzle, and can solve it, but many people don’t and get frustrated.

The end of the level, with a cannon and rabbits at the exit, is not too cool – redoing the game would remove those things, they’re annoying and unnecessary, even if the last rabbit by the door is neat.

The art and design for the boss involves much the same as the prototype. Each scream for a horn being taken off is different, each X is different. There is a lot of uniqueness.

For the music, repetition versus adding something to repeats of the notes. Reprises of the music, a rather dreamlike way. The game has a bit of a sadistic streak. The monsters look a bit grumpy, and sort of cry when they’re dying. Some lap of clarity into the classical platformer thing; you might feel bad about jumping on them. The monster has to die with the pit – what has to happen has to happen, he doesn’t die when you have the key because of the rewinding but has to die when you reach the door.

The Donkey Kong level. There are some neat ways around the fireballs, but the thing that annoyed people were the two doors. Most people are sleepwalking through it; you get kind of tired. It should be an alarm bell that there are two gates and that trying to open the right hand gate doesn’t work. The rules of the universe work like that, it wasn’t intentionally programmed in like that.

The reused jumping puzzle (a reprise of the previous one) but with back/forward time has the icons fade in and out rather then instantly change so you could see the change as you stepped through things.

The crazy two layers of cloud level, not asked many people about this. But it was a philosophical question; what happens when you ride a cloud you are moving on when you are moving through time too. Your speed increases when you’re on a cloud that is moving. There is an equation you can solve which is in the code to the game to increase your speed. 90% of players don’t really think about that. This game is about the rules of the universe dominating; so the music in the level might sound bad but in real life there isn’t music emanating from the world. The choice with the music is necessary to link to the time which is fundamental. Special relativity doesn’t hold in the world of Braid, if you cared. But it is the kind of thing you want to think about when you’re designing a platformer! 🙂

For keys, up to a certain point there are only keys which are sparkly. If you don’t have a sparkly key you need to see how it works. A special set of ladders mean you have to go to the left, which according to the universe rules moves back the key through mid air.

Another reprised level involve a parallel universe using shadows to display a different set of things.

The first level with no subtitle but only an icon of a guy next to a pit; you have to have your shadow die. Originally rather then a pit you die in you had a whole in a wall which you’d never seen before. That is why there is not title, the feeling of doing it with a pit of death and how well it worked couldn’t be put into words.

The reprise of the boss level; had a different way of solving it with shadow chandeliers.

In World 6, a reprise of the pit level, but with a twist. Redoing it with a puzzle piece at the bottom it makes it mundane and less special. This one hasn’t got that but has got the ring to slow down time. Having a feeling for how important the ring/Y can slowdown near it, with snowflakes slowing down – being next to the time vortex thing. The weather one was programmed separately to the other levels, and uses rewind data which is a scare resource, but not many people notice it.

Adding also complex interactions too; there are some intertwined puzzle elements with canons and monster canons. There are some parts also designed knowingly even though it wasn’t that fun. There is some brute force method to slowing down the up/down plants, with the ring. The major experience is this. The best consequence though is doing a lot of rewinding which makes it a lot easier. This is what speedruns are for; to make people have this kind of understanding. The speedrun is not just making things harder, extending playtime or doing things faster. It is having the deeper understanding of the puzzles. It made for a worse play experience for most players, but it clearly made the game deeper, only thought about it for about 10 minutes. Also myself came to the conclusion after playing it for the non-brute method; which was a wow moment to me.

There are less minimalist but more interplay between elements, like platforms interacting with the ring.

The second to last level went through some permutations. It started out much more silly. The vertical canon worked well as a break to the pattern.

The final world; there are some levels that lead up to this. A cutscene that plays. It is just like Donkey Kong; saying “Help”, so it is like that. There used to be bananas in the pit with the knight, but were not reimplemented by David. From the “Help” some dialogue came – like a palindrome where forwards it looks like the knight is getting angry and she is running but in reverse she is asking for help and escapes up the vines with him.


