avatar-optimization

[worn yew]

i feel like atlasing is super inefficient as well? huge waste of texel density

[buoyant holly]

it's fine if you do your Atlas properly

using the cats Atlas tool is not doing it properly

[worn yew]

yeah thats what i meant

[buoyant holly]

like if you're doing a manual Atlas or Texture baking you can get a lot better vram utilisation

[worn yew]

yeah of course

just whenever ive looked at the cats atlas output its a massive texture that wastes most of its space

[buoyant holly]

yes which is why I honestly prefer the cats baking tool over the atlas

[worn yew]

do you both keep outfits on one big material? or do you separate it out

wonder how that works if you have an outfit on a 4k texture but then also have like emission masks and so on

unless those are fine to keep at lower res

[ruby isle]

I don't have entire outfit toggles. If I make a separate outfit then it goes on a new avatar

[worn yew]

i dont mean for toggles

just a single outfit

[ruby isle]

My avatar is 1 single material

[worn yew]

what texture maps are you using

[buoyant holly]

basically everything in one material would be more optimized but it's probably not the end of the world if you use two materials so you can easily swap out the atlas for the atlas parts that are changing each time if you have multiple avatars with different outfits

[ruby isle]

Main, a black and white version used for an effect, and normal map

[buoyant holly]

myself the PBR set for the standard Shader

[worn yew]

might need to do some profiling

though it does feel like a lost cause with this game sometimes

[ruby isle]

Every little bit helps

[oblique patio]

Time is money, cats atlas is super fast

[lethal ginkgo]

Is there a reason LOD groups are not allowed on avatars?

[ruby isle]

There is no limit on uploading, it'll just be Very Poor if you go beyond the max

[buoyant holly]

and everyone on Quest will have to manually turn your avatar on

as there's no setting on quest to Auto Show Very Poor avatars

[tulip rose]

Question before I start working on this project. Do blend shapes get laggy when there are too many of them. I'm planning two avatar OSC projects, but I will need to render them out to a display of sorts, on/off. I'm using blend shapes to maintain quest compatibility. And i think it would be better then using a bunch of mesh renderers.
So for one, I would be animating a quad or two for a tile size of 7x6(x2) and the other would be 10x20. (Not the same avatar). Do you guys thing this would be a bad thing to try and attempt?

[leaden ember]

Added an outline via poiyomi. Looks like this when I zoom in a bit. Any idea where my mistake is?

[buoyant holly]

"except when you're using shapekeys!
There is an exception here! Calculating shape keys can be expensive, especially on avatars with more (>32,000) polygons or when running on lower-spec hardware (Quest). If all your shape keys are on your face, it can be beneficial to split your face mesh from your body mesh, and delete all shape keys from your body mesh. This can be a more advanced technique. Because you have more meshes, it may negatively affect your Avatar Performance Rank, but in reality you'll be slightly more optimized. Remember, the Perf Rank system isn't perfect-- it is just a surface-level recommendation system" https://docs.vrchat.com/docs/avatar-optimizing-tips#-except-when-youre-using-shapekeys

[tulip rose]

Thanks, it is quite a weird problem, cuz like cant use I can't just use a shader because quest lol

[lone tiger]

#development-advanced message

[heady smelt]

is having the face as a seperate mesh better beecause blendfhapes

[oblique patio]

Yes, but from my experience it can lead to weird shading

[ruby isle]

If you set the Anchor Point on the meshes to the same transform (typically the hips/chest) it should fix your shading issues

[heady smelt]

how do i set the anchor point

I probably skimmed over something idk if that makes me sound dumb

[slate crypt]

Anchor point of what

Oh right

Blocked

No help for you

[ruby isle]

In the Skinned Mesh Renderer component, there's an option for it

[echo geyser]

Nevermind, github I found is by shotariya

[exotic meteor]

Excuse me if this is in a wrong channel to ask this--
Always been a little confused and curious by this but especially in recent times with the Avatar Dynamics update. Compared to all the other limits, the hard cap is at 70k, which okay, I can understand. The jump from being just below 70k makes it be Good, but being even just one poly over makes it be Very Poor. What is the overall like, reasoning behind this, seeing as every other statistic for PC Limits is increased like double or sometimes more per quality rating.

Genuinely curious and would love to get some insight on this o-o'

[exotic meteor]

Oh- as compared to my idea of a suggested 32k/70k/100k/140k or something like that, just so it's incremental, like everything else on the list.

[buoyant holly]

I doubt they would be inclined to be increasing the triangle limits considering they went okay no more triangles after 70k

[exotic meteor]

Well in that case shouldn't they have at least made it something along the lines of 32k/44k/56k/70k? Just so it isn't a steep curve from Good to Very Poor?
In my case I lie comfortably right above the 70k mark at 82k, and can optimize without losing any visible quality down to 73k. Everything else on my model is Medium or Good, depending on my setup, but due to the steep curve of the polygon count it's automatically plummeted to Very Poor either way like it never was gonna matter in the first place.
I'd be more understanding if it was at the Poor level only, but from Good and down?

[buoyant holly]

they're unlikely they tweet those numbers anytime soon so no sense worrying about it

[exotic meteor]

Fair

[midnight lodge]

After running my first CATS bake, there is this BakeFix.cs script that needs to be applied on Unity, does anyone know how to use it?

