Server Icon
KQM | Genshin Impact
catalyst-n1-buffer
284 messages
Avatar
Ticket ToolTue, April 12, 2022 at 09:30 PM
@steno119 - As an author, it is your responsibility to complete the ticket

Guidelines
- Name your ticket with $rename <ticket name>
- When you are ready to submit the ticket, compile everything into one message following the format below. Then type $close or click the button; the ticket will automatically be moved to be reviewed.
- The ticket will be scrapped if: No activity >1 week or open for >1 month.
Write-up Format
Theory/Finding/Bug: Title of your submission

Evidence: Explanations with calculations and/or Youtube/Imgur proofs

Significance: Conclusion
Avatar
steno119Tue, April 12, 2022 at 09:30 PM
its time to open this shit up again
:pepeW:
continuing this ticket but changing the scope slightly
figured now would be a good time, since sim needs more detailed frame comparisons
will be adding more to this
in the coming days
Avatar
Delectable | N2J EnjoyerWed, April 13, 2022 at 08:01 PM
:QiqiPeek:
QiqiPeek7
Avatar
steno119Sat, April 16, 2022 at 01:46 PM
4 days later no update :ICANT:
Currently in the process of getting a new computer
Avatar
paichlearSat, April 16, 2022 at 06:00 PM
does no one read this
unknown.png
there better be something when i come back next week
monkaS6
Avatar
steno119Sat, April 16, 2022 at 11:15 PM
Well uhhhh
There is all the work I did in the previous ticket
:jerry:
I guess I should post it all to get started
Sorry pai
Finding: For catalyst characters, buffering a normal attack out of certain actions will lead to the n1 having significantly increased animation frames. So far, dash buffer n1, walk cancel buffer n1, and full normal attack string buffer n1 have been tested

Evidence: a couple buffering scenarios with Kokomi
https://www.youtube.com/watch?v=66CQ3c1dwzw

Significance: better optimization of catalyst attack strings.
Catalyst buffering bug
I'm too lazy to retype this so here have the write up copypasted from the ticket in KQM
https://discord.com/channels/763583452762734592/923418129219092520/925111350915657849

