Uploaded August 2020 | Updated September 2026, 2 weeks ago
This is tool-assisted.
The previous OoB vid was 2 or 3 seconds slower than an estimated optimal autoscroller. This attempt is more than 10 seconds faster, and isn't even great in terms of optimisation - a couple of obvious improvements are in there.
We think there could be a way to escape the moving saucer's gravity facing backwards which would mean a lot more distance covered right at the start by backwards bouncing off the cannonballs.
This is tool-assisted.
The previous OoB vid was 2 or 3 seconds slower than an estimated optimal autoscroller. This attempt is more than 10 seconds faster, and isn't even great in terms of optimisation - a couple of obvious improvements are in there.
We think there could be a way to escape the moving saucer's gravity facing backwards which would mean a lot more distance covered right at the start by backwards bouncing off the cannonballs.
![[TAS] SMG2: Space Storm Green Stars](https://i.ytimg.com/vi/qOEQi59gTmY/mqdefault.jpg "[TAS] SMG2: Space Storm Green Stars
Green Stars TAS WIP
Star order matters due to bubble cycles for GS3, where cutscene cycles are better by 9 frames, but also pull star cycles means GS2 would save 3 frames. So overall doing GS3 first is faster by 6 frames. Quick maths.
Green Star 1 - 5.88 IGT - 1:33:04.05 real-time split (1:03.57 segment)
The cycle of the ring of pull stars means grabbing a pull star immediately isnt necessarily faster.
After much testing, it was found to be marginally faster to do a land cancel before grabbing the pull star planet, as this leads to a slightly higher and faster first pull where the speed adds up later on, while also catching the cycle at a perfect time.
The rotation of Luigi in the bubble matters for star grab, and can make up to 4 frames of difference. In this case, it saved 3 frames to twist him around using the IR cursor.
Green Star 3 - 16.86 IGT - 1:34:16.10 real-time split (1:12.05 segment)
This is an improvement of 1.6s over my 2020 TAS: https://youtu.be/-KnxWlYubGA
The bubble cycle is either global or the shooter is simply close enough to stay active (you can see the bubbles firing from the first planet). This means hyper-optimisation of the whole level isnt necessary provided you get the fastest cycle.
Despite this, I did optimise the first section as best I could, including an early release from the bubble to save a couple of frames on the launch star.
The main improvement to get an earlier cycle was going around the wall and setting up a boosted triple jump.
Jumping over a void (or moving floor) from a moving floor will make you accelerate in the air even when falling forward. This - plus landing in neutral to conserve more speed - led to the double jump being as fast as a long jump, and the triple jump then almost reaching the speed cap of 32, all despite going the opposite way of the moving floor.
Green Star 2 - 14.80 IGT - 1:35:13.83 real-time split (57.73s segment)
This is an improvement of 19 frames over my 2020 TAS, despite no-cutscenes losing 3 frames on the pull stars: https://youtu.be/tT7OmwDp0vg
Interestingly for the second planet, I managed to get a 6f faster route by reloading the wrong savestate during playback. A setup for a different route idea ended up making the main route idea faster, thanks to landing on the edge of the wall in such a way to be able to long jump without a stuttering run, or just immediately falling off.
Backflip saved 1 frame.
#supermariogalaxy2 #toolassistedspeedrun")
![[TAS] SMG2: Fiery Flotilla GS2 in 17.11 (RTA route)](https://i.ytimg.com/vi/qmK9BPKPkCs/mqdefault.jpg "[TAS] SMG2: Fiery Flotilla GS2 in 17.11 (RTA route)
This route is 0.7s faster in real time than the Thwomp jump route, as the Thwomp takes too long to rise when skipping cutscenes fast.
The spin I do at the end is done by shaking for 2 frames and pressing Z on the second frame. Normally this would result in a normal ground pound with spin attack properties still on it (used in Major Burrows fights, for instance), but when doing it too close to the ground to be able to ground pound, it instead results in a very short spin.
