6:15 it's absolutely savage to turn around and look your opponent in the eye as you are going into the final screen as if to say "buddy, you arent going to catch up"
Some people watching this video might not realize that pressing left and right on the same frame is not just physically impossible, it is against the rules. This is a result of standardizing play on different controllers. People playing an emulator with a keyboard must not have an advantage over people playing with the original hardware, because the original hardware is the standard against which everything else is checked. That also applies to people with turbo controllers, modified controllers, or whatever. Since you can't press left and right simultaneously on an original controller without modifying it (e.g. by cutting off bits of the pad or just pressing your thumb down so hard you literally break it), it's not allowed. Also, the TAS is not even timed the same way as a real run. The TAS timer starts at power on, which means it cannot wait on the title screen the way people do RTA or it will lose time. So Happylee's TAS is not a 4:54.03 but a 4:57.31, under a different set of rules. Maru's TAS (which is a one frame improvement over HappyLee's earlier version) optimizes RTA time following RTA rules. (It turns out, you don't actually have to wait on the title screen at all for optimal RTA time anyway, but in principle, the timings are different.)
Yeah there's other games that a turbo controller can break. Super Metroid bomb torizo skip is TAS only, but a human with a 30Hz turbo controller can do it. It requires inputs that are beyond the limits of the human species - both in frequency and precision. Key to the trick is pressing a key 13 times in 0.42 seconds (faster than a human can mash buttons) but not just that, doing it on specific frames. Hence the importance of rules like this. Can it be done on the OG hardware? If not, it's not allowed.
@@sirgog Tetris players can do it. Even in 2021 they've just developed a new technique (rolling) that results in as many as 20 d-pad taps per second, which is 0.05 seconds per tap.
@@B3Band I'm familiar with rolling. It's not fast enough, even ignoring the precision requirements. Rollers are button mashing in the low 20s per second, the requirement is 30.
It's good to see this video explanation. I'll add on my own thoughts: When I'm making a TAS, I don't see it as "telling the computer which buttons to push at which frame". I just think of doing super game play and make a great movie with precise movements. As for L+R, it's perfectly normal for TAS and totally approvable under TAS rules, because almost all TAS are made with emulators. I was using FCEU or FCEUX, which allows L+R, just like many other emulators. Usually when we talk about TAS, we usually talk about super play under TAS rules, and we don't need to think whether it can be done in real-time with real controllers, because it's pretty normal if it couldn't.
damn, L+R fast accels make so much difference also I appreciate the use of maru's TAS since it looks much more similar to regular runs. much easier to see differences
All I know is that it saves at least 1 frame in fast accelerations (getting up to speed from rest) and a lot more than that with the fast acceleration in the water section (since the top speed is so low, and there is no water fast accel at all RTA). It also saves time in some clips. You will notice that HappyLee never saves a frame rule over Maru, but he comes very close in 4-2.
But it seems that with Hercules fingers, you could perform it on a regular controller? I mean the individual input bits can be set simultaneously, right? So while not feasible, it's not impossible.
@@christophclear1438 The game checks repeatedly for input each frame during vblank if I recall correctly. You can't really check individual button presses because the controller only has one data line; instead, the inputs are pushed one by one onto the stack, so you basically need to read 8 times for each byte. Once a nonempty byte is read, I don't think you can change it. (This is different in SMB3, for instance: there need to be two sequential identical reads). So maybe if you could press left when the left bit is being read and then right when the right bit is being read a couple microseconds later, you could get both inputs. Not sure.
@@christophclear1438 Doens't matter what fingers you have, the controller is not designed to read simultaneous left and right inputs so it can't be done (unless you modify the controller, which is cheating). This is illustrated by the analogy with string instruments: when you press a string down on two points, only one of those pressure points has an effect.
This will become like the FNAF Max mode community, with harder and harder mods for each game and fangame and fangames of fangames, all of their hardest modes beat in record time... Like that mod of five nights at candy's 2 (already a fangame of FNAF) that had a mode "Ringing Hell" which jacks up the difficulty and has the player have a 1/4 chance of instantly dying no matter what every 15 seconds, it was beat (easily 💀) by some absolute sweat out there.
