Uploaded August 2020 | Updated September 2026, 1 week ago
Nemi ( youtube.com/channel/UCtezaVs5vcwl3A6uwIFJPQg ) had been toying with this idea so I tested it to see how much faster it was, and even this unoptimised attempt saved 2.7s over the TAS demo!
It may be faster to avoid the third Octopus and triple bounce off the Toad instead, but this method made the angle to Toad much easier to figure out, as it was virtually straight up.
This is tool-assisted, mainly just savestates with Dolphin running at 25% speed.
Nemi ( youtube.com/channel/UCtezaVs5vcwl3A6uwIFJPQg ) had been toying with this idea so I tested it to see how much faster it was, and even this unoptimised attempt saved 2.7s over the TAS demo!
It may be faster to avoid the third Octopus and triple bounce off the Toad instead, but this method made the angle to Toad much easier to figure out, as it was virtually straight up.
This is tool-assisted, mainly just savestates with Dolphin running at 25% speed.
![[TAS] Super Mario Galaxy 2: Haunty Halls Green Star 2 in 11.90s](https://i.ytimg.com/vi/Sy6O893-okk/mqdefault.jpg "[TAS] Super Mario Galaxy 2: Haunty Halls Green Star 2 in 11.90s
Luigi backflips off a bump in the path to a Green Star in 11.90 seconds [IGT, 714 frames]
The camera pan at the start moves the second planet off-screen, stopping the paths cycles for a moment. This was helpful as it meant the path to the star needed less meandering, so I could backflip closer to the star.
The camera cancel on the second planet is frame-perfect, pressing up on the D-pad on the first landing (dust) frame. A jump cancel is also possible here, which is also frame-perfect.
It was 10 frames faster to land on the bump and backflip, rather than continue the long jump closer to the star but landing on lower ground.
I had had a cooler idea for the star grab involving the broken walls but it ended up over 1 second slower. Here it is on my second channel: https://youtu.be/TccadovLC7c
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
![[TAS demo] SMG2: Honeybloom GS2 in 58.90s](https://i.ytimg.com/vi/TP0NNyl4FDg/mqdefault.jpg "[TAS demo] SMG2: Honeybloom GS2 in 58.90s
Improvements were made thanks to cutscene cycles and the 11f advantage of how the in-game timer starts from cutscenes, starting further along with a few frames of flying speed before grabbing the vine.
The first improvement is being able to squeeze between the Choppah (I learned the name!) and the ledge at the end of the waterfall section, rather than having to go over the Choppah.
This squeeze also meant that the ledgegrab instantly climbed up, and also gave a huge boost of horizontal speed, up to 32 units/frame, that could be carried briefly in a single jump (but the jump was shortened as a longer jump wouldnt regain running speed as quickly when landing. More optimisation to be had here).
Another improvement was in the bouncy fruit section, where the first Choppah goes low enough that a bounce gets cut off of its height just enough to get to the cloud, while quickly getting to the faster hspd of falling as Bee Luigi.
And then P2 was used to hold the second Choppah in a good position to get the damage boost onto the next ledge.
The squeeze between the Choppah and the ledge at the end of the pole section required an early P2 A press so P2 would jump up, and resulted in a ledge grab spin bonk as the spin was still active when grabbing the ledge, which may be faster than spinning and waiting to grab the ledge as the squeeze makes Luigi land very quickly and in such a way that he starts running instantly, with the horizontal speed getting the same as the squeeze in the waterfall. A different landing would have been stuck running on the edge for a moment.
Overall, this run isnt highly optimised with many frame saves to be made and other little bits of movement to experiment with.
#SuperMarioGalaxy2 #TAS")
![[TAS] Super Mario Galaxy 2: Flip-Swap Green Star 2 in 17.38s](https://i.ytimg.com/vi/VJuskE1EpWE/mqdefault.jpg "[TAS] Super Mario Galaxy 2: Flip-Swap Green Star 2 in 17.38s
Luigi launches off flipping panels to collect a Green Star in 17.38 seconds [IGT, 1043 frames].