[buoyant holly]

that doesn't require any Unity scripts to use cats

as you would just be exporting a fbx

[midnight lodge]

I meant this little feature from CATS

[ruby isle]

All that script does is set your lighting anchors automatically. It's optional as you can do that yourself or not do it at all

[midnight lodge]

Right! But I don't know how to do it myself or with the script, how would you run it?

[ruby isle]

I've never used it but I assume it might appear in a menu up top

[midnight lodge]

Hmm, I'm not quite sure which menu you could mean

But, if I can do it manually, I will read the script and find whatever it is doing

[ruby isle]

To do it yourself, click on each mesh in the scene and in the Inspector in the Skinned Mesh Renderer component, set the "Anchor Override" to a bone on your avatar (commonly Hips or Chest is good)

[midnight lodge]

Yep! And actually, it seems that on importing the baked folder, it ran itself! So I had to do nothing

Thanks for taking your time to answer :)

[proper grail]

Last time i reseached this 4 years ago it was hinted that unity, in game, splits single skinned meshes until they are each under 70k tris. Very bad. Although the performance ranking adds every skinned mesh as they are one so it gets confusing and is more of a try to fit all cases suggestion system if anything

[ruby isle]

This happened wayyyy back in old Unity. I'm talking pre-2017 Unity where they used proper version numbers old. In modern unity, it doesn't split meshes up at all

And I believe it was ~65k polys

[exotic meteor]

Even with that reasoning, the requirements set to be at Good compared to Very Poor with the split difference being essentially 1 polygon between the two is just such a joke if they're realistically expecting anyone to follow it. Everything else about the limits are fine, and the system is working under the suggestion pretence to begin with because we can just you know; upload anyway.

[median wharf]

Anyone Know how to Atlas while only keeping mostly what's used in the uvmap of the model, like to avoid huge empty space or part of the texture that increase the size of the texture.

This could save so much time sometime

[stray tangle]

use the CATS Bake feature

will re-pack all of your islands into one new UV map and bake the original textures down to one new texture

takes care of your PBR textures too

[median wharf]

Thank you

[supple dirge]

I guess maybe the question is better here, sorry for the duplicate : Stupid question but hum... How many particles is too much? Like on average if they were to all be on screen at once during an anim what would be a reasonable number?

[slate crypt]

Me with 55,000 particles:

Jokes aside that’s the max particle count of my avatar, I’d say around 1,500 are visible at any given time

It depends on what materials they use, if they have collisions, trail renderers, lights, etc

Obviously particles with trails and lights are gonna hurt a lot more

[ruby isle]

Very Poor rank for particles is too much

[slate crypt]

2,500 particles really ain’t much

[floral mica]

why does cats bake just refuse to work

[stray tangle]

can you try using the latest blender LTS version? (2.93.9)

[ruby isle]

It is when there can potentially be that number x40

[floral mica]

that fixed it, tysm 🙏

[stray tangle]

👍

[tropic valley]

Depends on what you are doing.
If it is idle particles, probably a few hundred max.
If it is for some toy or something (like a nose boop or hammer), probably a bit more.
If you are doing full on particle animation type stuff, the sky is pretty much the limit. Although, if you start getting into the multi-millions it starts to get laggy for everyone, and weaker systems hit this limit sooner. Still, depending on what you are going for that could be more or less of an issue.

[lament garden]

Quad remesher is nice but I would also recommend retopoflow as it's free and you learn more about retopology.

[heady smelt]

Hey, I’m trying to make a pc avatar quest compatible using a quest converter tool.. and I’m trying to select the avatar thing however it says nothing is in assets or scene and when I click the eye thing it doesn’t work

[low shale]

Question regarding this, I've been searching for more information and have yet to find much.

I'm aware that shapekeys are very expensive, and that bone toggles are most definitely a better solution a lot of the time, but at which point does it become the preferable solution?

Say there's around 19 different things you want to toggle on one mesh that has approximately 50k~ polys on it. That would require a fairly significant amount of additional bones and set up and I'm curious if it's worth the extra effort.

thank you!

[calm spade]

Shapekey cost is only dependent on how many vertices are affected by it, which is all the vertices on the mesh the shapekey is on, so separating parts of the mesh if they have a lot of shapekeys is usually good practice (which is why a lot of people separate their face mesh from the rest of the avatar).

I'd say it's very situational, but if the avatar is built so that by default all things are on by default, and require a toggle to disable them, it's better than the opposite. Having bones where you can is great, but I know it's sadly not an option for a lot of things as you said.

Imo it's worth the effort, and if everyone did it, it would make a significant difference, but shapekeys aren't the most expensive part of non performant avatars usually, they just compound the issue due to the high poly nature of the models.

I've just recently come across some popular avatar bases which have hundreds of shapekeys to modify every single tiny aspects of the model, and people sadly aren't aware that you aren't supposed to upload the model that way, but instead apply all the shapekeys to the default model you want, and delete all the ones you don't need to reduce filesize !

[low shale]

gotcha, thanks so much for the response! the situation I'm in at the moment is a little tricky, as many parts of the mesh are painted to multiple parts of the armature. for example, the fishnet that goes down the left leg would require about 8 additional bones (because of twist bones / bones for dynamics), and that just covers that one toggle. the same or similar situation is there for the rest of the 18 things I would like to have the ability to toggle.

