User Files from mohoc

Upload All User Files

#638667544697225272 - Rework: The Bowl Is Empty TAS improvement ("long version") by Natetheman223

bowl-is-empty-long_fixed.ltm
In 04:44.47 (17068 frames), 1648 rerecords
Game: The Bowl Is Empty ( Linux, see all files )
27 downloads
Uploaded 11/9/2024 1:07 PM by mohoc (see all 26)
This is a reworked version of this submission by Natetheman223 which properly ends at the final message in the credits as soon as possible (saved 2 frames & cropped unnecessary inputs at the end).

#638658089145926423 - Minor fix: Super Mario Bros. Deluxe "warps" by KingKappa & Denial140

Super Mario Bros. Deluxe - L+R in 457.45 cycle fix.bk2
In 04:57.45 (17814 frames), 2640 rerecords
28 downloads
Uploaded 10/29/2024 2:28 PM by mohoc (see all 26)

#638653234684537309 - Rework: The Bowl Is Empty TAS by Natetheman223 (until final credits prompt)

MTBowl_including_end_credits.ltm
In 04:56.92 (17815 frames), 878 rerecords
Game: The Bowl Is Empty ( Linux, see all files )
26 downloads
Uploaded 10/23/2024 11:37 PM by mohoc (see all 26)

#638653234548162807 - Rework: The Bowl Is Empty TAS by Natetheman223 (until IGT stop)

MTBowlSHORT_fixed.ltm
In 04:07.33 (14840 frames), 880 rerecords
Game: The Bowl Is Empty ( Linux, see all files )
22 downloads
Uploaded 10/23/2024 11:37 PM by mohoc (see all 26)
This is a reworked version of this submission by Natetheman223 which properly ends at IGT stop (i.e. when the screen goes black at the beginning the final cutscene).

#638645327070068529 - Resync: Paper Lily - Chapter 1 TAS by Tegron

PaperLilyCh1_resynced_v1.x86_64.ltm
In 29:12.05 (105123 frames), 2638 rerecords
32 downloads
Uploaded 10/14/2024 7:58 PM by mohoc (see all 26)
This is a slightly reworked version of this submission by Tegron in the hopes of more consistent syncing.

File info

This movie was made on libTAS v1.4.6.
Additional libTAS settings
  • Runtime > Asynchronous events: check XEvents at frame beginning and SDL events at frame beginning.
Additional instructions
  • Pause at frames 2 and 33 for a few seconds so that the clock is guaranteed to disappear at frame 34.
FPS
  • Between 4 and 5 most of the time (i.e. it should take from 7 to 9 hours to go through the whole thing).
Total frames
  • 105123 (+19 from original movie)
Changes made to the submitted movie
  • 32100: 2 blank frames inserted.
    • Reason: late menu thread, prevents loading errors.
  • 33080: 2 blank frames inserted.
    • Reason: late menu thread, prevents loading errors.
  • 39738-39822: fishing timings updated (39752, 39788, 39822 instead of 39738, 39783, 39803).
    • Reason: adapt to different fishing RNG.
  • 39806: 15 blank frames inserted.
    • Reason: adapt to different fishing RNG.
  • 40198: two extra action presses, no insert.
    • Reason: catch the rabbit more consistently.
  • 90607-90720: go exactly the other way (180° symmetry), same death timing.
    • 114 'Right' inputs changed to 'Left', 69 'Down' inputs changed to 'Up'.
    • Reason: adapt to different monster spawn RNG in the third phase of Miss Knives' fight.