This is fast, but theres so much nuance to flip panel jumps, and waiting the cutscene a bit would give better electric beam cycles, so an IL time could go sub 17.
For the first panel jump, I spin cancel long jump on the panel to trigger another flip. This gives a strange push in mid-air in the direction the flip panel moves (in this case north-east). Displacement? I dont know the technical side of it.
This also means the blue panels are active, and landing on that one was necessary to avoid the electric beams - again, waiting the cutscenes could be better for in-game time as you could do a shorter panel jump without the spin cancel and go over to the red panels instead.
P2 comes in handy to hit the checkpoint so that Luigi doesnt hit it and lose all speed when he spins. This helps to set up a stronger blue panel jump that makes it all the way to the final panel. I learned about this blue panel strat from datteos IL of the main star: https://youtu.be/M-Fss5TKjWE
This final panel flings straight towards the star.
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
![[TAS] SMG2: Shiverburn Green Stars](https://i.ytimg.com/vi/VOCZo6UDmHo/mqdefault.jpg "[TAS] SMG2: Shiverburn Green Stars
Green Stars TAS WIP
Star order has no real effect - all three stars have the same start so start position difference doesnt matter.
Green Star 3 - 35.45 IGT - 1:51:51.80 real-time split (1:31.94 segment)
This is an improvement of 1.53s over my 2020 TAS, even without land cancel: https://youtu.be/L1khyJTw79E
No land cancel here.
This route to the Cloud Flower saved 6 frames over starting with the 2nd green platform, and getting a height boost off the blue platform.
The most interesting part was the route from getting clouds and skipping to the launch star, and how many clouds would be best to have at the end.
The route RTA uses - going back to spawn and then long jumping out of bounds from there - is possible with just 1 cloud, leaving 2 at the end.
This route - which is later used for GS2 - is barely possible with 2 clouds (even when entering gravity, I have to slow down for a few frames to avoid swinging out into the deathplane). This route is 92 frames (1.53s) faster than the other.
But on the last planet, having 2 clouds would allow for a double cloud long jump route that saves 0.65s over the standard 1-cloud strat.
So overall, the final route - 2 clouds at the start, 1 at the end - was 0.88s faster.
On the last planet, I saved 3 frames by turning the camera so Luigi was moving to the left in its perspective.
For some reason, you get ever so slightly more speed moving left/right to the camera compared to moving away from it, and much more compared to moving towards it. https://docs.google.com/document/d/14DAcbHw0OPSL2psmxpvAPtpAVdF_AYkYJyeE8GNZKjM/edit
This mechanic is difficult to make useful in other parts of the game, as often times the camera either cant move or its constantly fighting to get back to being behind Luigi, with its back-and-forth swinging making optimising movement an exhausting task without some kind of script or visualiser that we dont yet have.
Green Star 2 - 19.48 IGT - 1:52:55.11 real-time split (1:03.31 segment)
This is an improvement of 0.5s over my 2020 TAS: https://youtu.be/QdUV1xkWiuo
Not much to note besides the trick itself catching the edge of gravity right above the star into a cool precise spin-cancel ground pound.
Green Star 1 - 17.91 IGT - 1:53:56.86 real-time split (1:01.75 segment)
This is an improvement of 0.58s over my 2020 TAS: https://youtu.be/thg0ekOZxfg
On the 2nd cloud long jump to the button, I start walking up and then turning and long jumping before completing the turn. This makes it so Luigi turns around quicker when getting up from the ground pound, saving a few frames.
Despite the wayward zig-zag, the movement on the frozen volcano saved 2 frames over going straight over the higher lip to the left.
#supermariogalaxy2 #toolassistedspeedrun")
![[TAS] SMG2: Flash Black Green Stars](https://i.ytimg.com/vi/WFnBtg0WeTc/mqdefault.jpg "[TAS] SMG2: Flash Black Green Stars
Green Stars TAS WIP
Star order would matter due to Paragoomba and Jack OGoomba cycles, however use of 2P negates the differences. GS1 was still picked first from the pre-2P route.