For some catalyst users (e.g. Barbara), buffering a normal attack in a state when they cannot attack will sometimes lead to a very delayed n1. This happens during dashes, v...
needs to be tested:
- does jump buffer n1 lead to the delay
- how many frames does each buffer scenario increase the length of n1 by
- how does the n1 buffer inconsistency for tall females affect the above
Avatar
Magnus Artifex | Chiori C1 how?Mon, April 18, 2022 at 01:31 PM
Wait so if I’m reading this right N1 cancelling on catalysts suck?
Avatar
steno119Mon, April 18, 2022 at 05:31 PM
not quite
trying to mash n1 out of other actions
is bad
like dash -> n1 is slow
Avatar
Magnus Artifex | Chiori C1 how?Mon, April 18, 2022 at 08:57 PM
Gotchu
Avatar
daibangdenTue, April 19, 2022 at 09:30 AM
can we extend this to all weapon types and not just catalyst?
i know it'll be a lot to cover but this is a long ticket anyway
can we? :kokomiShy:
Avatar
steno119Tue, April 19, 2022 at 04:23 PM
do any other weapon types have this issue?
:Hmmge:
we can check claymores and bows ig
Avatar
daibangdenWed, April 20, 2022 at 02:30 AM
i've recorded some videos with bow characters, will make a comparison video later
although idk how to setup nohboard to show my input though
i'm not using obs
nvm i just checked for all bow users and seems like there is no delay
Avatar
steno119Wed, April 20, 2022 at 12:28 PM
Thought so, would've been very obvious to Yoimiya users if it did happen
I'm getting that you can't buffer any action out of jump
:starege:
or wait maybe you can hold on
I need a frame comparison
Avatar
washedFri, April 22, 2022 at 04:15 PM
don't think you can buffer out of jump
except for plunge attack
oh yeah and swap
Avatar
steno119Fri, April 22, 2022 at 05:32 PM
yeah, I wasn't able to
which answers the first question
how do we measure the length of an attack
like if I wanna know how much longer a delayed n1 is compared to a regular n1
should I look for a keyframe during n2
or some recovery frame during n1
Avatar
washedFri, April 22, 2022 at 06:18 PM
:keqhmm: does looking at what frame the N1 appears not work
or maybe measure the length of N1W
Avatar
steno119Fri, April 22, 2022 at 06:19 PM
wanted to include recovery frames
yeah n1w works
tho uh
is there a keyframe for walk
I guess you can just eyeball it for the most part
Avatar
washedFri, April 22, 2022 at 06:20 PM
the frame the weapon disappears i think
Avatar
steno119Fri, April 22, 2022 at 06:20 PM
right
that's a thing
yeah that works
Avatar
washedFri, April 22, 2022 at 06:20 PM
though yae and ning don't use their weapon i think?
Avatar
steno119Fri, April 22, 2022 at 06:20 PM
:pepeW:
forget Ning
idc
Avatar
washedFri, April 22, 2022 at 06:20 PM
:kleek:
Avatar
steno119Fri, April 22, 2022 at 06:20 PM
she has the weird left hand animations I don't wanna delve into
can generalize using Lisa
:peepoIgnore:
Avatar
paichlearFri, April 22, 2022 at 06:48 PM
stare at the animation for an hour then you'll start to notice, between the animation of each attack in an attack string there's like this one frame where the character just moves differently, most of the times they move like 3 or 4 times more than in the middle of an animation if that makes sense
wait do you mean length as in distance or time
if it's time just send the clip and i'll try, need to get a hang of catalyst animations for #frames anyway :kleek:
Avatar
steno119Fri, April 22, 2022 at 06:54 PM
well
I don't have a clip yet
was trying to figure out how I should go about recording to find it
but I'll get you one in a bit
Avatar
paichlearFri, April 22, 2022 at 06:55 PM
which character?
as for recording instructions there's this now, still a wip but it's almost done https://docs.google.com/document/d/1rhKDnkgkVYU5jLq6i_7FRz43DRwElyALX_j6hW5DwqY/edit?usp=sharing
KQM Frame Count Standards
This is a WIP, leave feedback in the #frames channel Glossary NA: Normal Attack; CA: Charged Attack; E: Elemental Skill; Q: Elemental Burst. Macro: software with automated input sequence that imitates keystrokes or mouse actions. Attack string: a series of attacks. e.g. Traveler’s NA string is N1...
Avatar
steno119Fri, April 22, 2022 at 07:03 PM
needs to be done for any tall female catalyst (Lisa pls, don't use Ning :keqingPain: ) and any non-tall female catalyst
Avatar
paichlearFri, April 22, 2022 at 07:04 PM
ah i can just record it myself then, do lisa and klee work?
Avatar
steno119Fri, April 22, 2022 at 07:10 PM
yeah
Avatar
paichlearFri, April 22, 2022 at 07:15 PM
!rm 4h catalyst na frames
Avatar
(!) Keqing's Eternal AssistantFri, April 22, 2022 at 07:15 PM
Alright pai, I'll remind you about catalyst na frames in 4 hours. ID: 39089434
Avatar
paichlearFri, April 22, 2022 at 10:14 PM
i have yet to figure out how to use a macro so these are not frame counts, they're just animation references

lisa:
27 - start of n1
78 - start of n2
125 - start of n3
175 - start of n4
232 - start of n1