[ruby isle]

At that point you might benefit using material changes to make it not render, but that wouldn't work on quest

At the same time, you should really create a separate avatar for the toggle instead of making it a realtime toggle

[low shale]

that would require a hell of a lot of additional materials given that this entire mesh is atlas'd to one texture

[ruby isle]

You can still do it using cutout and cutting it from the atlas

That would require a lot more texture files however

Again, I would strongly suggest making it a separate avatar instead of having these toggles on one avatar

[low shale]

they're not toggles for additional options. they're all part of one outfit

[calm spade]

Yeah, in this case I can see why you wouldn't want to have 18 additional bones, so shapekeys make sense. I guess the problem is people having a ton of realtime customization on their avatars, instead of building it static and export that

[ruby isle]

Right, so making a separate avatar for each outfit variation would be the better option really

Most of the toggles on the outfit can be assigned to a bone. It's just the main outfit that would be tough

[low shale]

the main reason to keep it real time is for the ease of the user to quickly make adjustments on the spot. For example, in this particular case, if someone didn't like the bandana on the leg or the straps on the body they would be able to quickly turn them off in game. creating a unique static avi with a desired combination would be.. impossible.. given the number of potential combinations of 18 different toggles

if it was a personal avatar though of course that'd be the obvious choice, but unfortunately it isn't.

maybe one day there will be a more automatic solution to this in the future. some kinda way to make the desired adjustments in unity via something like lyuma's emulator, lock them in, then upload as a static avi would be great. there's a whole lot of variables to consider with something like this so i guess we're stuck here for the moment.

[calm spade]

It's definitely a challenge for people selling avatars. I'm only upset when it's people's personal avatars and they have Blender knowledge

[low shale]

im just trying to do the best i can, someone who makes public avi's has ought to have optimization in mind

[ruby isle]

How many polys in this specific mesh? Is this the 50k one or was it total that you were mentioning?

Oh wait you said one mesh with 50k

[low shale]

a little less than 54.3k

[ruby isle]

Well I would suggest decimating it a bit. You can probably get it down to 25k quite easily

[ruby isle]

Then I guess if you need it to be toggles then use bones for what can be assigned to a bone and blendshapes for anything else

[low shale]

i've already brought the polycount down pretty dramatically, lowering it past the point it's at right now would make some pretty noticeable differences

[ruby isle]

You can retopo this super easy

#avatar-optimization message

[low shale]

already used :)

[ruby isle]

Use more then =)

Like I said, you can most likely get it to 25-30k pretty easy

For example you can collapse those faces on the sword

[low shale]

that's one thing i didn't do yet yea, but that would save maybe like

a hundred polys lol

[buoyant holly]

definitely some major Loops you could dissolve

[ruby isle]

Collapsing quads saves a lot more than you think

[buoyant holly]

like honestly you don't really need every other Edge Loop

[low shale]

every other? that's kinda stretching it, from my testing a lot of the areas with denser polycounts have received noticeable declines in quality when dissolving their edge loops. I could definitely do more, but getting it to the 30k area seems like too much of a sacrifice for the gain

[ruby isle]

If you are really wanting to be optimized, then reducing the polycount is huge

[buoyant holly]

fair so maybe not every other but there's definitely some Loops not doing much

this set of topology doesn't seem to be doing anything useful

[buoyant holly]

do you ever toggle those to where they would be separated

because otherwise they could be the same mesh

[low shale]

they are the same mesh yea

[low shale]

all of this here is

[buoyant holly]

let me rephrase that you could just strip out the underwear all together and project the texture of the underwear onto the fishnet

[low shale]

ah, yea that's not really do-able

toggling off the fishnets is an option so

[buoyant holly]

good to know

[ruby isle]

Also one thing I notice, is there any reason you have culling off?

[buoyant holly]

turning off culling effectively doubles the actual triangle count rendered

so that's a case of trying to cheese the Avatar rank system leading to worst performance

[low shale]

:(

a few reasons, most of which are like this

i could definitely just go and add backsides to these manually, then turn culling off

but then that'd add more polys to the mesh and im unsure how / if culling effects shapekeys or not

[ruby isle]

Culling doesn't affect shapekeys, but it does affect GPU rendering performance a lot

[low shale]

gotcha

[calm spade]

As long as you're keeping optimization in mind when making the models, that already puts you above 99.9% of the models used out there !

[low shale]

the meshes are separate atm just to delete the zero weight vertex groups via CATS to figure out what bones i'd have to put into place to make them bone-toggles rather than shapekey toggles

from my understanding, a new bone per group would have to be created solely for whatever needs to be toggled

surely shapekeys are more efficient than an additional 22 bones that'd have to be scaled for one toggle.... right? 😨 @ruby isle @calm spade

[ruby isle]

My opinion on what you should do is decimate this to 30k or less and use shape keys

[low shale]

is there something special about this 30k number? from what i've experienced / heard, the 64k mark is where stuff starts to get incredibly performance impactful, and i've read to aim for less than 50k on the docs.

i did manage to get it down to less than 47k tho! will be aiming for lower but not sure if i can hit that number

[ruby isle]

30k is the cutoff for me personally but 50k is generally a good point to be under. Your goal is the absolute minimum you can get away with, so there's no magic number for it

[buoyant holly]

so just keep chopping Edge Loops that don't contribute anything visually

very handy for doing the stuff to make sure to look at your model at the average view distance so you're not obsessing over Edge Loops that don't visually benefit unless you stick your nose right up against it

[low shale]

gotcha, thanks guys! really appreciated.