Green Star 1 - 16.78s IGT - 2:01:53.73 real-time split (1:13.05 segment)
This is an improvement of 2.97s over my 2020 IL TAS, despite cutscenes: https://youtu.be/WBvi2DMZO9M
Even against the Blimp boost tech explained later, climbing Yoshiless is faster by a couple of seconds.
A megasideflip is a glitch that still isnt quite understood on a technical level, but requires a certain camera type (XZ_PARA) and angle, and then sideflipping and holding a direction perpendicular to facing angle. See more here: https://docs.google.com/document/d/1a2wWBhFL7hTVEY9zJmtTk5CEbxiib3lim9hSOCkAWQw/edit#heading=h.54goeomx2quh
This megasideflip route is 1s faster than this zip jump triple jump route: https://twitter.com/PurpleSunSR/status/1801718542540411149
The megasideflip was made even faster by aiming to grab the ledge. Since this is inside the 2D section, and entering a 2D section from the front slowly forces Luigi into the 2D restriction, grabbing the ledge results in a 1-frame climb.
Grabbing Yoshi at the end would only be 0.5s slower than what was done here.
Usernameispassword suggested starbitting the Paragoomba to bounce off it in mid-air. This strat saves 0.37s over the mega-wallkick route done in the 2020 TAS.
Green Star 2 - 31.15s IGT - 2:03:07.98 real-time split (1:14.25 segment)
This is an improvement of 5.18s over my 2020 IL TAS: https://youtu.be/XJqTklGuRrg
Thanks to the Paragoomba cycles, it is 3 frames faster to bounce-spin off one to the wallkick up to the Jack OGoomba, rather than the double wallkick GS1 has to settle with (2P cant help GS1s cycle because it needs the Paragoomba to move to the left).
Taking Yoshi at the end of the climb ends up 0.8s faster than going the rest of the way Yoshiless (which would use a zip jump).
The Blimp boost tech involves freezing a Paragoomba so Blimp Yoshi can go through it, and then spinning it to make it solid inside Blimp, causing it to push Blimp out at high speeds (like a proxy in Spyro games). This one was 121 initial y-speed - Blimps normal is 7 y-speed.
UPDATE: Improved ending by 0.39s here: https://youtu.be/emYeiQOWsKY
At the end, backwards infinite flutter from the start is 0.5s faster than going over to the crystal for a zip jump.
I slightly angle the flutter to drift off of a straight line to the star in order to be over a void around the middle part.
After accelerating to top speed with fluttering, you eventually lose speed if youre fluttering close to the ground, but maintain it over void or high above ground. This saved a few frames.
I also turned the camera here - moving to the side of the screen is slightly faster than moving towards or away from it. This saved 0.2s. See more here: https://docs.google.com/document/d/14DAcbHw0OPSL2psmxpvAPtpAVdF_AYkYJyeE8GNZKjM/edit
at 150k rerecords now
#supermariogalaxy2 #toolassistedspeedrun")
![[TAS] Super Mario Galaxy 2: Spin-Dig Green Star 2 in 38.45s](https://i.ytimg.com/vi/WJgs-aUh5bc/mqdefault.jpg "[TAS] Super Mario Galaxy 2: Spin-Dig Green Star 2 in 38.45s
Luigi and P2 drill and backflip to a Green Star in 38.45 seconds [IGT, 2307 frames].
At the start, I tried long jumping and bouncing off the digger, but its slower than just jumping down and triple jumping up, especially when you can triple jump up without having to spin, which saved a few frames.
I turned the camera just before talking to the Luma so it didnt have to pan as far to its resting point, making the textbox appear 1f earlier.
On the second planet, I used the IL strat of getting the drill clip on the underside: https://youtu.be/BcsOtQoJuXU?t=25
This saves 0.75s over clipping from the first drill.
After making this, I had found I could have HGPd the drill 3 frames earlier. I found this out while comparing camera cancel with a new jump cancel idea, with camera cancel still coming out 1 frame ahead. I also may have found 1 frame of timesave for both drills when doing cutscene cycle tests.