This was done to get a little bit further along with some of the remaining long jump speed without losing too much of it with a regular spin, which meant I was far enough along for the second lava boost to go directly up into the star, rather than slowly arching towards it. This saved 8 frames.
Thwomp jump is faster with full cutscene cycles, so it is still in the run, and doing GS1 first is 0.5s faster than doing GS2 first, because the Thwomp jump can go much further left - much closer to GS1 - than this ? block triple jump wallkick.
This doesnt seem like it makes sense but Im tired.
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
![[TAS] Super Mario Galaxy: Buoy Base Secret in 40.53s](https://i.ytimg.com/vi/qnbGV_tTlUs/mqdefault.jpg "[TAS] Super Mario Galaxy: Buoy Base Secret in 40.53s
Mario finds the secret of Buoy Base in 40.53 seconds [2432 frames].
There are two main parts to this: the torpedo and the Bullet Bill.
The torpedo spawns as soon as youre vertically in line with it. A good visual cue is reaching the ring with starbits on.
Optimising this was interesting, as going straight down would mean the torpedo would come out too early and home in when youre on the wrong side, going away from the pipe.
So I made sure to get some distance before activating the torpedo.
There may be a frame to save if you can turn the camera to where it ends up when watching the pipe whirlpool spawn. Cameras have micro-optimisations like that in other places.
I had to swim into the bottom of the cage so the torpedo would be on the perfect path to hit as soon as possible. Not doing this wouldve lost a frame.
Bullet Bills still turn when frozen by P2, so the key was optimising the right path to make it turn quicker and letting go when its angle is at least somewhat facing backwards.
The backflip was late enough that it didnt change the Bullet Bills path, at least, not enough to lose time.
#SuperMarioGalaxy #ToolAssistedSpeedrun")
![[TAS] Super Mario Galaxy 2: Tall Trunk Green Star 1 in 8.83s](https://i.ytimg.com/vi/qshjDR8ch4M/mqdefault.jpg "[TAS] Super Mario Galaxy 2: Tall Trunk Green Star 1 in 8.83s
Luigi climbs a tree without a run-up this time in 8.83 seconds [IGT, 530 frames].
-1.52s (91f) improvement over previous record. Tall Trunk GS1 joins the sub 10 club.
Diamentos found that the technique of holding neutral (or pressing C) before the double jump will add more height and momentum to the triple jump, making this route much faster and it doesnt need a run-up from behind the spawn stump: https://clips.twitch.tv/BrainyGentleTildeDxCat
This is a technique that was found over the development of the Battlerock 1 OoB trick.
This wasnt hyper-optimised with a couple of slightly later landings than necessary, and I think there is some more potential in the extended triple jump that would remove the need for a walljump, and possibly even without a spin (the spin in the run was used to land earlier and further right than the walljump would have).
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
![[TAS] Super Mario Galaxy 2: Supermassive Green Star 3 in 19.58s](https://i.ytimg.com/vi/rSjkTOOWvjc/mqdefault.jpg "[TAS] Super Mario Galaxy 2: Supermassive Green Star 3 in 19.58s
Luigi and P2 use checkpoints and Koopa Troopas to collect a Green Star in 19.58 seconds [IGT, 1175 frames].
I delayed the first long jump as it got the following single jump a bit closer to the ledge, saving a few frames.
For the wallkick on the ? block, I used neutral wallkick strain to be closer to the block when spinning, landing earlier.
Having P2 grab the checkpoint and then dying is faster than the pipe strat done in RTA runs by ~1.5 seconds (in real time; IGT is more like 3.5 seconds).
But there is a different strat using angled wallkicks that is ~0.3 seconds faster in real time than the death abuse strat (seen here: https://youtu.be/KwHCkBmDdJU), so death abuse only has use for in-game time - the timer stops when the life counter appears and starts again when its gone, so in-game time is 2 seconds shorter than real time.