Known desyncing factors

General gameplay
  • Solution: in libTAS: Tools > Slow Motion > 12% (i.e. 7.5 FPS max).
Autosaves
  • Solution: in libTAS: Settings > Runtime> Asynchronous events: check XEvents at frame beginning and SDL events at frame beginning.
  • Side effect: max FPS drop to around 5.
Late 'retry' menu threads after intentional deaths
  • Solution: insert two blank frames at least a dozen frames prior to the menuing.
  • Side effect: changes RNG in the later parts of the run (notably the fishing minigame around frame 39700 and the monster spawns in the third phase of Miss Knives' fight).
    • Note 1: when such a delay happens, it seems to be consistently late by two frames.
    • Note 2: there are seven intentional deaths: the three Red traps, the painting chase and the three last phases of Miss Knives' fight. So far I only got delays at the second and third Red traps. Making a v2 of the resync where the 'retry' menu threads are always assumed to be late might lead to even more consistency.

Savestates to avoid desyncs from autosaves

Since autosaves are notorious desync points, it is best to make savestates in libTAS prior to them.
The table below describes two successful sync attempts with the resynced movie. The given frames correspond to the latest initialized thread before every autosave. You can access this piece of information by opening the Settings > Debug... window, setting Level to Debug and, in the Print section, (only) checking Threads.
If you get a thread timing which is late by only a couple frames then you will probably desync soon. But no matter what, even if you got a completely different timing, wait until you can tell that you desynced for sure.
When running the resynced movie, you can make a savestate a couple hundred frames prior to these timings and, if you get a desync, load your latest savestate or the one before it. Personally I don't make a savestate before every autosave. I usually wait 3000 to 10000 frames between my savestates.
Note that the Movie > Pause Movie at frame... option in libTAS can be useful to not miss the next desired savestate timing.
In the 'Save #' column (p) indicates that this autosave is a 'persistent' one and not a 'retrysave' one. In the 'Run 2' column '-' indicates that the timing was the same as in Run 1.
Save # Run 1 Run 2
1 11670 -
2 14150 -
3 14778 14641
4 19953 -
5 22971 23008
6 28506 -
7 30019 -
8 30174 30019
(p) 9 31277 -
10 31393 -
11 31393 -
(p) 12 32177 -
13 32380 -
(p) 14 33165 -
15 33468 -
16 34475 -
17 36425 36450
18 37505 -
(p) 19 38920 39016
20 39650 -
21 44431 -
22 44802 -
23 47841 -
24 56803 56929
25 57607 -
26 59480 -
27 59480 -
28 61291 -
29 61691 -
30 62770 -
31 69690 -
32 69884 -
33 74399 -
(p) 34 75639 75513
35 86903 -
(p) 36 90622 90555
(p) 37 90932 -
38 94947 -
39 94947 -
40 94947 -
(p) 41 99541 99331

Running the movie in 1080p

The base resolution is 720p. If you want 1080p follow these steps:
  • Go to ~/.local/share. If there is a Paper Lily folder delete it.
  • Open game executable PaperLilyCh1.x86_64 without libTAS.
  • Choose any language then go to Settings > Screen Settings.
  • Set Fullscreen to No and Resolution to 1920x1080.
  • Quit the game and open file ~/.local/share/Paper Lily/engine.cfg (any text editor should work).
  • Remove the line about translation. It should be the third line. If you chose English, it should be core/translation_selected="en".
Now you can run the game in 1080p with libTAS. As a reminder:
  • In Settings > Runtime check XEvents at frame beginning and SDL events at frame beginning. Leaving Prevent writing to disk checked is recommended.
  • In Settings > Video > Screen resolution choose Common 1920x1080 (16:9).
  • In Tools > Slow Motion choose 12%.

EDIT: added the encode and 1080p instructions.
EDIT 2: added the 'Savestates to avoid desyncs from autosaves' section.
EDIT 3: added the libTAS version (v1.4.6).