At the Luma, I again turned the camera to save 1 frame of camera cutscene movement.
On the third planet, I turned the camera at a certain point to get the yellow appearing platforms off screen to freeze their cycles. This meant that I reached the second row of platforms as they reappeared, and when they reappear, they move slightly upwards - side note: the collision is the same size despite the model expanding - so a backflip could gain a bit more vertical speed. This saved 7 frames.
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
![[TAS] SMG2: Lava Lair Comet in 0xA presses [Mario]](https://i.ytimg.com/vi/WhZmjb_kUiY/mqdefault.jpg "[TAS] SMG2: Lava Lair Comet in 0xA presses [Mario]
Mario speeds through Lava Lair to grab a star in 0 A presses.
This improves on the previous 1xA count by using the Whomp as a sliding tool to get to the meteor.
This star was an interesting one, as beforehand the two 1xA routes were either: Jump on spring at the start and use the Whomp with an 0.5xA strat for the meteor, or use the pull star with B to skip the spring and jump on the meteor.
It turned out that the Whomp strat was possible 0xA, but I was sceptical that thered be enough time on the limit to bring all the strats together.
As stated, pull stars can be grabbed with B. The IR appears a second time to manipulate Marios rotation inside the bubble so hes facing away from the spring; backwards speed is faster in the air.
I stayed on the moving platforms rather than the lavaboost that is done in other stars so that I could fireboost into a clock later on and still have the health to lavaboost into the star. This lost 3 seconds in real-time but gained 7s profit on the time limit.
I got lucky with the Whomps cycle, although P2 could have helped if necessary.
The slide requires standing on a certain point as the Whomp stands up; a bit too far up or down would still slide, but not get any speed in the air. A quick turn can be done just before sliding to face backwards.
I couldnt get that around the meteor far enough to hit the gate immediately, but was able to use some gravity trickery as the next one came up to be able to pound that instead.
Lava boost into snake block clip at the end for style points.
#SuperMarioGalaxy2 #AButtonChallenge")
![[ABC TAS] SMG1: Melty Molten 2 in 0.5x A presses (Luigi)](https://i.ytimg.com/vi/WuQdDXih1zk/mqdefault.jpg "[ABC TAS] SMG1: Melty Molten 2 in 0.5x A presses (Luigi)
Luigi gets through the meteor storm in 0.5 A presses.
The half A press only counts for the level itself, essentially noting that the A press for selecting the star should be held throughout the whole level.
Its required for a higher spin to get the star.
On the Thwomp planet, its possible to get clipped into the floor. By getting under the edge of a Thwomp as it comes down, Luigi will sidewarp (pushed with a lot of speed in 1 frame) instead of getting crushed.
This sidewarp clips thanks to the cornered floor; running from the top side of the corner means Luigis tilt doesnt update to the diagonal floor the Thwomp crushes, so instead of going up-right, the sidewarp goes straight right and through the corner.
This was found by Annoying (A Good Duck) who discovered a consistent clip if you die and then hold left on the respawn: https://youtube.com/watch?v nGMh75Dhrg
Its necessary to get the clip facing left in order to get the faster backwards acceleration to get past the big block.
Afterwards, a simple backwards spin knockback got up the small block.
The Life Mushroom isnt necessary - I got that to try a star grab strat using lava boosts.
During the chip section, I did edge grinding to save frames, running along the edge to start falling off but being able to stay on, using precise angles to stay in the fall off (+ ratio) state - this has a slightly faster speed than running.
The really slow grinding... thats definitely RNG manipulation. Trust me, bros.
At the end of the Bullet Bill planet, I entered the launch star while falling off a ledge to get displacement, leading to sliding along the floor and bonking on the next planet which saves some time.
While I avoided coins for most of the ball section, I couldnt find a way to get to the final platform while avoiding the coin ring - it was either too steep an angle or too little speed turning to roll onto the lava platform.