[buoyant holly]

like as a test I'd see if removing every other Edge Loop makes it look significantly worse or not on the leggings

[ruby isle]

'But I need my 10000 polys in this tiny earing people can barely notice on a headset!'

[low shale]

the leggings are particularly dense because of clipping issues, i plan to dissolve some of the bits closer to the ankles/upper thighs since they don't bend but the knees, ankles, and butt area wont be touched

the alternative of preventing clipping would just be making them have more space between the legs but

[buoyant holly]

if you weren't going for toggles I would say it would be more optimized just to have the legging texture be part of the leg

[low shale]

i originally intended for the socks and fishnets to be a texture but pandabear's base has some particularly bad seams for it

[buoyant holly]

then maybe go rebake it with some better seams?

[low shale]

im not sure if re-doing the entire uv, body texture, the socks & fishnet textures, and pbr stuff is worth saving 5k polys 😅

[buoyant holly]

Fair

you're still doing like better than most of the user base

but you might have to go that whole extra mile if you were trying to make like a quest model

[low shale]

https://docs.vrchat.com/docs/quest-content-optimization

seeing as this is a thing

i don't intend to make them quest compatible lol

[ruby isle]

Wait if just your toggles are 50k, isn't your avatar above 70k polys and Very Poor?

[buoyant holly]

that section is a little out date the hard cap would be 20,000

[low shale]

yes, this avatar will be rated very poor. I don't intend to be making the full version of the avi something event ready or rated underneath this area, just something significantly more optimized than a lot of other similar models currently available

[low shale]

and gotcha

[buoyant holly]

that is helpful information to know what your goal is

[low shale]

perhaps i should've mentioned that, my bad. helps to start from a perspective of making it as optimized as it can be in the first place regardless though lol

[calm spade]

I think bones would be better, but then it's a matter of what the VRChat perf settings end up being with more bones.

[low shale]

as a guestimate it'd be easily > +100 bones

so the performance ranking would take a hit, but i'm not too concerned with making it < very poor. they'll be very poor regardless of what i do regarding toggles, so it's more so a matter of actual performance

[ruby isle]

Unless you optimize it to stats under Very Poor

(Which you should do)

[low shale]

right

[buoyant holly]

because everybody is going to assume your avatar is terrible at Very Poor

[low shale]

i don't mind that lol, the ranking is unavoidable with what i plan to add.

[ruby isle]

If I see red at all in the perf rank then I immediately think it's terrible

[buoyant holly]

then you're probably putting too much in one Avatar

[ruby isle]

Everything can be done under Very Poor

There is yet to be a good excuse for VP

[low shale]

a lot on one avatar is what people like

[ruby isle]

And then they wonder why their game runs like crap

[low shale]

yea lol

[buoyant holly]

yes and those people are why we have bad frame rate

[low shale]

i just try to deliver what people want at the best performance i can

[ruby isle]

What people want is not what people need

I'm sure many people would love to have an avatar with literally every possible toggle and feature ever made, but it's not something they need

[low shale]

definitely not, and i agree. but i can't force what people need down their throats

simply put, if i put less on an avatar less people are interested in it

[lime egret]

Because I’m lazy? xD

[ruby isle]

I said good excuse

[lime egret]

Because I’m lazy? 😦

[low shale]

i would love to have a tool in unity that provided checkboxes for what's wanted and not wanted that would automatically generate static versions of any avi i made

buuut

[lime egret]

Hey I atleast made multiple different copies of my avatar with different things rather than cramming it all on the one avatar. I do plan on optimising it and making it quest compatible I just haven’t yet.

[low shale]

i assume almost everyone using this tool would just check every box because why wouldnt they want to have it all available on the fly in game

if they wanted to make a change they'd have to load the project again, generate a new variant, upload... people are lazy

not many people would opt for that i dont think

but i will make < very poor version of the avi with everything stripped from it for those who'd want it

[ruby isle]

But by catering to lazyness, you are making the experience for everyone else worse

[low shale]

by not catering to laziness nobody cares about what i make

[ruby isle]

That's a bit selfish

[low shale]

those who don't want to see avatars that are very poor can hide them, it's not like they're forced to see it.

[ruby isle]

That can often cause social disconnects

You shouldn't be relying on other users to filter the avatar and break their social connection for your own gains

[low shale]

then those who dont want to feel said disconnect can use fallback avatars

[ruby isle]

That also causes social disconnects

Not seeing what a person is wearing no matter what it is will cause a social disconnect

[low shale]

well since i plan to have an optimized version, it could be used as a fallback

it'd be the same avatar, just with less bells and whistles

i dont think the difference between having a hoodie on vs not having a hoodie on, missing particle effects, or dynamics, would cause a social disconnect

[ruby isle]

Just saying you have a lot of opportunity to optimize below VP that you won't take because "by not catering to laziness nobody cares about what I make"

My avatar was at 30k polys and I still decimated it to 20k in favor of performance

No such thing as a limit

[low shale]

i am optimizing where i can, that's why im here. but like i said, it wont be possible to get them beneath very poor.

mainly because of poly count

everything else should be less than that

[buoyant holly]

have you tried dissolving Edge Loop at to see what you can shave off

[ruby isle]

If I can get a 500k poly avatar to 70k, you should be able to get yours there too

[low shale]

dang a 500k avatar must've had a lot of earings

😆

[buoyant holly]

like I suppose question do you really need geometry shoelaces

https://twitter.com/McPhersonsound/status/1532391420081819649?s=20&t=XTwvsoECQ0H_QoIyGCuLEw

Garrett “akalink” McPherson (@McPhersonsound)

This ended up being a dumpster fire of fixing the textures manually in gimp, but I was successful in getting a 200k polygon shoe down to 700 polygons by retopo and diffuse+normal baking. Will never look as good as the original, but it's good enough for me.