Jonathan: Won’t answer about the stars, so don’t even bother asking!

Audience: Is it intentional that the two doors above the 3 door puzzle intentional being off screen?

Jonathan: The camera focuses on the main task at hand, the actual puzzle could have been broken off but was thematically the same so was separated a bit. Having too much irrelevant information and seeing the pattern mean it would have been moved.

Audience: Why do some flowers contain rabbits and some not?

Jonathan: The rabbits in code are mimics, and so are mimicking perfectly nice flowers. Earlier levels have no mimics so the why is just a judgement call. It didn’t feel right having all flowers being rabbits. Adding that feeling of newness in a level. Segwaying into a topic, with hindsight it is a technique of dramatic climax paying attention to ever flower. (Skenarian?) This is where I get int rouble but playing a game made by people who don’t understand RPG’s like some barrels always containing the same amount of coins; it isn’t an ethical choice really, they don’t understand that there is some dramatic arc from getting more or less coins.

Audience: About the epilogue?

Jonathan: There are tons of pattern breaks. The world is falling apart. Violating rules like foreground and background objects being walkable. How willing and what percentage of players realise there are puzzles there? There are no rewards for the puzzles there.

The cloud in the level is behind conspiracies; because there a reason for everything. The cloud is for having a vantage point to stand on; the castle doesn’t just have blocks but something antagonistic to me available, and looking down on the text. It’s a pattern break. Everything in the game means something apart from this. Most of the patterns are consistent until they are broken. There is no gameplay purpose to this cloud.

Audience: The Oppermimer quote in the Epilogue, is it important?

Jonathan: With story, not explaining it here, but every part of the text is important. It took 3 years perfecting details. Not going to talk about it much here; the reason I make videogames is I want to make some art things and communicate certain feelings. Is topological right? It often is very hard to verbalise. The sort of thing doesn’t work well in linear language. Even though there are text bits in the game, I think of the game as a whole thing. There are perhaps things I could have done better but I don’t know even years after. Is the story there for everyone to have their own interpretation? It is counter to reality to have this, but at the same time you have a story you intend. To use an analogy, Gravities Rainbow, very complicated and I don’t understand very much of it but I’ve got a lot of it. Reading cliff notes and re-reading it I might get even more out of it. There is a continuum of levels you can get out of the game; from gameplay not the text, which you can just run past. The game means a very definite thing to me, the story elements even if they were very crazy. The design decisions were made in 2005; Gish was 5 or 10 thousand copies. Piling on text for a culture of players who probably don’t want to read was important. A long way to not really answer a hard question.

Audience: At what point you knew you’d spend 3 years of your time on it?

Jonathan: Even 2 years into it I thought I’d be done in 6 weeks. Always 6 weeks, bang out all the art, it’d be awesome. The final game is way better then envisioned; the final game had nice art. There was a programmer ready version in December 2005, the ending was there and so forth. Totally be lying if it was said it was worked on full time. Got really burnt out; don’t work too hard. The game benefited from the calendar time; if it could have been squeezed into 1 year, it would not have been as good. It is not easy, having a lot of reworking of space and puzzles. Lots of not obvious things that took a long time to work out. Don’t have a way of saying it’d be not better if it was focused though.

Audience: What time did you come to the decision to go with the emergent rules?

Jonathan: It is usually the philosophy for 99.99% of game designers to get hammers to the design and chisel the design to get it into shape. But found that the 4 or 5 days to prototype, really fast when rewind time was chosen; a really rich subject matter. Because I was the programmer and the designer at the same time, some things came from just looking at the code and lots of “What happens if I do this…” and “What happens if I do that…” and so forth.

One thought on “Braid – Director Commentary with Jonathan Blow”

Leave a Reply

Your email address will not be published. Required fields are marked *

Website and journal of Andrew Armstrong