#638637681293120891 - Feedback: Mario's Super Picross - Level 1 (Mario) TAS by InputEvelution

SuperPicrossMario1TAS_feedback.bk2
In 03:06.96 (11236 frames), 2407 rerecords
29 downloads
Uploaded 10/5/2024 11:35 PM by mohoc (see all 26)
This is both an improvement and feedback post to this WIP by InputEvelution. The improvements are located in 1-A (1 frame), 1-D (2 frames), 1-F (2 frames), 1-G (3 frames), 1-I (3 frames). 1-J (2 frames), 1-K (1 frame) and 1-L (1 frame), totalizing 15 frames. (Of note that while this movie file is only shorter by 12 frames, it does lead to the post-(1-L) menu inputs 15 frames faster.)
In isolation the routing of each level looks very solid. I only managed to save one frame in a couple levels by balancing the total load between both players a bit better. To me this shows that the author has a good understanding of what an optimal two-path solution should look like, and how to obtain it empirically. However, there is something more important to note, and this was where most of the found time save came from.
Hammer animations are cyclic! And both must be completed before the level can end.
More precisely, Player 1's hammer animation is on an 8 frame cycle while Player 2's hammer animation is on a 7 frame cycle. After a hammer completes its last animation cycle, it visually disappears after exactly 14 frames. Then the end of each level unfolds with unchanged timing. So the timing of a level is conditioned by the completion time of the last cycle of each hammer animation. This behavior can be compared to the "frame rules" of Super Mario Bros., except there are two concurrent "frame rules" here, and you have control over when each individual "frame rule" cycle starts.
What this means is that there will always be a "limiting" hammer between the two, while the other one might have some margin: either some room in its last animation cycle or even enough time for an extra cycle. Knowing this, it is possible to track the advancement of the hammer animation cycles, identify the limiting hammer and, if possible, balance the load between both players so that the limiting hammer completes its last animation cycle as early as possible (sometimes enough to change the identity of the limiting hammer).
Basically:
Because of these hammer animation cycles, this looks like a pretty difficult optimization problem. Significantly more so than your traditional shortest path problem.
While this game as a TAS does not directly translate to a common computer science problem, it could be interpreted as a two-path covering problem on a planar graph, coupled with a batch scheduling problem on two machines with unit-time tasks. So brute-force seems inevitable.
While this game looks bottable, in my opinion it is possible to get a very clean result by hand. A good optimization strategy could be to ignore the hammer animations at first and find a solid solution with as few idle frames as possible (as it has been done by the author), and only then slightly tweak the level route to optimize the completion of cycles for both hammers.
Now that the hammer animation shenanigans are understood better, and given the quality of the first optimization phase achieved by the author, I can only be optimistic about the future of this project.
Finally, as a side note, it might be possible to use the dusting move in order to change the timing of the hammer animation cycles and save a few more frames. Though this is only a short in the dark.

#638615159712607211 - Mario Party 8 "Star Battle Arena" v1.0

v1_0.dtm
In 23:52.72 (85963 frames), 19682 rerecords
Game: Mario Party 8 ( Wii, see all files )
30 downloads
Uploaded 9/9/2024 9:59 PM by mohoc (see all 26)
This is my second movie of this game. Not a final product, but very close to it.

File info

  • Game version: USA (rev 1)
    • SHA1: 2252edd39961d3de9cf8ce529b721856960b8715
    • MD5: a2feb2d57ec670615ac9f1d13e1db49b
  • Vi frame count: 51357
  • Lag frames : 567
  • RNG manipulation: 168 frames in 30 fps sections and 25 frames in 60 fps sections

Emulator settings

  • Dolphin 5.0 with Lua Core v4.3 by MikeXander (syncs on 5.0 stable)
  • Dual Core: off
  • Idle Skipping: off
  • JIT Recompiler
  • DSP HLE emulation
  • Wii aspect ratio: 4:3
  • No GameCube Controller, one Emulated Wiimote

General info

This is "Star Battle Arena", the solo mode of Mario Party 8 which leads to the game credits. You face an opponent in the six game boards, and then defeat Bowser in a final boss fight.
This improves my previous movie by close to twenty seconds from a faster King Boo board layout, better RNG manipulation and attempting to reduce lag. I made and used this Lua script to assist me in the RNG manipulation process, with great success!
Routing wise this is very close to my previous attempt, with the exception of King Boo's board which I was able to complete in two turns without needing a candy or a DK space.

