Interview of Lord Tom for GamersGlobal
This interview was answered by Lord Tom for a German gaming website called GamersGlobal in August 2010. The interviewer is Christoph Hofmann. Links and formatting were added afterwards.
Tom, you’ve been TASing since 2007. What still motivates you?
I like games that I played or at least had heard of as a kid growing up. Games where there is complicated route-planning or item-usage management are most interesting to me. There's a big thrill in finding new glitches and techniques that no one else has encountered, in all the millions of times people have played these games in the past 20+ years. Finally, ending up with something that's fun for me to watch, as well as for other people to appreciate and comment on is a big part of it.
And what introduced you to the scene?
Back in college, in 1998 or so, I spent some time trying to beat Super Metroid as fast as I could. On a whim one day, I searched online to see what the world record was, which led me to discover SpeedDemosArchive.com. I spent a few months eagerly watching those movies, but not really considering contributing since I didn't have a console and was never very good at playing games realtime. During this time, I kept seeing references to TAS's, but didn't know what they were. Finally, I followed a link to TASvideos.org and was blown away. I was especially intrigued by NES Legacy of the Wizard, since it used a technique of combining speed scrolls that led to truly ridiculous speeds. Poking around, I found a very minor improvement in one of the early screens, which led me to jump in and complete my first TAS.
Which you have improved since then, too. Your most famous run however is the TAS of Super Mario Bros. 3 you did together with Mitjitsu. What prompted you to run that particular game?
I'd watched the Super Mario Bros. 3 runs many times, but I realized (since it had been years since I'd played the game) that I had no idea what was going on with all the glitching through walls in the final castle. I found an online map, and analyzed the route the published run used, and was surprised to see that it appeared longer than an alternative route using an entirely different path. I was initially stymied - the route required jumping off a wall with very little speed, and using an invincibility star to get past one of the bad guys in the castle. Finally, though, I was able to get to Bowser, with a time that beat what was up there, so I put my discovery up on the tasvideos forums.
But a TAS to Super Mario Bros. 3 isn’t just any run…
I was nervous about taking on one of the best-optimized and most-visible runs on the site, but with the Metroid and Rygar runs under my belt, I felt up to it. Unbeknownst to me, Mitjitsu was already at work on the run, and had developed an improvement to World 1, so he contacted me and suggested we work together.
Mitjitsu did most of the primary TAS-ing, whereas I found improvements and developed new techniques. For instance, Mitjitsu had performed the "floor glitch," where Mario drops into the floor to get pushed ahead of the screen on an auto-scroller, at much lower speed than was normally required - but he hadn't been recording, so he didn't know exactly what had happened. I hammered away until I discovered how to do it consistently. Overall, though, it was very informal and collaborative, and each of us pretty much worked on every segment.
As you mentioned, your route is only possible because of the newly found low-speed wall-jump glitch. How did you find it?
Like many glitches I've found, it was dumb luck at first. Prior runs didn't have any examples of wall jumps being done with so short a runway (3 blocks), but I was just playing around in Bowser's castle and accidentally performed a wall jump off the wrong (left hand) wall. This proved that a 3-block runway was possible, and fortunately I was recording at the time so I was able to sort out what had happened. As it turns out, this trick is just, just possible under perfect conditions - if the gap between walls was 1/16th of a pixel less, this trick and therefore this route wouldn't work at all!
What other new glitches or tricks allowed you to chop off more time from the previous records?
After finishing the first tank in World 8, we found that the boss behavior was bad, which was going to cost us all our world 1 savings. This sent us back to world 1 looking for a few more frames to let us manipulate that boss's behavior better. I found that in the World 1 fortress, P-speed could be maintained longer en route to the warp flute by jumping as late as possible at one point (jumping starts a counter for P-speed to expire when Mario has the leaf).
A bigger find that didn't save time but let us get more entertainment in was figuring out how to bounce off cannonballs fired just offscreen to the left. This, along with more careful planning, let us get 99 lives (unlike the published run, but like its predecessors) while completing World 8 even faster. This was also the hardest segment of the run as we weren’t even sure that it was possible.
Sounds like a lot of work. How long did it take you to perfect and finally record the run?
All told, I'd say I spent 60 or so hours on this run, with perhaps half of that spent on the first tank in World 8 (getting more lives and figuring out the low-speed floor glitch).
While the game is very linear, it does feature a few random elements like the cards at the end of a level. How did you circumvent them?
The randomness in Super Mario Bros. 3 revolves entirely on the frame (point in time) at which various things happen - there's an underlying random number generator that's updated 60 times a second. The 2 big "luck manipulation" tasks in Super Mario Bros. 3 are, like you said, getting the right card (mushroom, flower, star) at the end of each level and getting the hammer brothers on the World maps to move correctly. These tasks are much more significant for the "warpless" run, but for the cards we sacrifice 1 frame in 8-1 to avoid getting 3 stars (which would trigger a long animation). For the hammer brother, we need him to only move 1 space each time - moving farther takes longer! Finally, the boss for the first World 8 tank has set patterns - it is necessary to go into the tube at just the right frame to make him throw his boomerang as soon as possible. This is because only when all the hammer's sprites are offscreen will the treasure chest appear.