@@Gabriel64468 8-3 is interesting because the quickest RTA strategy to reach the flagpole reaches it with 243 remaining on the timer, which would actually lose time, since you would get 3 fireworks. So RTA runners and Maru reach the flagpole on timer 242 and do a flagpole glitch to save an extra frame rule. HappyLee's L+R TAS can reach the flagpole on timer 244 which achieves the same frame rule as 242+FPG. But the timing to get 244 is too tight to reach the flagpole on 244 AND do FPG, which is why L+R doesn't help save another frame rule in 8-3.
@@sdrawkcabmodnar the first room fast accel is impossible to do by humans, every other frame including the 2 frames in the turnaround have been done by humans (frame 63). But the current record only has Frame 65 for 8-4 IL
Can Happylee's tas be improved? i noticed that Maru is a bit in front at 2:20, or can someone explain the reason? Edit: it happens again in another point in the run, could this not matter because it does not save a framerule?
New Speedrun: Super Mario Bros. Extended Any% 8-4, Warpless 8-4, Any% D-4, Warpless D-4, Worlds A-D Any%, Worlds A-D Warpless, without the Minus World ending.
It's kind of weird that for all the incredible complexity, most of the time saved comes from a single frame of input with L+R instead of just R. (At the start of the water room.)
I have a suggestion that you may have already thought of. What about a visual indicator of the frame rule cycle? Perhaps it could be a number that counts up once per frame rule, or maybe a highlight around the timer each time the 21-frame cycle begins for one to three frames? Also, I feel like your algorithm is slowly getting better. In some of your earlier videos, I noticed a tiny bit of horizontal jitter on the Marios that seems to now either be minimized or completely mitigated. This shit looks *sooo* smooth, and I love it.
I like the idea! I'll have to think about how best to implement it. I eventually need to implement a reasonable 'timing tower' where I have each runner with their positions / splits. The frame rule system is integral to the game, so how best to show it is important. Thanks!
@@flibidydibidy perhaps a sort of “heartbeat” inspired animation? Not necessarily in the shape of a heart, perhaps adapt the coin shine cycle to be bright at the beginning of a frame rule and slowly fade out across the 21 frames?
I imagine you have tried experimenting with coins and I see that you are keeping them on-screen, even when both runs collect them (e. g. 1-1 underground at 01:35), but I was wondering if you could elaborate on why you decided to go this way, instead of showing them with some transparency as you do for enemies and include the fact that they can disappear if collected. I mean, I imagine that enemies are harder to deal with than coins, no?
Hi Cristian, insightful questions! Broadly speaking I place all sprites into the same 'nametable' (if you're not familiar with NES terminology, that's the regular gridded background). This 'nametable' is currently immutable, so beyond drawing it with a different palette (to get the coin color cycling), and a couple of hard coded ugly special cases (pipes, bullet bill glitch, the castle's 'flag' at the end of the level), I do not currently have an easy way of adjusting it on a per frame basis. Eventually I would like to handle the nametable in a better way - this will allow for transparent coins (which is a good suggestion!), and also for blocks breaking, bowser's bridge crumbling, etc. Once I add support for that in the drawing code, however, I would still have to add support for 'seeing' that in the classification code. Currently my image processing pipeline can not identify if the nametable has been changed.
@@flibidydibidy Thanks for the answer. Do I understand correctly if I say that the problem with coins is that they are part of the background, while enemies are not? Anyway, I think that if you could make some videos where you explain these details it would be very interesting.
Yes the coins in super mario bros are background - not sprites :). I do plan to explain this stuff! I have one more big mario video planned, then hopefully some more tech oriented videos. Thanks!
He was showing off that by using l+r to accelerate faster at the beginning and going as fast as possible you can end with a 244 on the timer. If he did the flagpole glitch it would take time and leave him with a 243 on the timer which would get him three fireworks and lose time. Good question!
If i had one thing i dislike about the videos is, like, if someone hits a box it stays normal. Well, more accurately if all of the people do it stays normal. Or coins don't disappear etc. I know its probably extra work but it could help visualize a slightly deeper quality. Watching stuff change if everyone does it. Could be cool :)
@@flibidydibidy here's a suggestion for that. Sprites/blocks should be tinted a color similar to the Mario's dominant color when interacted with, since sometimes it's hard to tell which of the sprites goes with which Mario
Happylees tas isn’t one frame from saving the framerule, he slowed down for entertainement but the most optimal way to do it would be ahead of marus tas
Human speedrunners are getting really close to the perfect Maru TAS run, 4.54.8 is the current record but in practice runs Nifski has demonstrated 4.54.4 is possible.