Known improvements

By far King Boo's board has the biggest potential, with up to ten seconds to be saved with an ideal layout: one less coin block animation (4s), two less branches (2 x 1.5s), no shop on the way (1s) and at least one King Boo place two rooms away from the start instead of three (1s). The board layout is entirely dictated by RNG, so there is no telling if such a layout can be found in a reasonable time, or if it is obtainable at all. Plus for now the board layout must be checked manually, which is quite time consuming. My goal for this second attempt was to use no candy and no DK space, and in this regard it was successful.
Other than that I lost 1 frame compared to my planned route. In turn 2 of Shy Guy's board my opponent landed one space further than planned (7 frames slower). However it was a red space instead of a blue one, which saved 6 frames due to a shorter animation. While this was 1 frame slower than planned, I kept it because I was not able to manipulate RNG well enough in order to get a faster time locally with the planned strategy.
Although much better than in my first attempt, I feel that my RNG manipulation could still be a bit better in the last two boards. Though at this point it is ruled by my luck in my... RNG manipulation process...
Finally I am just getting started with lag management. I would say it was fine overall except in the final boss fight where I got nineteen additional lag frames compared to my first movie. So I will experiment and try to improve this part at the very least before submitting this movie.
EDIT: minor rewording

#638615107657589700 - Mario Party 8 RNG manipulation helper v1.0

MP8_RNG_manipulation_helper.lua
Game: Mario Party 8 ( Wii, see all files )
28 downloads
Uploaded 9/9/2024 8:32 PM by mohoc (see all 26)
My first Lua script. Enjoy!

#638601179598592372 - Mario Party 8 "Star Battle Arena" v0

v0_51930_frames_341_5_manip_21_extra_605_lag.dtm
In 24:11.93 (87116 frames), 6354 rerecords
Game: Mario Party 8 ( Wii, see all files )
1 comment, 33 downloads
Uploaded 8/24/2024 5:39 PM by mohoc (see all 26)
This is my first full-game movie on Dolphin, I hope that it syncs and that you will enjoy it!

File info

  • Game version: USA (rev 1)
    • SHA1: 2252edd39961d3de9cf8ce529b721856960b8715
    • MD5: a2feb2d57ec670615ac9f1d13e1db49b
  • Frame count: 51931
  • Lag frames : 605
  • RNG manipulation: 323 frames in 30 fps sections and 37 frames in 60 fps sections

Emulator settings

  • Dolphin 5.0 with Lua Core v3.5.1 (syncs on 5.0 stable)
  • Dual Core: off
  • Idle Skipping: on
  • JIT Recompiler
  • DSP LLE recompiler
  • Wii aspect ratio: 4:3
  • No GameCube Controller, one Emulated Wiimote

General info

This is "Star Battle Arena", the solo mode of Mario Party 8 which leads to the game credits. You face an opponent in the six game boards, and then defeat Bowser in a final boss fight. This RNG gauntlet has been quite a popular speedrunning category with 58 entrants on speedrun.com and a WR time in 50:15 by StillSoftSpoken. In contrast a TAS of this game mode is expected to take around 24 minutes.
This movie file took inspiration from the strategies used by both Mizumaririn in their WIP up to the end of King Boo's board and the full TAS run of 豪華客船乗組員 on Nicovideo (part 1 up to the end of King Boo's board, part 2 for the rest). Mizumaririn had the fastest strategies and times known at the time on the first three boards, while the Nicovideo TAS was the state of the art for the rest of the run. I also tested some creative strategies from Truepixel__'s playlist but none were actually faster (and it is a shame given my love for several of these strategies).
Routing wise this TAS reused the known fastest strategies on DK, Shy Guy and Bowser's boards. I brought minor improvements to the strategies in Goomba and Koopa's boards: respectively landing on a red space in turn 1 instead of a blue space (6 frames faster) and better coin route/hotel investment (roughly a couple seconds saved). A detailed description of the board routes is given in this Google doc and a detailed time breakdown of the movie is available in this Google spreadsheet.