After the original run by Morimoto, the movie has already been improved a few times. Were you able to translate parts of their runs into yours, or did you start from scratch?
We absolutely took a lot from the previous runs. Pretty much every game like Super Mario Bros. 3 that's been improved multiple times develops a huge trove of tricks and techniques. Some turn out to be improvable, others don't, but being efficient and getting the best time possible requires a thorough knowledge of all that has gone before. It's important to note that this includes realtime speedruns, too - one of the biggest timesavers we discovered in the upcoming Super Mario Bros. 3 improvement was noted by a realtime speedrunner years ago in a forum at SpeedDemosArchive.com.
So there are still significant improvements possible?
For every run on TASVideos, you always have to assume the answer is a big yes! Time and again since I joined, I've seen runs thought to be perfect sit for a while before someone finds a new glitch or technique. There is a shortest possible time for every title, and Super Mario Bros. 3 may be getting close to that, but only time will tell.
I am actually working with Tompa on the improvement to this run right now; so far we have found about 1 second worth of improvements.
Any other runs you’re currently working on?
I'm also creating an Super Mario Bros. 3 "walkathon", which completes the game using the B-button only to use items (i.e. no running, fireballs, tail-flips, etc). No one has done this before for Super Mario Bros. 3, but the Super Mario Bros. one has been quite popular, and not being able to run completely changes the game's strategy, so I'm hoping it comes out well.
As opposed to normal Speedruns, TAS-Runners use emulators – in your case the specially-made FCEUX - to make and record their run. Could you explain to our readers how that works?
Our emulators allow us to play games either in slow-motion or (most often nowadays) in "frame advance" - where we can specify which buttons to press 60 times each second, and can manually advance through the game one frame at a time. This lets us slow things way down and observe game behavior very carefully. But another major element of TAS-ing is trial and error. Re-recording lets us mark points in the gameplay to go back to, so we can try different things or erase mistakes. Take a wall-jump in Super Mario Bros. 3 - jumping off a wall requires precise positioning. This is bad news for a realtime run, where a player might only succeed 10% of the time. But in a TAS we're able to try the walljump once, check Mario's x and y locations in memory and determine if we're too high, low, left or right, then make a tiny adjustment and try again (and again, and again...) Once we succeed, we continue on, and when the movie plays back all it shows is the successful walljump. Emulators keep track of the number of retries (or re-records) the movie authors use in making a movie. Most movies nowadays have several thousand re-records, some have a hundred thousand or more!
In addition some TAS, like the one of Mega Man 2, use Bots to find the optimal inputs. Did you use additional tools?
No, we did not use bots or any other technical help for this run. Super Mario Bros. 3 is all about geometry. We really just watched Mario's x and y locations in memory very carefully, to the 16th of a pixel, at all times. Beyond that, making the run was just a huge cycle of brainstorming ideas, testing them, implementing them, and then seeing if they could be even better.
Any advice for someone who wants to start making a TAS?
Watch a lot of TAS's and read the comments to get an idea of the techniques people are using. Then read the FAQs at TASVideos.org or ask questions on the "Newbie Corner" forum. Pick a game that you're already familiar with and that has an existing TAS (it is much easier to improve an existing TAS than to do one from scratch). Then, load the game up in the emulator and play around with it to thoroughly understand its mechanics and start looking for glitches, time-savers and the like. A big mistake beginners make is to start trying to make a run too early - without fully understanding the game or having an overall plan.
What TAS would you show someone who has never before watched one?
 NES Super Mario Bros. "warps" by klmz in 04:57.33 is the latest Super Mario Bros. run. It's a very widely known game and the run is short, fast-paced, and makes use of many obvious glitches while still preserving a lot of familiar gameplay.
Normal Speedrunners when asked about TAS mostly respond with a negative comment, declining TAS the right to call themselves speedruns? What is your take on that?
I've more heard of this than experienced it. I don't attach much importance to the name "speedrun", but whatever they're called it's important that the audience understands what they're watching, i.e. a real-time speed-run or tool-assisted.
But what’s more fun to watch in your opinion – a TAS or a "normal" Speedrun?
Neither - it totally depends on the particular movie. I like TAS's because I'm more involved in that community, have insight into how they're made, and they appeal to my technical nature. But realtime speedruns have the element of excitement that goes with being able to watch a real player either execute successfully or deal with mistakes or other unexpected occurrences.
Actually, two of my favorite runs to watch are recent speedruns at SDA of two games that I TAS'd - Legacy of the Wizard and Faxanadu. Both games are very complex and difficult, and it was very rewarding to see the speedrunners get great times and even use some of the tricks I found for my TAS.