In terms of “can” yes he can. Anyone can. In terms of “is it allowed?” then no, he can but he’s not allowed by rules. The emulators have by default the “L+R” and “U+D” disabled. In order to change that you must then edit the default configuration of the emulator. Needless to say, it’s pretty easy to catch a player who has modified the controller or the emu configuration to do these things. Many people may think that this modification “would be hard to catch” but in reality it’s easy. Mario experiments an odd behavior that can be measured when those inputs are input, subtle for the casual untrained eye but pretty obvious for top level players.
So is what we see for turnaround room what the racers are gonna see or is that just for broadcasting because a big part of smb1 racing any% is tying to 8-4 then having a faster 8-4 and unfamiliar pixels to stop on can be a bit wired to get used to
So they get all the same framerules and all the actual difference is in 8-4, right? Can you please do commentary on 8-4 about what and where does happylee do to get ahead of human theory?
8-4 is special in that it's not affected by framerules. It counts every frame until you touch the axe, so loss or gain in time is done by frames instead. Happy Lee's TAS uses L+R as an input which makes Mario goes faster in frames. This is because Mario moves faster facing the opposite of your input (i.e. moving right but facing left).
Is there an AI that is being trained in SMB speedrunning? maybe such AI could come up with new strats/discoveries that could make speedruns even shorter. Like OpenAI in dota2, the bots have started doing things that didn't make sense in the beginning but later humans understood their "reasoning" and implemented bot behaviour in their own game play :D Maybe something similar could be achieved in SMB too (which is way less complex than dota2)
pretty late but something like a top 3/top 5 comparison for minus world ending would be interesting, especially since the wr is the only submitted run with -2 ffpg
isn't it possible to press L+R on a controller if you simply break the plastic dpad in half? what if someone did that and got a new world record? he really did it on a real nes hardware on a real controller, just the dpad plastic is broken.
because HappyLee's TAS needs to avoid getting 243 which make 3 fireworks goes off so the TAS touch the flag at 244 on the timer and 243 with FPG isnt possible by few frame becuase FPG make Mario touch the flag pole 6 frames later hope my dumb english will help
WHAT IF WE HAD MULTIPLAYERS LIKE THIS FOR ALL THE OLD SCHOOLS SUPER MARIO BROS SERIES SUPER MARIO LAND SERIES & SUPER MARIO WORLD SERIES BACK IN THE 80’S & 90’S
SMB1 Speedruns are not like improvisations at all. They attempt to play a piece of music and sometimes make mistakes which they cover up, not unlike concert musicians. Improvisations are creating music on the fly or with a general framework in mind. There's no 'on the fly' aspect to SMB1 Speedruns.
For very high level SMB you are absolutely correct, but for the majority of speedrunning in general (even SMB above about 4:58) I think the analogy is close enough. ;) Thanks for the feedback!
On 8-3 getting a 243 on the time gives off 3 fireworks which looses time so Maru has to slow down and 244 is impossible without left and right however doing th fpg requires Mario to touch the flag pole slightly later resulting in 243 so that’s why happylee didn’t do the fpg on 8-3
@@starting7725 A 243 FPG is possible in Super Mario Bros. Extended, and it doesn't create fireworks. Watch the TAS here! ua-cam.com/video/MJgLHFj6r1Q/v-deo.html
When a spiney is in front of the score my code can't identify it, and currently just leaves it blank. Realistically I should just use the last valid score now that I think about it.
I'm guessing HappyLee doesn't go for the flagpole glitch at 8-3 bc he doesn't make the framerule anyway? Otherwise why doesn't he do the flagpole glitch?
I like that they're grey
cool guy
The most clutch gamer alive will do a 4.54 in 9 attempts
They're metal (robot)
Yes they are robots
Hello big boy
6:15 it's absolutely savage to turn around and look your opponent in the eye as you are going into the final screen as if to say "buddy, you arent going to catch up"
Wow, I can’t believe this had 337 likes but NO COMMENTS.