klee:
14 - start of n1
52 - start of n2
92 - start of n3
179 - start of n1
like if you open the videos, go to those frames, then move between them and 1 frame before them it should give you a good idea of how it looks like
although i'm not 100% sure whether or not it's accurate, it's kind of just what i've been using and what makes sense. hm canceling each animation separately might be a better idea
yea so if you look at this, it's obvious which frame the dash animation starts on, and if you look at 1 frame before that, it's identical to 1 frame before the the frames i sent in the other message
idk if i'm making sense but after each attack the character will kind of stay in place for quite a few frames only moving slightly, then they will suddenly move a lot, that's the frame the next animation starts on
putting the 2 videos side by side might make it clearer, the video on the left is playing normally and i cropped the one on the right to match the left
n4 is the most obvious start for lisa, the rest are less obvious but you can still tell when the animation starts just by going through those few frames a few times
Avatar
steno119Sat, April 23, 2022 at 12:02 AM
yeah makes sense
Avatar
daibangdenFri, April 29, 2022 at 10:26 AM
i think i found somethings :hmmge:
character attack animations will come out faster if they are in idle stand compare to when they are in the middle of other animations
you can see the sudden change in the movement of Donut when Kokomi enters idle stand, if she attack before that her attack will come out slower
idk if you have inclued this in the finding or not, but i thought it would be good to share :hmmge:
works on other characters as well
so the reason why spamming left click make character attack slower is because you are doing n1 during their previous n3 animation
after doing full attack string it actually you put in a stand that equal to idle stand which allows you to do an n1 without having animation duration increased
the reason why it does that idk
:kokomiShy:
doesn't work for Sucrose i think
for the full attack string
above
so your n1 is never be delayed but n2 is
Avatar
steno119Fri, April 29, 2022 at 02:23 PM
weird
Avatar
washedTue, May 3, 2022 at 07:42 PM
this may or may not be related, but i've had an issue of the game eating my skill and burst inputs for seemingly no reason