Freezing the Koopa Troopa at the end to bounce off it saved 0.6s over a triple jump.
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
![[TAS] SMG2: Fluffy Bluff Green Star 2 without Cloud Flower](https://i.ytimg.com/vi/rT8SBF4Pwbw/mqdefault.jpg "[TAS] SMG2: Fluffy Bluff Green Star 2 without Cloud Flower
aka Luigi makes his way around the large cliffs for 4 minutes.
Could this be done with a crystal float? Maybe, I didnt spend long testing that idea. Could it be done with a megasideflip? I dont know if they are possible here. Could it be done with a backwards jump off the cloud? Maybe, I didnt get very close with it though.
Sometimes its good to have a silly outlandish method of completing a challenge.
While slope climbing - as it works in SMG1 - is patched, it is still possible to jump off a slope and be able to spin, if that jump is on the last frame before sliding off the slope.
Sometimes a slope has a bump before the edge that Luigi can slide off of, so it might look like a slope jump is not coming off the edge but it is coming off the end of a slide.
I could have made this much faster by not messing around with different ideas, but whatever lol.
#SuperMarioGalaxy2 #PowerUpChallenge")
![[TAS] SMG2: Rolling Masterpiece Green Stars](https://i.ytimg.com/vi/rZX99eGgevk/mqdefault.jpg "[TAS] SMG2: Rolling Masterpiece Green Stars
Green Stars TAS WIP
I included the World movement because it is mind-blowing movement with the small IR click saving an astonishing 2 frames and breezing past Freezy Flake.
Star order doesnt matter for this galaxy.
Green Star 2 - 58.83 IGT - 58:07.10 real-time split (1:55.59 segment)
1.8s improved from 2020 despite starting from cutscenes: https://youtu.be/mTI_hX9rb5o
Backflip was 25 frames faster than long jump spin - a spinless long jump cant land directly onto the ball without a little more height as well as slowing down significantly, otherwise it just gets the forced jumping mount. Spinning from backflip was 21 frames slower.
The triple jump chain into the launcher was 2 frames faster than rolling and doing 1 jump.
Optimising launcher entry involves more than just aiming for the middle as too much speed will cause some slow swirls, but also having too little swirling causes the ball to vibrate in the middle, taking longer to lock in and be launched.
Because of this, its faster to aim slightly off-centre and slow down in the air to avoid the vibrating while still minimising the time spent swirling.
In the 2020 TAS, I had thought that jumping through the red brushes was a little bit slower, but what I missed was getting the trajectory just right to both minimise the speed lost from hitting the brush and to hit the side of the yellow brushes when landing.
It turns out that this hit avoids the usual landing slowdown, making the route much faster, about 1s faster in fact.
2P can freeze the starball. This is used to quickly turn around from the keys cutscene.
Trajectory of the next launcher is too fast to jump straight out and land safely, but some speed is retained with the movement done.
The jump straight onto the top disc requires all that run-up and a jump on the last possible frame.
Like the yellow brushes, hitting the arrow sign when landing avoided the landing slowdown to have more speed for ruler bouncing. In fact, it kept too much speed so I weaved in the air to slow down enough to catch the edge of the ruler to get a faster fall towards the star.
Green Star 1 - 35.08 IGT - 59:25.26 real-time split (1:18.16 segment)
1.1s improved from 2020.
Skipping cutscenes means a different starting position. A direct backflip wasnt possible, so long jump spin was fastest by 6 frames over a backflip spin.
For the star grab, theres no jump height difference that depends on horizontal speed, so jumping directly underneath would have worked just as well as the big run-up.
Happy new year!
#supermariogalaxy2 #toolassistedspeedrun")
![[TAS] SMG1: Beach Bowl Comet in 39.60s](https://i.ytimg.com/vi/rf4fRG5cBNg/mqdefault.jpg "[TAS] SMG1: Beach Bowl Comet in 39.60s
Mario utilises the Fast Foes of the Stone Cyclone to collect the star in 39.60 seconds [2376 frames].
