TAS discussion

[Karisa]

Discuss Touhou TASes here.

List of Touhou Tool-Assisted Score Records


Recent news: kopiapoa counterstopped WBaWC on Normal with Youmu-Otter. The video also includes two real-time Lunatic counterstops, for comparison I suppose?

[KirbyComment]

Since late April this year I learned how to make TASes myself and have done 5 scoring TASes of DS and 1 survival TAS of GFW. I have them archived in this playlist. They aren't included in arcorann's list and survival TASes by default aren't. It should be noted that kopiapoa, chum and LET have done scoring TASes of StB which also aren't included there. I plan on doing a GFW Lunatic LNNN TAS of a route other than A1 and more DS scoring TASes in the future.

[Karisa]

Considering Arcorann's said feel free for others to update the list, and new sections have been added over time, seems like those TASes would belong on his page? Not sure if Arcorann is still around to ask.

Also, I didn't know you'd learned to TAS. It's a shame GFW no ice is so slow from low shot power, since aside from that it's an interesting watch. I liked how you also went for full spell (I'm surprised that's possible on power level 1) and no laser damage.

[Arcorann]

I haven't touched the page in years. While I don't really have as much of an interest in survival-only TASes myself, I don't mind if you add them.

Having said that, this thread seems like a decent place to compile things as well, so I guess we can maintain both.

[KirbyComment]

I see. I didn't want to make completely new sections to the page in case Arcorann wasn't okay with that. The photogames have a decent amount of TASes, even from before I started making them. Chum also has some TASes for StB that may be lost and not uploaded anywhere. I also only just today found out kopiapoa has a handful of DS scoring TASes which is pretty awesome. As for survival-oriented runs, only runs without weird patches were added, but I'm sure there's more that I don't know about.

[Tom]

One of the issues preventing TAS replays from gaining traction is the availability of the Hourglass program which makes these runs even possible.  With the program not being actively developed such runs will become harder and harder to produce going forward.
If someone has any thoughts on the matter or suggestions they're welcome to share them

[KirbyComment]

Since I have been PMd with questions about this, and to answer Tom's comment with how I made my TASes. I used a program called VMWare Workstation 15 Pro with a virtual installation of Windows XP Professional 32-bit (64-bit doesn't seem to work). I believe I have seen either kopiapoa, NNN or both use Hourglass in some cases in Windows 7 (presumably in a VM) but whether this consistently works across games or is specific per game, I don't know. Anyone curious about that and skilled in Japanese would be able to ask them I think. If you have a decently strong computer TASing in a VM should be doable for everyone. You can use the above VMWare program and look up a license code yourself to be able to use it for an unlimited time.

Then to produce the TASes I used Hourglass r81. I have tried r83 as well, which should be optimized for Touhou specifically, but I found it to make save-state usage alot slower and eventually crashing Windows XP as well. This might just be my setup and might not happen to everyone.

With r83 you can save replays while making the TAS, but in r81 when you try to save a replay, it crashes the game in MoF onward. This problem is easily circumvented by just not recording inputs for saving the replay and simply saving the replay manually in real-time after your run has finished. Hourglass r81 does save replays properly in StB at least in my testing. As far as both r81 and r83 are concerned, MoF runs into problems getting it to run and from what I remember SA doesn't work at all.

A few subtle things to keep in mind while producing your TAS: your input recording can desync. The solution for this is to either keep restarting the input recording until it works, or use the splice-feature mid-recording to line it up again. This will require some trial-and-error. It would also not be a bad idea to make copies of your input-recordings. To save on time and worries, also often make Virtual Machine save-states on top of your Hourglass savestates. If anything unexpected happens, or you make a mistake and can't revert to an Hourglass save-state to fix it, then VM-savestates can fix your problem. This may depend on how much effort you plan on putting in your TASes, but the more effort you put into it, the more I advise to do this, personally.

[KirbyComment]

MoF Extra ReimuC 1,005,770,760 by kopiapoa and NNN
(Rerecord count: 8,871 - Replay - Video)

This may be the first Touhou collaboration TAS as of now. This also improves kopiapoa's score of 999,379,340 from late 2013. A big improvement is that they were able to stay below Suwako during her ring spell and squeeze through the masses of rings. Not sure if this is theoretically possible for Marisa.


UM Demo Lunatic Sakuya 6,537,257,840 (testrun) by kopiapoa
(Replay) - Video)

This replay is not going to work because it uses the scroll card, which 100% of the time will desync replays when bombs are used. They for some reason increase the mallet recharge duration. Collett right now got 5,923,855,290 so we can assume the TAS replay is very unoptimized. Collett's replay naturally also desyncs.