Twitter · Jun 2, 2022, 3:59 PM

[low shale]

Yea i could definitely change that

[buoyant holly]

that would require some retopology and texture baking but just crossed my mind

as far as places you could shave lots of triangles off

like what's the current triangle count on the whole Avatar

[low shale]

most of the poly reduction would be through replacing those details with textures

[buoyant holly]

which would work as most people aren't going to be sticking their nose in your avatar that would be rude

[low shale]

atm with all of this, exactly 200,058

i am still not done decimating though, but am running out of places to reduce a ton

[buoyant holly]

I would definitely call that that should be separate avatars for some of those combos

[low shale]

spoiled because it's maybe slightly nsfw? idk. i censored it pretty aggressively lol

Well, the clothing combos would be fully disabled if they were toggled off

not hidden via shapekeys since they're their own meshes which i intend to have dissolve toggles for

[tropic valley]

haa~ can we please stop using such blanket statements as there are always going to be edge cases that cannot work in any other way?

[ruby isle]

I can count the amount of edge statements on one hand

[tropic valley]

maybe for you, but still, that is why you don't use blanket statements. Just say like 99% or 99.9% or something

[ruby isle]

If it is for me it is for everyone too. Everyone else has the tools and possibilities I do too

[tropic valley]

not everyone does the exact same things you do with an avatar.

[ruby isle]

No they don't, but they can. What I do is just what normal game developers do for optimization

[tropic valley]

Uhh..... I am not talking about optimization here, in that sense you are correct. I am talking about what you make on avatars, mostly on the unity side of things as most blender stuff can be optimized under the limits

[ruby isle]

Anything in Unity can be optimized. How do you think actual Unity games don't run at 20fps?

[tropic valley]

uh..... what

how was I spamming wtf

[ruby isle]

Bot borkd

[tropic valley]

...... aparently

well, I'm just going to answer your question and be done, bot got me annoyed xD
Game devs using unity have the ability to control every aspect of the game which is impossible here.

[ruby isle]

Yes and no. We are creating assets for the game, which is the exact same thing an artist does for a game. The artist has no control over the game design systems in place, they just make the content and it gets put in.

While we don't have the ability to control some systems, the ones that count in terms of optimization we do have control over

[tropic valley]

yeah but we also need to be rendering at times up to 80 avatars that are fully simulated either by animations or ik and given that lods don't get made (are they possible actually? or is that stuff not permitted for avatars) the avatar renders the same quality level super far away as it does right up close, and most people (rightfully in my opinion) make avatars with the expectation that others are going to be reasonably close.

Anyways, I agree that doing the best possible on optimizing (within reason) would be great, although I mostly just wanted to remove the blanket statement.

[ruby isle]

I do think LODs should be added to avatars, but regardless, in that case you need to adjust for the medium solution. I can 100% tell you nobody needs over 70k polys. As someone who does avatar optimization for others, I have yet to come across one that could not be brought down

[tropic valley]

duplicate avatar for using action layer emotes (don't say to just use the main avatar, there is literally a bug with root motion on index controllers that this is a possible workaround)

[ruby isle]

Workaround != good

[buoyant holly]

that's a terrible fix

[tropic valley]

https://vrchat.canny.io/avatar-30/p/bugbuild-1004-avatars-30-emotes-on-index-do-not-have-root-transform-applied-prop

that is the bug in question

[tropic valley]

how else do you propose doing it?

also, some people might want to watch that kind of stuff without having to be stuck with the avatar moving differently from them

[ruby isle]

Just wait until the bug is fixed

And yes I know it's been an issue for years

[tropic valley]

It is almost 2 years old...... that is not a viable solution.
Also, I don't really have emotes, so I don't use that. I was just using an example of polygons being over 70k.
Also, vrc is cpu limited in most cases, adding a lot of polygons (assuming not with tons of blendshapes or stuff) is not the worst thing in the world

[ruby isle]

I don't have emotes either so it doesn't affect me. But they're not something that are major issues. You could just not use an animation that moves your root motion instead.

Also, vrc is cpu limited in most cases
Exactly. More polygons = more blendshape calculations = worse CPU usage. This exact stuff can be seen if you go into blender. With an avatar with around 30k polys, if you slide up and down a blendshape it will be super smooth. Do the same thing on a 100k avatar and blender chugs.

[tropic valley]

which is why I precisely specified without blendshapes.....