The star grab should not have worked. The weakspin (little twirls after spinning) was frame-perfect. Hitboxes really are crazy. (Sorry for no real technical explanation besides weakspins having a little extended hitbox and star hitboxes being so random).
ty Xander for encode
#supermariogalaxy #abuttonchallenge")
![[TAS] SMG2: Cloudy Court Green Stars](https://i.ytimg.com/vi/X295KachVO8/mqdefault.jpg "[TAS] SMG2: Cloudy Court Green Stars
Green Stars TAS WIP
Star order matters a bit here, as the land cancel affects the different starting routes, but is no different between GS3 and GS2 (once getting to their proper split at the drums), but 8 frames slower in GS1.
Green Star 3 - 48.73 IGT - 54:20.85 real-time split (1:44.65 segment)
2.43s saved over a rough 2020 TAS
While the starting route is a little later bouncing on the first drum than the route used in GS2, it makes up for it with the backwards facing bounces going further to the right to make the big skip a little faster.
This version of the skip is 1.1s faster than the IL route starting cloud long jumps from the cymbal.
An alternative version that starts with a cloud sideflip was only 2 frames slower.
From the sloped tower in the 2D section, I landed on the very edge where Luigi would long jump off a slightly lower floor and get a low long jump.
The effect of this compared to a normal long jump then carries on through the rest of the star making for different jumps up the crusher. Overall, the route saved 4 frames.
Green Star 2 - 18.65 IGT - 55:22.83 split (1:01.98 segment)
0.45s saved over my 2020 TAS: https://youtu.be/_K_sa1GU4DY
As explained above, this starting route bounces on the first drum a little earlier, but is faster here.
Frame-perfect cloud spin cancels and long jumps get a much higher boost than most mid-air cloud ljs (about 23-24 initial yspd compared to anywhere between 13 and 18 - and those 18spd jumps have to be very delayed).
Theyre enough to start jumping from almost under the last drum, and save 1 second over a single cloud lj from the cymbal.
Green Star 1 - 5.35 IGT - 56:11.51 split (48.68 segment)
2f saved over my 2020 TAS: https://youtu.be/55cF7iXMaFA
Landing in neutral after the first cloud spin cancel conserved a little bit of speed, enough to be able to instantly long jump. However, much more height and a better angle was gained in the long jump by delaying it a few frames.
Originally I had two routes tied at 5.36 - one was the 2020 route (lj + 2 cloud sideflips), another was instant backflip + backwards single jump + cloud sideflip.
However, @EpicRandomGamingChannel (Xander) then challenged me to untie one of those routes, helping me out in VC.
We ultimately found a slightly better 2020 route by doing an instant backflip from the long jump - this has less y speed but more horizontal speed than the first sideflip (which goes away from the star), and this meant an earlier cloud sideflip (that goes toward the star) could get under the star to hit it at a low weirdly extended point and save the frame.
#supermariogalaxy2 #toolassistedspeedrun")
![[TAS] Super Mario Galaxy 2: Flash Black Green Star 2 in 36.33s](https://i.ytimg.com/vi/XJqTklGuRrg/mqdefault.jpg "[TAS] Super Mario Galaxy 2: Flash Black Green Star 2 in 36.33s
Luigi, Yoshi and the P2 Luma zoom across a planet to collect a Green Star in 36.33 seconds [2180 frames, IGT]
This isnt highly optimised, but was part of testing three different routes for this star: Full Yoshi, Late Yoshi, and Yoshiless.
Yoshiless is the fastest method for GS1, so it would make sense if it was also fastest for GS2, but because of the infinite flutter on the second planet, Late Yoshi is the fastest, being faster than Yoshiless by nearly 2 seconds, and faster than Full Yoshi by nearly 7 seconds.
[The less optimised parts were more around the start, so they didnt affect the differences between routes an awful lot.]
P2 is useful for two reasons: 1) it can freeze the Paragoombas, so one doesnt get in the way up to the wallkicks (and saves losing 3f of hitlag), and another can be bounced off with Blimp Yoshi, and 2) it can break Yoshi eggs earlier.
Jumping out of the bonk landing is ~0.5s faster than waiting to be able to move.
#SuperMarioGalaxy2 #ToolAssistedSpeedrun")