Known improvements

I lost 21 frames compared to my planned route. In Goomba's board my opponent got 10 coins from their coin block instead of 5 (15 frames slower), and in turn 2 of Shy Guy's board I landed on a blue space instead of a red one (6 frames slower). Both inaccuracies were kept because I was not able to manipulate RNG well enough in order to get a faster time locally with the planned strategies.
My RNG manipulation was also not ideal, especially in the first two boards. This makes sense because this was my first experience with it and my progress is already noticeable by comparing the first half of this movie to its second half.
I intend to improve this in the future, notably with help from better RNG manipulation. I have ideas to automate and make my RNG manipulation process more efficient with Lua scripting.
EDIT: added encode link + minor rewording

#638567371142506129 - All Make Up, No Damage Normal (Legacy)

All_Make_Up_No_Damage_Normal_Legacy.bk2
In 11:39.54 (41782 frames), 315 rerecords
54 downloads
Uploaded 7/16/2024 2:31 PM by mohoc (see all 26)
This is an overhaul of the 2015 "Cheat Bot 100% run" by Matt Shepcar. Most of the level input sequences are reused, with minor time improvements (less than a second overall) and some 'quality of life' changes (e.g. less shakey camera movement).
Emulator used: Bizhawk 2.9.1 with mGBA core
Twenty of the thirty level input sequences were obtained (and kindly made available) by Matt himself. The other ten were significantly redone with help from KuruBot by E-Sh4rk.
The ten majorly redone levels are all of Grasslands, Jungle 1 and 2, Cake Land 2, Cave 3, Cloud Land 3, Ice Land 1 and Ghost Castle 2.
Compared to Matt's video all level times are matched, and even sligthly improved in seven of them: Ocean 2 (-0.25s), Ocean 3 (-0.18s), Cake Land 2 (-0.05s), Cake Land 3 (-0.15s), Cloud Land 1 (-0.03s), Ice Land 3 (-0.07s) and Ghost Castle 2 (-0.15s).
This is the first step of a bigger 100% TAS project. The next steps involve adding all the other levels done with no damage, then perfecting the presentation of the run with even less shakey camera movement and better downtime entertainment.
EDIT: Due to loading time discrepancies between Bizhawk 2.9.1 and 2.3.1, I did not notice that several levels were still one or two frames longer than in Matt's video. Here is a detailed breakdown:
- seven levels are worse by one frame: Cave 2 & 3, Cloud Land 2 & 3, Ice Land 1 & 2 and Machine Land 3,
- two levels are worse by two frames: Machine Land 2 and Ghost Castle 3,
- the twenty-one other levels are matched or improved.

#54179529352803165 - Best Ending Normal w/ Training v1.2.1

Best_Ending_Normal_With_Training_v1.2.1.bk2
In 06:11.58 (22194 frames), 0 rerecords
667 downloads
Uploaded 3/28/2019 11:21 PM by mohoc (see all 26)
An 8.15s improvement to this submission, exclusively from better menuing in the first minute of the run.