[ruby isle]

Yeah without blendshapes polys aren't a big deal. But being below 70k is still a good thing for render budgets

99% of avatars have blendshapes as well

[tropic valley]

I can turn my render res up to 200% most times and still be cpu limited, so I personally don't see the benefit in optimizing polycount that much (not counting blendshapes okay, but they should be on their own mesh in the first place). I also know that my system is not everyones, but I do think it is starting to be on the lower side of average ¯_(ツ)_/¯

[tawdry drift]

does removing blendshapes you don't use for your avatar do anything in regards to its optimizing

[tropic valley]

yes, lower vram and less cpu calculations

[tawdry drift]

finally got below 70k

hmm okay

im gutting everything i possibly can

well now i can include a "kick me" sign yay

[ruby isle]

I can too, but that doesn't admit everyone going way overboard with Graphical-intensive stuff. Some people can't render at 200% resolution. I actually do know someone who is GPU as well so they would be at a loss if that was the case

[buoyant holly]

we're still stuck with the ranking system that puts you at the ranking if just one of your stuff has it so even if polygons optimizing is fairy dust we're stuck with it if you don't want people thinking you're have a dumpster fire Avatar

[tropic valley]

I mean, if your gpu can't handle a few k extra polys, you are already having a bad time. I mean, the only person I have heard of being gpu bottlenecked was trying to run vr on a gtx 660, and at that point, there is no reason to optimize for that as literally everyone would need to be extremely decimated to perform better

[ruby isle]

The person I know with a GPU bottleneck has an i9 (I think?) and a 3090. They are definitely not having a bad time. There is always a reason to optimize no matter what conditions are there. You are sharing resources with up to 80 other players. Be mindful

[tropic valley]

uhh..... that person needs to get that gpu rma'd if that is the bottleneck
I know people with 3090 5950x and the bottleneck there is the cpu
With 80 people, the cpu is dieing way more than the gpu cause the gpu handles parallelism better than the cpu

[ruby isle]

They are able to get 60fps in a lobby of 80 players, their GPU is fine

[tropic valley]

then they don't have a gpu bottleneck?

[ruby isle]

They do

[tropic valley]

?????

[ruby isle]

I was literally there when they were checking their stats

Their GPU was at 100%

VRC targets 90fps by default. They have an index though so I think they were targeting 144

[tropic valley]

that doesn't matter though....
If my cpu has a frametime of 100ms, and the gpu has a frametime of 45ms, the usage does not matter. I could have 30% cpu usage and 100% gpu usage, at that point, the one being waited on longer is the cpu which is where the bottleneck is

[ruby isle]

If your GPU is 95-100% you are GPU bottlenecked. Your CPU will never exceed 50-60% in vrc

https://www.reddit.com/r/buildapc/comments/syx95s/comment/hy0eqew/?utm_source=share&utm_medium=web2x&context=3

[tropic valley]

That is not true though......
Utalization is kind of a bad stat to go by, so I am going to use Frame Time
Frame Time is the time it takes to render 1 frame.
This means that CPU Frame Time is the time it takes for the CPU to render 1 frame

In VRChat, the cpu often has a super high CPU Frame Time. This is because a lot of stuff needs to be done on the cpu and is not multi-threaded enough to have full utilization.

So let's compare frame times.
If I take a general prison/murder lobby[this is with everything shown and in the lobby facing everyone], cpu frame time for me is around 50ms or .05s. My GPU frame time is around 25ms or .025s.

This nets you around 20fps.

Now I am just going to pick a hypothetical usage for this example as it would demonstrate better what I am talking about.

Let assume that the GPU is being used at 100% and the cpu is used at 60%.

Conventional knowledge is that you are gpu limited; however that is not actually the case. You are waiting 2 times longer for the cpu to render the frame than you are on the GPU.

So now lets continue, if you decreased the cpu frame time to 25ms.

You would then increase the fps to around 40fps, or even if it did not increase all the way to 40fps, at some point it would take longer to render one frame on the gpu than it would the gpu limiting framerate, not the cpu.

This happens even though the gpu is fully utilized because the time it takes to render a frame is much less than the cpu.
So basically, whichever device has a higher frame time is the one that is the current bottleneck.

You can see these timings if you open the advanced steamvr graph or if you buy the fpsvr utility

[ruby isle]

You are missing one crucial detail:
If the GPU has a smaller frametime, by logic, the GPU will not be near 100% as it is there waiting for the CPU to prepare a frame to render.

[tropic valley]

not necessarily, the gpu can be almost or fully utilized at the lower frame rate, but if you cannot get more frames to the gpu to render, you are still limited by the cpu.

You can actually test this if you are curious. Go into a world (game worlds are great because they are super high on udon and relatively performant for quest) and turn your render resolution up to a super high percent. You will notice that you don't really lose fps until you get to a certain point, at which you start losing fps proportionally as you go higher.

If you have fpsvr or open the steamvr advanced timings graph, you can test this with a lot more information

[ruby isle]

So if the GPU has a 25ms time and the CPU has a 40ms time, what is the GPU doing the other 15ms?

[tropic valley]

waiting

it cannot do anything

[ruby isle]

Exactly

This means it is not working

Now what happens to the load percentage when something is not working?