i decided to investigate further, and it looks like if you press skill on the first frame of walk animation, the game ignores your input. i wonder if this is related to walk cancel delay N1?
steno can you try recreating the catalyst walk cancel delay N1, but by starting with W first, instead of doing N2 and then W
so just
W
N1 right after
Avatar
steno119Thu, May 5, 2022 at 12:56 AM
yeah sure
sorry I just saw this now :kleekicant:
:mike:
why did I say yes
:peepoLeave:
here it is with just n1
@washed
the amount of delay
seems to be different
Avatar
washedThu, May 5, 2022 at 01:37 AM
wait i can't tell
is it slower with the W
Avatar
steno119Thu, May 5, 2022 at 04:09 AM
yeah, its not always consistent but yes
W -> n1
leads to slow n1
JeanHands1
Avatar
washedThu, May 5, 2022 at 04:14 AM
:derpfei:
Avatar
LeahSat, May 7, 2022 at 12:43 PM
:smadge:
does that mean i can't kokomi walk cancel anymore
Avatar
daibangdenSat, May 7, 2022 at 01:11 PM
um no :hmmok:
walk cancel is still walk cancel
if you input both w and n1 at the same time it leads to slow n1 which is why you don't do that nor hold w all the time
Avatar
paichlearSat, May 7, 2022 at 01:30 PM
wait, what about holding walk the whole time during n1, does that still make the animation longer?
Avatar
LeahSat, May 7, 2022 at 01:31 PM
im actually not sure
Avatar
paichlearSat, May 7, 2022 at 01:31 PM
because doing that with some ca/e/q makes the animation shorter
Avatar
LeahSat, May 7, 2022 at 01:31 PM
not n1 related but i sometimes find that kokomi n2 can have a delay before coming out if you input too quickly
not syre if this ties into the current TCL finding about the n2 delay though
Avatar
paichlearSat, May 7, 2022 at 01:33 PM
can you leave that in #frames? along with evidence if possible
Avatar
daibangdenSat, May 7, 2022 at 01:42 PM
well ye, if you input too fast, like with any claymore user, you wont get 2 attacks from 2 mouse clicks, you have to wait, it's more of timing that input delay
Avatar
paichlearSat, May 7, 2022 at 01:51 PM
i know that if you input too fast the game will ignore it, but i've never seen inputting too fast delaying an attack
Avatar
steno119Sat, May 7, 2022 at 02:54 PM
this is actually the point of this ticket
but yes you can consistently avoid it
the n2 coming out slowly is just as a result of n1 animation being longer
which happens when you buffer n1 out of stuff
or w
Avatar
LeahSat, May 7, 2022 at 02:55 PM
:worry:
Avatar
LeahSat, May 7, 2022 at 02:56 PM
:keqhmm:
yeah that's what i meant
Avatar
steno119Sat, May 7, 2022 at 02:56 PM
I didn't show walk cancels in that vid since its still being researched but
basically you can avoid it consistently
as long as you fully release w and wait like 3 or 4 frames
before you input another n2
Avatar
LeahSat, May 7, 2022 at 02:57 PM
waiting :kekwait:
this is why i will never play kokomi with beidou
Avatar
steno119Sat, May 7, 2022 at 02:57 PM
its not actually slow you just need to time it
:KEKhands:
Avatar
LeahSat, May 7, 2022 at 02:58 PM
i do really quick walk cancels on kokomi
Avatar
LeahSat, May 7, 2022 at 02:58 PM
:worry:
Avatar
steno119Sat, May 7, 2022 at 02:58 PM
this is what it looks like, you can't see the delay at all
macro
Avatar
LeahSat, May 7, 2022 at 02:58 PM
i do it manually
no wonder
Avatar
steno119Sat, May 7, 2022 at 02:58 PM
yeah, I do too, can't get 11 tho
my highest was 10.5
but at least we know 11 is possible
Avatar
LeahSat, May 7, 2022 at 02:59 PM
11N2W here
:coool:
Avatar
steno119Sat, May 7, 2022 at 03:00 PM
op
:KEKhands:
Avatar
LeahSat, May 7, 2022 at 03:00 PM
what do you guys use to count frames
Avatar
paichlearSat, May 7, 2022 at 04:09 PM
oh that makes more sense
Avatar
paichlearSat, May 7, 2022 at 04:09 PM
adobe premiere, but any video editor works
Avatar
LeahSat, May 7, 2022 at 04:09 PM
:keqhmm:
ive got ticket burnout after submitting something that took 5 months to crack so give me another 8 years
:keqWheeze:
Avatar
steno119Sat, May 7, 2022 at 04:43 PM
I use Da Vinci Resolve, but like
even VLC Media Player works
:RaidenWheeze:
as long as you know the right hotkeys
Avatar
daibangdenSat, May 7, 2022 at 05:00 PM
my 10n2w + n1 :kokomiShy:
Avatar
daibangdenSat, May 7, 2022 at 05:02 PM
also how about this one? the n1 animation becomes longer eventhough i didn't buffer it out of any action other than itself? :hmmge:
Avatar
steno119Sat, May 7, 2022 at 09:33 PM
no idea
:pepeW:
Avatar
paichlearSat, May 14, 2022 at 08:20 PM
@steno119 hello this ticket still exists
i see a writeup in pins, what else are you trying to do?
Avatar
steno119Sun, May 15, 2022 at 12:41 PM
Ah fuck
Yeah there's a couple more things I wanna do to figure out when it happens and exactly how many frames it slows n1 by
Avatar
paichlearSun, May 15, 2022 at 02:13 PM
every action -> n1 for every character?
what are you using to determine the duration of n1, release frame, hitmark, start of n2, or something else?
Avatar
steno119Sun, May 15, 2022 at 02:52 PM
Release frame is very character dependent, hitmark doesn't apply for ranged chars, so I was planning on looking at start of n2
Avatar
paichlearSun, May 15, 2022 at 03:12 PM
start of n2 is just as character dependent as release frame is though, and afaik most catalysts do have hitmarks when their na/ca land on enemies or objects, they just don't have hitlag
anyway so to test it, n2 without buffer and n2 with buffer to see how many frames later the n2 starts right?
Avatar
anju06Sun, May 15, 2022 at 03:31 PM
but then Sucrose doesnt have n1 hit mark if you spam left click :keqhmm:
how about start of n1 to start of n2? for catalyst users just look at the sparkle effect on their hands or something
Avatar
paichlearSun, May 15, 2022 at 04:48 PM
well this is data from walk/dash/jump press -> 17ms left click spam https://youtu.be/CedXQkQ96IY
Kokomi N1 Buffer
00:00 No buffer
00:17 Walk
00:34 Dash
00:52 Jump
unknown.png
n2 start - n1 start is consistent enough (14-17f), idk what happened with the 24 though. details here https://docs.google.com/spreadsheets/d/1fKfOkf06GCpzRTwp439chyqEdYf6SA-rg9_G8Lmha_c/edit?usp=sharing
i think when exactly it's buffered matters, might need to macro the whole thing and not just the left click spam
something like this, but it'd need to be tested and adjusted for every action
unknown.png
the delay between walk and n1 click can be 0 but the delay between n1 and n2 can't go below 60ms (it's inconsistent at 60, that's why it's 70 in the screenshot)
actually from this test (3 trials for everything) n1 basically got delayed only once, i'll go through the footage again later to see what's different with that one
hm the only difference i'm seeing is that n1 is buffered earlier in this video, if the left click input is 10 frames after the dash input then the animation doesn't get extended, the 2nd and 3rd trial from my video are both 9 frames after but in this video it's 4 frames after, that makes sense https://www.youtube.com/watch?v=66CQ3c1dwzw
Catalyst buffering bug
I'm too lazy to retype this so here have the write up copypasted from the ticket in KQM
https://discord.com/channels/763583452762734592/923418129219092520/925111350915657849