#54086380840886059 - Best Ending Normal v0

Best_Ending_Normal_v0.bk2
In 06:08.01 (21980 frames), 742 rerecords
692 downloads
Uploaded 3/24/2019 6:40 PM by mohoc (see all 26)
Kuru Kuru Kururin done in 6:06.33 (21980 frames) as stated in the Best Ending Normal rules on speedrun.com (complete all thirty main levels, save the birds).
The TAS starts at Power On with the GB intro.
This was put here for legacy.
OOB viewer available here

#52647444906124211 - Any% Normal v1

Any_Normal_v1.bk2
In 05:20.87 (19165 frames), 281 rerecords
765 downloads
Uploaded 1/18/2019 11:24 PM by mohoc (see all 26)

#52269947472336163 - Ghost Castle 1 done with wall clipping + OOB goal zone

w10_ghostcastle_1_oob.bk2
In 00:14.78 (883 frames), 688 rerecords
766 downloads
Uploaded 1/1/2019 11:23 PM by mohoc (see all 26)
Ghost Castle 1 done in 614 frames (from selecting the level on the world map to the first frame when Kururin looks happy).
It is definitely improvable since it used the "slow" early wall clip available here (that waits for an additional 180° to adjust position): http://tasvideos.org/userfiles/info/52237142160912306
This has been done with help from Kururin OOB view v1.4 (beta) available here: https://github.com/E-Sh4rk/KururinTAS
Savestate: world map on Ghost Castle 1.

#52269764288546723 - Machine Land 1 done with wall clipping + OOB goal zone

w9_machineland_1_oob_v2.bk2
In 00:09.51 (568 frames), 459 rerecords
773 downloads
Uploaded 1/1/2019 11:11 PM by mohoc (see all 26)

#52269660021613854 - Cloud Land 2 done with wall clipping + OOB goal zone

w6_cloudland_2_oob.bk2
In 00:09.39 (561 frames), 657 rerecords
777 downloads
Uploaded 1/1/2019 11:04 PM by mohoc (see all 26)
Cloud Land 2 done in 303 frames (from selecting the level on the world map to the first frame when Kururin looks happy).
This has been done with help from Kururin OOB view v1.4 (beta) available here: https://github.com/E-Sh4rk/KururinTAS
Savestate: world map on Cloud Land 2.

#52269565877650126 - Cave 3 done with wall clipping + OOB goal zone

w5_cave_3_oob.bk2
In 00:09.23 (551 frames), 511 rerecords
738 downloads
Uploaded 1/1/2019 10:58 PM by mohoc (see all 26)
Cave 3 done in 260 frames (from selecting the level on the world map to the first frame when Kururin looks happy).
The second wall clip used a similar method as in this Ghost Castle 1 wall clip: http://tasvideos.org/userfiles/info/52237142160912306
The few weird inputs before the second wall clipping were done for adjusting position.
This has been done with help from Kururin OOB view v1.4 (beta) available here: https://github.com/E-Sh4rk/KururinTAS
Savestate: world map on Cave 3.

#52269414949784408 - Cave 1 done with wall clipping

w5_cave_1_matt.bk2
In 00:05.54 (331 frames), 1113 rerecords
725 downloads
Uploaded 1/1/2019 10:48 PM by mohoc (see all 26)

#52237265785563966 - Ghost Castle 2 done with two wall clips

w10_ghostcastle_2_oob.bk2
In 00:10.73 (641 frames), 802 rerecords
769 downloads
Uploaded 12/31/2018 12:04 PM by mohoc (see all 26)
Ghost Castle 2 done in 312 frames (from selecting the level on the world map to the first frame when Kururin looks happy).
I used a first wall clip to escape the starting zone and go out-of-bounds, then another one to go inbounds again and reach the regular goal zone.
The previously known wall clipping techniques did not work here for some reason, so I had to find new ones.
There are probably faster strategies, but this is a good start.
Savestate: world map on Ghost Castle 2.

#52237142160912306 - Ghost Castle 1 early wall clip

w10_ghostcastle_1_EarlyClip_Wait180.bk2
In 00:06.20 (370 frames), 606 rerecords
757 downloads
Uploaded 12/31/2018 11:56 AM by mohoc (see all 26)
A new way to escape the starting zone in Ghost Castle 1.
The method given in Grasslands 3 (http://tasvideos.org/userfiles/info/52177347080337677) does actually not work here because of the springs on the bottom wall. So I wait for an additional 180° to give me time to adjust my position.
There are probably faster strategies, but this is a good start.
Savestate: world map on Ghost Castle 1.