[tropic valley]

it goes down

[ruby isle]

Right

So by logic, the GPU cannot be at 100% if the CPU is the bottleneck

[tropic valley]

But that is not true, as the cpu starts producing more frames, the gpu will not be able to keep up and will start being the bottleneck, but that does not matter until the gpu cannot produce enough frames to keep up with the cpu

[ruby isle]

Let's do the math slowly here.
It takes 40ms for the CPU to calculate and prepare a frame. It takes 25ms for the GPU to render a received frame.
40ms(CPU) - 25ms(GPU) = 15ms left.
The GPU has 15ms left to sit there and do nothing.
The CPU is still preparing the next frame for the GPU.
A bottleneck is something that stops the workflow of another thing. The CPU is stopping the workflow of the GPU

[proper grail]

I think theres a bit of a misunderstanding, in this case frame time is not being inverted to get an fps value. Its more of a how many instructions am i giving it measurment

[ruby isle]

Moreso how long it takes to perform said instructions

I don't even care about FPS values. It's about which is the bottleneck in this case

[tropic valley]

https://developer.valvesoftware.com/wiki/SteamVR/Frame_Timing

Valve's Documentation on frame timing
read the parts under variables in the struct

[ruby isle]

Yeah this is reprojection

[tropic valley]

yes, and frame timings, so as long as the gpu timing is lower than the cpu timing, the cpu is the one that has the bottleneck

[ruby isle]

Yes you're absolutely right. The CPU is the bottleneck. This is what I was telling you before. The GPU has a lower time, CPU is the bottleneck

[tropic valley]

(╯°□°）╯︵ ┻━┻

[arctic silo]

┬──┬ ノ( ゜-゜ノ) There you go

[tropic valley]

this is literally what I have been saying the entire time

[ruby isle]

So by logic, the GPU cannot be at 100% if the CPU is the bottleneck

The GPU will not be at 100% if the CPU is the bottleneck, is what I'm saying

[tropic valley]

kill me

......

but as long as the gpu can put out frames faster than the cpu, the cpu is the bottleneck

[ruby isle]

I know exactly. But the GPU will not be working 100% if this is the case

[tropic valley]

why couldn't it though?

[ruby isle]

Because it can't be using the entire frame to render something when the CPU can't even send the instructions fast enough

Look at it this way:
A cook makes a pizza in 10 minutes. It takes a customer 5 minutes to eat it. When the customer is done eating, they are there waiting for the next pizza. The cook is the bottleneck. They are working 100% of the time making pizzas. The customer is not spending 100% of their time eating

[tropic valley]

hmm... but that analogy is too grossly simplified for the subject at hand.

Lets think about this in a slightly different way.

Lets force the gpu to do more work per frame by increasing the resolution it is rendering at.

Doing so means it takes longer to render each frame because it has to do more work, and it is already fully utilized so it cannot use inactive parts of the gpu for rendering (because they are all active).

However, as long as you don't increase the resolution too much, the gpu will still be able to render the frames faster than the cpu is producing them, even though the gpu is fully utilized, it just takes more time to do so.

This is similar to the frame timings. As you increase the amount of work that the gpu has to do by rending frames faster on the cpu, you increase the time that the gpu is spending rendering those frames because it is doing more work; however, as long as the amount of work stays low enough, the cpu will still be the bottleneck.

[ruby isle]

What you just described means literally nothing at proving what usage percentage reflects to bottlenecks. When you increase the resolution, you are increasing the work. You're basically tilting the bottleneck more to the GPU side.

I am always CPU bottlenecked in VRC, but obviously if I crank my resolution to 500% I'm gonna be GPU limited right? My GPU will be at 100% usage then. But it will not be 100% if I am not cranking my resolution because natively the game is CPU limited on my hardware

[tropic valley]

Okay, I need to call it here. I need to get some stuff done.
However, I will do some testing/research this weekend when I should have some more time and either admit I was wrong or continue attempting to prove my point. (probably going to take this to dm's tho as it doesn't fit the channel anymore)

[ruby isle]

I would be interested if you find any info proving me wrong. I'd love to learn more about the whole process of rendering frames.

[proper grail]

Very overthinking the meme statement of - Its doing nothing, its not at 100%

[ruby isle]

I've been trying to tweak my render resolutions to hit my GPU harder since I have the extra cycles however it's very hard since the CPU is very variant on it's frametimes

[ruby isle]

Well, what other possible scenario could this be false? What scenario has something not using 100% of its power and still being strained

[tropic valley]

Literally learning now that I need to do some more research xD

[proper grail]

gpu util is a very rough estimate of what power state it is in

[ruby isle]

Knowledge is power xD

[tropic valley]

I was hoping to not bring randomness into this because there is so much variance and randomness in the whole pipeline, especially talking about vrc (well randomness is not the right word necessarily, but close enough)

[leaden ember]

Top part of the teeth/guns won't move with the rest of the mesh. Weight painting does not seem to work (part stays red for some reason)

Here's how it looks in weight painting mode

Any ideas how to fix it?

[high kiln]

paint it to head not hips

also u have #avatar-rigging

[leaden ember]

Thank you!