4:50 I like HappyLee showing off while waiting for the Bullet Bill
Immagine doing this during a real run
It’d be easier to get 8-4 il wr
That was insane! I have heard about HappyLee TASas and I have seen some pieces here and there, but that I never saw!
oh hey there power twower
That was insaine i had to watch it back to believe what i saw again
Some people watching this video might not realize that pressing left and right on the same frame is not just physically impossible, it is against the rules. This is a result of standardizing play on different controllers. People playing an emulator with a keyboard must not have an advantage over people playing with the original hardware, because the original hardware is the standard against which everything else is checked. That also applies to people with turbo controllers, modified controllers, or whatever. Since you can't press left and right simultaneously on an original controller without modifying it (e.g. by cutting off bits of the pad or just pressing your thumb down so hard you literally break it), it's not allowed.
Also, the TAS is not even timed the same way as a real run. The TAS timer starts at power on, which means it cannot wait on the title screen the way people do RTA or it will lose time. So Happylee's TAS is not a 4:54.03 but a 4:57.31, under a different set of rules. Maru's TAS (which is a one frame improvement over HappyLee's earlier version) optimizes RTA time following RTA rules. (It turns out, you don't actually have to wait on the title screen at all for optimal RTA time anyway, but in principle, the timings are different.)
Yeah there's other games that a turbo controller can break. Super Metroid bomb torizo skip is TAS only, but a human with a 30Hz turbo controller can do it.
It requires inputs that are beyond the limits of the human species - both in frequency and precision. Key to the trick is pressing a key 13 times in 0.42 seconds (faster than a human can mash buttons) but not just that, doing it on specific frames.
Hence the importance of rules like this. Can it be done on the OG hardware? If not, it's not allowed.
@@sirgog Technically, they found a way to do the BT skip with only 12 arm pumps in NTSC. That's still impossible, though.
@@EebstertheGreat Yeah still far beyond human physical capabilites. 7 times within 0.42 sec would probably get done on rare occasions.
@@sirgog Tetris players can do it. Even in 2021 they've just developed a new technique (rolling) that results in as many as 20 d-pad taps per second, which is 0.05 seconds per tap.
@@B3Band I'm familiar with rolling. It's not fast enough, even ignoring the precision requirements. Rollers are button mashing in the low 20s per second, the requirement is 30.
Very cool comparison. You can really see HappyLee's TAS inching ahead at every fast accel in 8-4.
So essentially they both reach 8-4 at the same framerule, meaning those L+R fast accels on previous levels won't save any time. Pretty interesting.
a couple of them (1-1, 4-2) are literally 1 frame off saving a framerule lol
@@whitegx-k5iyooo ur pfp, you like rubiks cubes too?
It feels like Maru is playing seriously while Happylee just wants to show off
Well. HappyLee's TAS was perfect for years. So in my opinion he decided to edit it and add some show offs just for fun.
It's good to see this video explanation. I'll add on my own thoughts:
When I'm making a TAS, I don't see it as "telling the computer which buttons to push at which frame". I just think of doing super game play and make a great movie with precise movements.
As for L+R, it's perfectly normal for TAS and totally approvable under TAS rules, because almost all TAS are made with emulators. I was using FCEU or FCEUX, which allows L+R, just like many other emulators.
Usually when we talk about TAS, we usually talk about super play under TAS rules, and we don't need to think whether it can be done in real-time with real controllers, because it's pretty normal if it couldn't.
When you made your tas in 2011 did you ever think humans would get close to it?
@@GardenChess No. Back in 2011, no one can do things like flagpole glitch in real-time.
damn, L+R fast accels make so much difference
also I appreciate the use of maru's TAS since it looks much more similar to regular runs. much easier to see differences
this is what true mario coop looks like in a perfect world
They ain't cooperating much, they are just both doing their thing?
might've been worth explaining what effect L+R has
All I know is that it saves at least 1 frame in fast accelerations (getting up to speed from rest) and a lot more than that with the fast acceleration in the water section (since the top speed is so low, and there is no water fast accel at all RTA). It also saves time in some clips. You will notice that HappyLee never saves a frame rule over Maru, but he comes very close in 4-2.