For some catalyst users (e.g. Barbara), buffering a normal attack in a state when they cannot attack will sometimes lead to a very delayed n1. This happens during dashes, v...
the n2 didn't happen though, which means that the second left click input was too early, which also makes sense
without another action before the n1 you can use 70ms delay between n1 and n2 clicks but with dash -> n1 it needs to be longer, but 70ms works with walk
how long idk yet, i increased it by ~20ms at a time and even at 300ms n2 isn't coming out
wait it works at 500ms :artesians:
!rm 5h test this again but with timed clicks instead of spam clicks
Avatar
(!) Keqing's Eternal AssistantSun, May 15, 2022 at 07:56 PM
Alright pai, I'll remind you about test this again but with timed clicks instead of spam clicks in 5 hours. ID: 40768571
Avatar
paichlearMon, May 16, 2022 at 02:22 AM
@steno119 my computer decided to lag when recording this but it turns out kokomi's n1 duration is only affected by dash, these are entirely macro'ed with the shortest consistent delays. it's in the same sheet as above so take a look and lmk if you think the methodology is ok. btw the gray numbers are just the same as the on-screen frame counter and the black numbers are the delta from the start of n1 to whatever the column is https://youtu.be/pF9OV2fJ_uA
unknown.png
if it works i can test this for all the catalyst users except yae (which is the only one i don't have)
Avatar
paichlearMon, May 16, 2022 at 02:31 AM
apparently as long as you're not standing completely still (with the weapon fully appearing on the character's back, it can't be still fading in with the yellow thing) n1 will be longer sometimes, requiring n2 to be input a few frames later
just noticed that while i was testing with 70ms delay between n1 and n2 input, it's not exactly canceling walk with n1 that makes n1 longer, it's "waiting for walk to end" that makes n1 longer
tl;dr standing completely still or holding a directional key = shorter animations. anything else could make the animations longer
Avatar
steno119Tue, May 17, 2022 at 04:06 PM
okay sorry for responding late I will take a look later today
Avatar
paichlearWed, May 18, 2022 at 02:12 AM
so walk for less time?
Avatar
steno119Wed, May 18, 2022 at 02:13 AM
yes
the walk input and the attack input need to be almost instant
something like 2 - 3 frames between the walk input and the left click
also wow dash buffer is only 5 frames longer? that seems a little :keqhmm:
Avatar
paichlearWed, May 18, 2022 at 02:16 AM
2-3 frames between the start of walk input and left click?
i thought only releasing w mattered :ZhongliConcern:
btw what's the delay between button down and button up for
Avatar
paichlearWed, May 18, 2022 at 02:23 AM
it's 9 frames longer if you look at start of n1 to start of n2
Avatar
steno119Wed, May 18, 2022 at 02:23 AM
still way less than I was expecting
Avatar
steno119Wed, May 18, 2022 at 02:24 AM
just to make sure my inputs are actually visible to OBS Input overlay
if I just send raw clicks
it doesn't even show them
for your purposes you won't need it
Avatar
paichlearWed, May 18, 2022 at 02:25 AM
hm why does my overlay show it
azure's doesn't show either if there's no delay between down and up
so the duration of the walk affects the duration of n1 animation?
or the position of the character when n1 is buffered
Avatar
steno119Wed, May 18, 2022 at 02:57 AM
this
my guess is that
walk startup (like the first 2 frames) are considered inactionable
which means if you try and n1 out of it
it considers it as buffering n1
Avatar
paichlearFri, May 20, 2022 at 12:21 AM
since kokomi's n1 isn't affected by walk but is affected by dash i tested both, turns out how long you hold walk or dash for doesn't change how much n1 is delayed by
unknown.png
n1 start -> n2 start is the same here when buffering n1 while dashing
unknown.png
15 frames when not buffered and 24 frames when buffered :keqhmm:
here's n1 being buffered during walk, dash, and jump for every catalyst user (except yae, i need someone else to record her or lend me an account that has her) https://docs.google.com/spreadsheets/d/1fKfOkf06GCpzRTwp439chyqEdYf6SA-rg9_G8Lmha_c/edit?usp=sharing