This is highly-optimised, as it is optimised in the parts that matter most; the more unoptimised parts dont necessarily matter because they have to wait for a cycle anyways.
That cycle is the Tox Box cycle, and a fling off of it can catch the gravity of the upper section and swing around, bumping another Tox Box to get the right line straight to the crystal.
Then the crystal can be broken in one spin with the right position.")
![[TAS] Super Mario Galaxy 2: Upside Dizzy Green Star 1 in 12.96s](https://i.ytimg.com/vi/rw64kFfrQNg/mqdefault.jpg "[TAS] Super Mario Galaxy 2: Upside Dizzy Green Star 1 in 12.96s
Luigi flips around gravity planes to collect a Green Star in 12.96 seconds [IGT, 778 frames].
This is the IL strat that has been known for years, optimised to beat the 13 second mark.
The strat is very janky. For some reason - possibly an effect of the bonk - Luigi twirls after changing gravity, and this makes his wallkick angle go whatever way it wants, including backwards.
Interestingly, delaying the first long jump and getting a slightly higher bonk allows the camera to follow Luigi up to the gravity change, and it *seemed* to affect the angle and speed of the backwards wallkick too, although Im not convinced that thats the case.
This could be improved by 3 frames if this same fast wallkick can be done with the earliest long jump, with which I had no luck after many attempts (I was stuck with a 13.16 for a long time). In contrast, I got this fast wallkick after only a couple of minutes with the later long jump.
A frame of neutral joystick is used just after wallkicking - holding right straight away delays the acceleration to top speed for a few frames.
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
![[TAS] SMG2: Wild Glide Green Stars](https://i.ytimg.com/vi/s1fLQyWmg2k/mqdefault.jpg "[TAS] SMG2: Wild Glide Green Stars
Green Stars TAS WIP
The movement on the map uses an IR click and is one of the bigger differences of using IR, about 0.2s saved.
As there arent any cycles and the start is the same, there is no significance to star order.
For those that care, language difference is 28 frames from Japanese, 46 from Chinese.
Movement: I had done a Fleet Glide GS2 TAS a while ago, finding that the most optimal consistent speed was when the Y axis of the Wiimote gyro was at 498, a little bit above being completely level, which is the highest tilt that will accelerate.
But nemi had already noted (https://youtu.be/VmpawoTiNjI) that this wasnt necessarily indicative of speed, and that the best movement appears to be alternating between accelerating and lifting up.
Using the first straight section at 0:53 as a testing ground, I found that alternating between 498 (accelerate) and 496 (lift) every 44 frames was optimal, although, compared to other alternating times, it was only by fractions of a unit across 3 seconds of gameplay, so theres not much difference to RTA runs.
Compared to not alternating, though, it was 60 units further ahead at 12.15 on the timer, so about 1.3 frames faster.
The average speed went from 47.27 to 47.4.
The issue is that, particularly for GS2, this movement is too high for the star and some obstacles, so at some point I needed bigger downward tilt which lowered the average speed of the whole run and lost a lot of that potential timesave.
Green Star 2 - 1:04.65 IGT - 35:28.88 real-time split (2:00.75 segment)
The first level longer than 1 minute IGT.
Turnaround double jump was 4 frames faster than running to Jibberjay.
As stated earlier, I had to start tilting downwards a lot more nearer to the end, but a frame was saved by going further down to lift up with more speed into the Green Star. This animation also brings Luigi further forward which helped in the frame save.
Green Star 1 - 35.38 IGT - 36:47.35 real-time split (1:18.47 segment)
Backwards jump saved 8 frames here.
This star didnt need as much downward tilt so the optimal movement was more helpful. I did a slight lift up into the star but it didnt save any frames this time.
#supermariogalaxy2 #toolassistedspeedrun")