Yeah. Woops. It makes you accelerate faster ;)
But it seems that with Hercules fingers, you could perform it on a regular controller? I mean the individual input bits can be set simultaneously, right? So while not feasible, it's not impossible.
@@christophclear1438 The game checks repeatedly for input each frame during vblank if I recall correctly. You can't really check individual button presses because the controller only has one data line; instead, the inputs are pushed one by one onto the stack, so you basically need to read 8 times for each byte. Once a nonempty byte is read, I don't think you can change it. (This is different in SMB3, for instance: there need to be two sequential identical reads). So maybe if you could press left when the left bit is being read and then right when the right bit is being read a couple microseconds later, you could get both inputs. Not sure.
@@christophclear1438 Doens't matter what fingers you have, the controller is not designed to read simultaneous left and right inputs so it can't be done (unless you modify the controller, which is cheating). This is illustrated by the analogy with string instruments: when you press a string down on two points, only one of those pressure points has an effect.
Please make more of these videos. They're so fun to watch.
You can tell how in synch these videos are by the music canceling itself out on mono speakers. Cool comparison!
oh god when humans finally save the last frame of 8-4 will people start doing "perfect speedrun + score" runs?
No lmao
No they’d just do other categories like warpless and all starts lol
This will become like the FNAF Max mode community, with harder and harder mods for each game and fangame and fangames of fangames, all of their hardest modes beat in record time...
Like that mod of five nights at candy's 2 (already a fangame of FNAF) that had a mode "Ringing Hell" which jacks up the difficulty and has the player have a 1/4 chance of instantly dying no matter what every 15 seconds, it was beat (easily 💀) by some absolute sweat out there.
well maru had more points for most of the game so who really won
Lee.
Lee.
Lee.
Good point lol
The fact that no one cares for point on mario kinda answers it lol
2:08
I like how they jump at opposite times
I hope this guy gets more attention soon. He put in the work and deserves it.
When the jumping noise in the water section hit I laughed so hard I fell out of my chair
When half of the music is missing but you still wanna play at the optimal level:
Your videos are incredible! Thank you for making them 🙏👏
Framerule making a huge difference is fun to see. All of the extra accelleration basically didn't matter until the very end.
Yeah even on 8-3.
It looked like that would surely lead to a meaningful timedifference, but no
@@Gabriel64468 8-3 is interesting because the quickest RTA strategy to reach the flagpole reaches it with 243 remaining on the timer, which would actually lose time, since you would get 3 fireworks. So RTA runners and Maru reach the flagpole on timer 242 and do a flagpole glitch to save an extra frame rule. HappyLee's L+R TAS can reach the flagpole on timer 244 which achieves the same frame rule as 242+FPG. But the timing to get 244 is too tight to reach the flagpole on 244 AND do FPG, which is why L+R doesn't help save another frame rule in 8-3.
“Maru is restricted to human viable strats”
Also Maru: *does humanly impossible inputs in first room of 8-4*
where
@@sdrawkcabmodnar the first room fast accel is impossible to do by humans, every other frame including the 2 frames in the turnaround have been done by humans (frame 63). But the current record only has Frame 65 for 8-4 IL
@@ilect1690 fast accel is not impossible lol, niftski even did it on his IL we
@@michael.pintabona5949 it isn't but the setup for saving an extra frame is
@@ilect1690 it is believed by many top runners that it will be achieved some day.
I really like this version of the music!
So basically, L+R can add wiggle room, but you don't really save time until 8-4.
Happy Lee sure makes his tases entertaining. I laughed a few times during warpless.
The ending of 8-2 how happylee almost dies and goes back up the wall is nuts
"almost dies" lol
It's on purpose
Nice! Thanks for putting this together
Your metaphors are OP! I love it
I've been wanting this so bad!! Thank you!
It's funny how the only place the physically impossible TAS really saves time is the 8-4 water area.
6:10 rip headphone users
0:07 "just before I answer that"....
PLEASE SAY PARALLEL UNIVERSES!!!
0:57 yes you can! LeKukie has shown it off multiple times
irl
Can Happylee's tas be improved? i noticed that Maru is a bit in front at 2:20, or can someone explain the reason?