everything is tested with both n1 and n2 as early as possible using macros, ofc you can delay n1 more by adding more delay between n1 and n2 input but that's already known and not specific to catalysts (kokomi's n1 gets extended by 1 frame per second of delay, didn't frame count other characters)

the testing tab has some random things that prove the button hold duration doesn't affect the n1 animation duration and that the delay has a range that also doesn't affect it

tl;dr out of these 3 actions, only dash affects the duration of n1 (extends it by 9-10 frames on average, with ningguang being the exception of 5 frames)
@steno119 is there anything else you want to test? i couldn't get walk to extend n1 duration, if you have a video of that lmk and i'll try again
Avatar
steno119Sat, May 28, 2022 at 08:15 PM
@paichlear here
Avatar
paichlearSat, May 28, 2022 at 08:44 PM
:ZhongliConcern:
unknown.png
do you mind if i upload this to yt btw
ok so apparently neutral position/standing/idle animation -> n2 (whether or not walk was input doesn't matter, it depends on whether or not the walk animation actually happened) is the same as walk hold (any duration) -> n2, which for kokomi n2 start is 15 frames after n1 start on average
unknown.png
as for delayed n1 in your video, it's also 24 frames, same as dash buffer. the theory that each character only has 2 n1 durations is still true
i think it might be that 1f delay between w up and m1 down
because in none of my tests i added a delay between w up and m1 down, that's why i never got a delayed n1 from buffering it during walk
only tried different delays between w down and w up
nvm it's not the time between w up and m1 down, and apparently 1f delay = 4f delay on the input overlay
unknown.png
Avatar
paichlearSat, May 28, 2022 at 09:17 PM
do you remember the macro for this or how exactly you pressed w and m1? if not can you try to replicate it but write down the macro this time
i tried adding a delay between w down and w up, w up and m1 down, and both. can't get extended n1 either way