Edit: There now is a TAS by kopiapoa that scores 7,322,479,840 and has also posted a SpoilerAL SSG that allows replays to be played back without desyncs. It may take a bit of effort but I've gotten this to work by running SpoilerAL with the latest version of Locale Emulator. Then you run SpoilerAL while Compatibility Mode is set to Windows 7 and then you run SpoilerAL with Locale Emulator in Japanese (as Admin). Not sure if Locale Emulator is strictly needed, but if you are running into trouble, you may have to turn the patch off and back on mid-replay, then select Replay Again from the pause menu.

[KirbyComment]

HRtP Lunatic Jigoku 17,586,190 by Pearl
(Rerecord count: 300-400 - Video)

This is a TAS done using without frame-advance with save states only made between stages. It is basically meant to be a "human theory" TAS rather than truly optimal. The TAS also doesn't do the infinite ball trick to get infinite points, as it would take roughly 36,000 hours to reach 99.9m. Pearl used Anex86 in a Windows XP VM to accomplish this TAS.

[KirbyComment]

UM Extra Sakuya 7,606,658,690 by kopiapoa (testrun)
(Replay - Nico - YouTube)

kopiapoa made an UM TAS pretty quickly after release. Likely the reason for him not doing Lunatic (first) is because counterstop was reached day 1, so it wouldn't really be interesting to make a TAS. Rerecords are unknown. The highest human score is currently 6,142,683,420 by すいれん (suiren) to compare. The TAS replay desyncs because after the boss entry dialogue ends, the rice and mallet card cooldowns go away and for some reason that causes the run to try using the mallet while nothing comes out.

[KirbyComment]

UM Lunatic Sakuya 9,999,999,990 by kopiapoa (testrun)
(Rerecord count: 4,154 - Replay)

Noone would be surprised about this score. However this TAS brings up an interesting question about how score is actually tracked in this game. For seemingly no reason, the TAS replay actually displays the true score in replay menu. This score is 18,014,175,680. However, when you take any other replay done by a human that reaches over 9.99b, the replay file simply shows 9.99b, as far as I know. This even holds true when the run counterstops in Stage 5. That is at least what this tweet seems to suggest. Now when you view the TAS replay in-game, the score counterstops when it reaches 9.99b, however when the Clear Bonus for Stage 5 shows up, it actually starts scrolling up the score partly to 10,995,725,190 (the real Stage 5 score is 12,445,792,950) but then switches back to 9,999,999,990. This score remains as such even after the Stage 6 boss. When you load up this replay to start at Stage 6, it does actually show the true score of 12.4b, but then one second later just switches back to bait you. When you take collett's replay as the humanplay example, the score never changes beyond the counterstop (reached in Stage 6). I'm not sure if this inconsistency is because kopiapoa used hacks to display the true score mid-gameplay. Someone can confirm that if anyone knows.

[Karisa]

This sounds like what'd happen if the replay was created with a counterstop removal. The game tracks the score at the end of each stage, and corrects any desyncs between stages (assuming the stage still properly finishes)... having a patch on during the replay would cause such a "desync".

The same happens with the EoSD 1.3b TAS from years ago, though I think loading that replay from stage 6 freezes the score at the end-of-stage-5 value, rather than reverting to 999,999,999.

[Citizen564374]

Hi, just posted something recently on TAS videos what I consider a novelty. After some reconsideration I thought it's some idea to also post on TH oriented site where more people could be interested or know what to do with it.

PoFV NoMiss/NoHit playthrough with Marisa on Normal, 78:01 in game time with score 1,169,883,280. This is a link to topic on TASvids and last post has a temporary (30 day) link to replay file:
http://tasvideos.org/forum/viewtopic.php?t=22662

Sadly I'm a bit limited to record an actual video of it myself and decided to just share it like this for the time beeing.

One more remark is that apparently the replay can desync on JP version, it doesn't on the last EN patch from gensokyo. And would I just notice after a year, could swear it works either way but there you go. It would be even more important to know for anyone stumbling upon this on TAS forum, more fuss with sharing this than I'd imagine...

[KirbyComment]

IN Lunatic Youmu 9,999,999,990 (10,788,110,780) by kopiapoa (testrun)
(Rerecord count: 106,619 - Replay - Nico - YouTube)

Unexpectedly kopiapoa actually completed a TAS of IN Lunatic! He was working on this in 2015 but abandoned the project. He started it back up again in 2022 and now finished a testrun of it. I remember that Scarlet Team has more scoring potential than Youmu in real time. I don't know if this is true for a TAS environment too. Even still this TAS manages to score even higher than Mirufu's PCB TAS (when score is uncapped). That is pretty impressive given that IN's potential is limited a bit more by there being a score cap on Spell Card Bonuses.

[Karisa]

Youmu has faster unfocused speed than Scarlet Team, which might help reach more bullets in time? TASes of graze-oriented games seem to favor high movement speed.

Though I'm not sure how the speed averages out, when they need to periodically focus to keep the gauge in youkai range.