Edit: it happens again in another point in the run, could this not matter because it does not save a framerule?
Happy lee TAS IIRC was created about 10 years ago
@@willie333b so long ago yet is still the best
Nah it’s too slow to save the framerule
Happylee's TAS can sort of be improved, but those improvements never save a framerule.
Even if it’s a little slower it still can’t be enough to save the framerule
Congrats for 5000
New speedrun:
Get the most points possible while still getting a sub 5 minute run
Sheeeeesh
New Speedrun:
Super Mario Bros. Extended Any% 8-4, Warpless 8-4, Any% D-4, Warpless D-4, Worlds A-D Any%, Worlds A-D Warpless, without the Minus World ending.
With separate categories for Mario and Luigi because of the different physics.
that water section was really satisfying to watch
It's kind of weird that for all the incredible complexity, most of the time saved comes from a single frame of input with L+R instead of just R. (At the start of the water room.)
That is an amazing analogy
I've never looked at two TASes and thought 'let's make them fight'
I have a suggestion that you may have already thought of. What about a visual indicator of the frame rule cycle? Perhaps it could be a number that counts up once per frame rule, or maybe a highlight around the timer each time the 21-frame cycle begins for one to three frames?
Also, I feel like your algorithm is slowly getting better. In some of your earlier videos, I noticed a tiny bit of horizontal jitter on the Marios that seems to now either be minimized or completely mitigated. This shit looks *sooo* smooth, and I love it.
I like the idea! I'll have to think about how best to implement it. I eventually need to implement a reasonable 'timing tower' where I have each runner with their positions / splits. The frame rule system is integral to the game, so how best to show it is important.
Thanks!
@@flibidydibidy perhaps a sort of “heartbeat” inspired animation? Not necessarily in the shape of a heart, perhaps adapt the coin shine cycle to be bright at the beginning of a frame rule and slowly fade out across the 21 frames?
4:50 they were wall jumping before it was implemented
2:08 すこ
like your fpg and bbg video could you perhaps make videos explaining other tricks? i.e blazit, pl8-1, tas 8-2 and other stuff like that
I'd love to make 'deep dives' on some of the levels where I get into that stuff. Thanks for the suggestion!
5:36 where the video starts
I imagine you have tried experimenting with coins and I see that you are keeping them on-screen, even when both runs collect them (e. g. 1-1 underground at 01:35), but I was wondering if you could elaborate on why you decided to go this way, instead of showing them with some transparency as you do for enemies and include the fact that they can disappear if collected. I mean, I imagine that enemies are harder to deal with than coins, no?
Hi Cristian, insightful questions! Broadly speaking I place all sprites into the same 'nametable' (if you're not familiar with NES terminology, that's the regular gridded background). This 'nametable' is currently immutable, so beyond drawing it with a different palette (to get the coin color cycling), and a couple of hard coded ugly special cases (pipes, bullet bill glitch, the castle's 'flag' at the end of the level), I do not currently have an easy way of adjusting it on a per frame basis.
Eventually I would like to handle the nametable in a better way - this will allow for transparent coins (which is a good suggestion!), and also for blocks breaking, bowser's bridge crumbling, etc. Once I add support for that in the drawing code, however, I would still have to add support for 'seeing' that in the classification code. Currently my image processing pipeline can not identify if the nametable has been changed.
@@flibidydibidy Thanks for the answer. Do I understand correctly if I say that the problem with coins is that they are part of the background, while enemies are not? Anyway, I think that if you could make some videos where you explain these details it would be very interesting.
Yes the coins in super mario bros are background - not sprites :). I do plan to explain this stuff! I have one more big mario video planned, then hopefully some more tech oriented videos. Thanks!
It would be cool to see an input display in the corner of each screen to see the differences.
Why didn’t happylee do fpg on 8-3? Wouldn’t that save a frame rule? Or make sub 454 possible?
He was showing off that by using l+r to accelerate faster at the beginning and going as fast as possible you can end with a 244 on the timer. If he did the flagpole glitch it would take time and leave him with a 243 on the timer which would get him three fireworks and lose time. Good question!
@@flibidydibidy because you touch the flagpole 6 frames later with fpg
Also, widescreen SMB !!! Wow !