#52179138490002530 - Grasslands 1 done with wall clip

w1_grasslands_1_matt.bk2
In 00:04.51 (270 frames), 4 rerecords
733 downloads
Uploaded 12/28/2018 9:14 PM by mohoc (see all 26)
Replica of the beginning of Grasslands 1 from Matt Shepcar's "cheat bot" video: https://www.youtube.com/watch?v=DXB5DTYUei4
It could become the starting segment of a potential new "fastest completion" TAS of the game.

#52177977673438241 - Cake Land 2 early wall clip

w4_cakeland_2_EarlyClip.bk2
In 00:02.31 (138 frames), 2 rerecords
714 downloads
Uploaded 12/28/2018 7:59 PM by mohoc (see all 26)
Replica of the beginning of Cake Land 2 from Matt Shepcar's "cheat bot" video: https://www.youtube.com/watch?v=DXB5DTYUei4
It collides with the top wall of the starting zone, then it clips through the right wall.
It is also used in Jungle 3 in Matt Shepcar's "cheat bot" video.
Doing a 180° of the inputs makes it work for other levels, notably: Cave 1, Cave 2, Cloud Land 1, Star Land 3, Ice Land 2, Machine Land 1, Ghost Castle 3.
(or any level where you start turning counterclockwise with both top and right walls or both bottom and left walls available in the starting zone).
Savestate: world map on Cake Land 2.

#52177778535894963 - Grasslands 2 done with wall clip + OOB goal zone.

w1_grasslands_2_oob.bk2
In 00:06.38 (381 frames), 74 rerecords
723 downloads
Uploaded 12/28/2018 7:46 PM by mohoc (see all 26)
Grasslands 2 done in 169 frames (from selecting the level on the world map to the first frame when Kururin looks happy).
It clips through the left wall of the starting zone, then it reaches an out-of-bounds goal zone.
I used the method that I found in Grasslands 1 in order to clip through the wall: http://tasvideos.org/userfiles/info/52177554086850103
Savestate: world map on Grasslands 2.

#52177554086850103 - Grasslands 1 early wall clip

w1_grasslands_1_Earlyclip_UL.bk2
In 00:04.51 (270 frames), 6 rerecords
747 downloads
Uploaded 12/28/2018 7:31 PM by mohoc (see all 26)
A new way to escape the starting zone in Grasslands 1.
It collides with the top wall of the starting zone, then it clips through the left wall.
I did a 180° of the inputs from the following wall clip in Grasslands 3 and it worked: http://tasvideos.org/userfiles/info/52177347080337677
It could potentially be used in other levels, notably: Grasslands 2, Ocean 3, Jungle 1, Jungle 2, Cake Land 1, Cake Land 3, Cloud Land 2, Cloud Land 3, Star Land 2, Machine Land 2, Machine Land 3, Ghost Castle 2.
(or any level where you start turning clockwise with both top and left walls available in the starting zone).
Savestate: world map on Grasslands 1.

#52177347080337677 - Grasslands 3 early wall clip

w1_grasslands_3_EarlyClip_DR.bk2
In 00:02.06 (123 frames), 2 rerecords
804 downloads
Uploaded 12/28/2018 7:18 PM by mohoc (see all 26)
A new way to escape the starting zone in Grasslands 3.
It collides with the bottom wall of the starting zone, then it clips through the right wall.
I took inspiration from this wall clip in Ice Land 3: http://tasvideos.org/userfiles/info/52172978644903455
It could potentially be used in other levels, notably: Ocean 1, Ocean 2, Cave 3, Star Land 1, Ice Land 1, Ghost Castle 1.
(or any level where you start turning clockwise with both bottom and right walls available in the starting zone).
Savestate: world map on Grasslands 3.