If i had one thing i dislike about the videos is, like, if someone hits a box it stays normal. Well, more accurately if all of the people do it stays normal. Or coins don't disappear etc. I know its probably extra work but it could help visualize a slightly deeper quality. Watching stuff change if everyone does it. Could be cool :)
Yeah! I agree it bothers me too. I'm hoping to address that in some upcoming software refinements. Thanks!
@@flibidydibidy here's a suggestion for that. Sprites/blocks should be tinted a color similar to the Mario's dominant color when interacted with, since sometimes it's hard to tell which of the sprites goes with which Mario
I like the swimming part where they both mash A 8 million times
3:24 wait, if maru's tas is ahead, and happylee's tas is just 1 frame off saving another FR, does that mean that...?
Happylees tas isn’t one frame from saving the framerule, he slowed down for entertainement but the most optimal way to do it would be ahead of marus tas
Пацанский TAS радует глаз.
Human speedrunners are getting really close to the perfect Maru TAS run, 4.54.8 is the current record but in practice runs Nifski has demonstrated 4.54.4 is possible.
The actual 2-player mode in SMB
3:42 Holy hell those pivots over the pits!
Woulda been better if the blue tas was green but otherwise great video!
Random fact:
Happylee had more points in 1-2, 8-2, 8-3, and 8-4 while Maru had more in 1-1, 4-1, 4-2 and 8-1
wait HOW DID YOU NOTICE THAT lol
@João Felipe Because i know both languages?
Will we ever see 4:53.xxx, even in a TAS?
Not unless new strats are found
We will see a 4:53, but in Super Mario Bros. Extended. We are currently seeing the 4:53 in a TAS here!
ua-cam.com/video/MJgLHFj6r1Q/v-deo.html
It's really interesting that the time difference is only because of the L+R fast acceleration.
Gotta love the music
No intro version pls
ua-cam.com/video/8vKw6Z6AGE0/v-deo.html
I feel like a category allowing L+R might be interesting.
Since Niftski plays on keyboard, can't he press L+R simultaneously?
In the emulator, it is disabled for RTA runs
He’s not allowed
yeah but that would be cheating
In terms of “can” yes he can. Anyone can. In terms of “is it allowed?” then no, he can but he’s not allowed by rules.
The emulators have by default the “L+R” and “U+D” disabled. In order to change that you must then edit the default configuration of the emulator.
Needless to say, it’s pretty easy to catch a player who has modified the controller or the emu configuration to do these things. Many people may think that this modification “would be hard to catch” but in reality it’s easy. Mario experiments an odd behavior that can be measured when those inputs are input, subtle for the casual untrained eye but pretty obvious for top level players.
He can, but it’s not allowed
TAS Battleee o.0
So is what we see for turnaround room what the racers are gonna see or is that just for broadcasting because a big part of smb1 racing any% is tying to 8-4 then having a faster 8-4 and unfamiliar pixels to stop on can be a bit wired to get used to
Racers would be playing on original hardware, on their own screens. The combined view is for visualization purposes only.
why didnt lee's tas do the flagpole glitch at 5:26?
Hello! If he did he would have got fireworks and been slower.
3:26 happylee was slightly slower 👀 same framerule tho
Can you try doing a warpless comparison next?
Happys warpless is fucking nutso
I never understood how fast L+R fast accel was until now
It saves 3 frames on flat ground but L+R is VERY useful for both slowing down and speeding up.
Ok
@FlibidyDibidy What other content do you want to make with this software?
Other games, more marios. I'm building out the system to work 'live' but that's a big ask honestly. It will take time! Thanks,
4:50 happylee goes godmode
Why did the red stop at 4:50?
I explain the bullet bill glitch a bit here: ua-cam.com/video/XtcIXYQ9gtU/v-deo.html
Thanks!
So they get all the same framerules and all the actual difference is in 8-4, right? Can you please do commentary on 8-4 about what and where does happylee do to get ahead of human theory?
8-4 is special in that it's not affected by framerules. It counts every frame until you touch the axe, so loss or gain in time is done by frames instead. Happy Lee's TAS uses L+R as an input which makes Mario goes faster in frames. This is because Mario moves faster facing the opposite of your input (i.e. moving right but facing left).
my 50-year-old uncles playing Mario after saying they haven't played in years
What does holding L+R even do?
In some spots the tas uses l+r to do a fast acceleration that saves a few frames
i want to see a human do that wall jump out of the pit in 8-2...
lol that was epic!
Is there an AI that is being trained in SMB speedrunning? maybe such AI could come up with new strats/discoveries that could make speedruns even shorter. Like OpenAI in dota2, the bots have started doing things that didn't make sense in the beginning but later humans understood their "reasoning" and implemented bot behaviour in their own game play :D Maybe something similar could be achieved in SMB too (which is way less complex than dota2)
If we allow for people to modify the controller, THEN the perfect run can use the L+R thing
big jump ahead at that last fast accel..
Why didn't the blue tas go for the pole glitch in 8-3???
It would’ve lost a frame rule
Because he would get fireworks, and grabbing the flag is the same speed as slowing down to avoid fireworks but doing fpg
Tbh, I like TAS better than real time speedruns
happylee: *being a tryhard trying to finish seconds off maru*
Maru: *still finishes only 200ms off happylee*
Why are you grae?
Who says I’m grae?
You are grae.
ur videos are great
Thank you!
pretty late but something like a top 3/top 5 comparison for minus world ending would be interesting, especially since the wr is the only submitted run with -2 ffpg
2:09 this is cool
isn't it possible to press L+R on a controller if you simply break the plastic dpad in half?
what if someone did that and got a new world record?
he really did it on a real nes hardware on a real controller, just the dpad plastic is broken.
Can anybody explain to me why HappyLee did go for the top of the Flagpole at 5:25 ? Isn't that the slowest way to finish a stage?
because HappyLee's TAS needs to avoid getting 243 which make 3 fireworks goes off so the TAS touch the flag at 244 on the timer and 243 with FPG isnt possible by few frame becuase FPG make Mario touch the flag pole 6 frames later
hope my dumb english will help
Bruh HaapyLee pulled off some Mario 64 wall jumps at the end of 8-2
yeah maru is so much better it literally waited for heppylee to climb up that's max level dominance
The dominance is on Lee's part. Man was literally showing off his skills.
@@Plasmariel still
WHAT IF WE HAD MULTIPLAYERS LIKE THIS FOR ALL THE OLD SCHOOLS SUPER MARIO BROS SERIES SUPER MARIO LAND SERIES & SUPER MARIO WORLD SERIES BACK IN THE 80’S & 90’S
but wouldn't using a keyboard and emu makes l+r possible? surely an advantage? p.s love the vids and the machine learning stuff, cool
um l+r not allowed
It does but you’re not allowed to
SMB1 Speedruns are not like improvisations at all. They attempt to play a piece of music and sometimes make mistakes which they cover up, not unlike concert musicians. Improvisations are creating music on the fly or with a general framework in mind. There's no 'on the fly' aspect to SMB1 Speedruns.
For very high level SMB you are absolutely correct, but for the majority of speedrunning in general (even SMB above about 4:58) I think the analogy is close enough. ;) Thanks for the feedback!
Why HappyLee on 8-3 jump on the top of flag. He loses time and now same with Maru. It is possible to do 4:53 TAS run.
On 8-3 getting a 243 on the time gives off 3 fireworks which looses time so Maru has to slow down and 244 is impossible without left and right however doing th fpg requires Mario to touch the flag pole slightly later resulting in 243 so that’s why happylee didn’t do the fpg on 8-3
@@starting7725 A 243 FPG is possible in Super Mario Bros. Extended, and it doesn't create fireworks. Watch the TAS here!
ua-cam.com/video/MJgLHFj6r1Q/v-deo.html
I like the part when they ran
Why does happy Lee’s score stutter? I noticed this in 4-1. Idk if it happens anywhere else
When a spiney is in front of the score my code can't identify it, and currently just leaves it blank. Realistically I should just use the last valid score now that I think about it.
Tool You used to compare?
They're still so close, it's insane.
Race For Peach, Who Will Make It First? Blue Mario Makes It.
I'm guessing HappyLee doesn't go for the flagpole glitch at 8-3 bc he doesn't make the framerule anyway? Otherwise why doesn't he do the flagpole glitch?