[slate crypt]

They should be following the head bone thus you need to paint them to the head

[ruby isle]

2 skinned meshes

[frail orbit]

I've made an avatar and I've split the mesh in order to be able to have those parts of the mesh be able to be toggled on and off, however theres now 3 skinned mesh renderers, and this avatar is being developed mainly for the quest (it's telling me 2 skinned mesh renderers is the limit) , so it's now showing as very poor, is there any way I can have the toggles I want without going over the limit for skinned mesh renderers? I've never made any toggles before

[calm spade]

You should have a single mesh and scale bones instead

[wicked plank]

if they are all going to use the same mesh you could try swapping materials instead maybe

[amber hemlock]

A viable solution is getting more people to vote on it so it stays at the top of the trending page on canny. Those are the ones that get fixed.

https://feedback.vrchat.com/avatar-30/p/bugbuild-1004-avatars-30-emotes-on-index-do-not-have-root-transform-applied-prop

I'm so suck of this bug, i've put off animation work for 2 years now :(

i wish they hadden't filed it as an Avatar 3.0 bug, since it's not one.

wonder if that's why it's ignored

[ruby isle]

It's ignored because it's not high priority

[amber hemlock]

I know. Just sucks putting off projects for so long

accurate location of avatars across the network should be a major freaking priority for a vr social platforn

[ruby isle]

I agree but they're stretched pretty thin as is making all this new stuff

[buoyant holly]

crunch compressor textures

[stray tangle]

that is a lot of polys for a quest avatar. you should definitely consider optimizing that

same for amount of meshes

[radiant shadow]

nice ripped avatar (notice file structure)

AuxiliaryFiles/ExportedProject comes from a certain program/location

[rain matrix]

gottem

[vast depot]

Damn caught in 4k

By yourself! how does that even happen man?

[proper grail]

Bruh how you rip free avatars

[buoyant holly]

the model was acquired illegally basically

if you ripped it out of vrchat itself it's illegally acquired

grabbing it from a website that ripped it out vrc

[lost zenith]

Some avatars use bases but often times you have to pay for those bases artists deserve compensation for the assets they create if they ask for it, and if it's ripped it could very well be someone else's personal customization of the avatar

[buoyant holly]

and the file structure your screenshots demonstrates matches a illegally ripped Avatar

[lost zenith]

Its copyright theft

Unfortunately lawmakers don't jump in to protect small artists but that doesn't make it any better to take from artists

[radiant shadow]

well now we know what page you been on

ALL their content is ripped

[buoyant holly]

what website did you grab it

if the Discord was prevent you from posting yes it's a illegal site

[lost zenith]

You could always get the Hobkin model from the artist for literally $10 bucks on their patreon

[buoyant holly]

that doesn't really change my point much

you were trying to mention a bad website

but yeah that answers the question you were using a a illegal model that could get your account ban

well one you would be denying the original Avatar maker Revenue

that's a Hopkin model so yes you have to pay to access it on his patreon

https://www.patreon.com/Hobbert

[hazy sundial]

here from #avatar-help

[heady smelt]

umm 25% empty space sir

UNNN acceptable

[lethal fossil]

Will atlasing my materials break cloth change toggles?

An avatar I bought has options to change clothes and their colors/designs, I'm wondering if atlasing would break that

[heady smelt]

It shouldn't break any toggles pertaining to turning off an on a mesh, but if some toggles are like material swaps then yeah it will 'break' them

it will break them because atlasing will make it all onto 1 material so

material swaps wont work

[lethal fossil]

I assume color/texture changes happen by material swaps usually (sorry still relatively new to dev)

[heady smelt]

its fine

you could upload multiple diff avatars with their own clothing set

if that is an issue

[lethal fossil]

That's what I just thought about recently, would be some compromise

[heady smelt]

I also think there is a way to have it all on one avatar and atlased

[lethal fossil]

I was also thinking about trying to make a fallback out of the 290k poly avatar

[heady smelt]

290k burh

thats gonna be impossible to make a fallback out of without severe comprimise

since fallbacks need to be Good or Excelent rating for QUEST

or well now they have pc fallbacks im not sure exactly how that works

but still

[lethal fossil]

Oh, one more thing, when I optimized the polys on my hoodie the texture got screwed up, how do I re-align it?

[heady smelt]

if you autodecimate it too much it will do that

[lethal fossil]

I don't really care tbh, nvm how jagged he'll be just want something custom 👍🏻

[heady smelt]

lol

[lethal fossil]

I did it manually very slightly but I assume there's less surface on the mesh to the png then. Just don't know how to go about fixing it

[heady smelt]

u dun broke the maths

I don't really know how you would go about fixing that besides a very tedius process adjusting single verticies at a time on the UV map

so I just dont decimate that far to avoid thaaat

but u have to so

unless you want to retopo

[lethal fossil]

Oh well. Thx anyway 🥲

[median wharf]

#avatar-optimization message You can also bake differently with other programe but i don't have tutorial in my hand.

[hazy sundial]

only now got a chance to test it and it doesnt work

I mean yes it does creates an atlas

but I doubt it is good for vrchat in any way

or I'm just dumb af

yeah aint good

guess no fallback avatar then

blender is my worst enemy and believe me I tried it

with cats plugin and stuff

but it cant just properly import/export my model normally

well no

not cats

some other one

but u get me

[worthy bridge]

not sure if it's fine posting here but is it possible to use the quest baking in CATS without losing out heavily on the resolution?

mostly asking since I spent some time yesterday upscaling the textures.
I wonder if removing the unused area could solve this(?)

wait wth I dont remember having all of those duplicate textures in there

ffs it was fine last time when I made the atlas

was perfectly fine when I made the atlas originally now it somehow screwed itself up

[worthy bridge]

why does blender keep causing new headaches ffffffff

[heady smelt]

is this good enough for fallback

read write is enabled

[worthy bridge]

fallback requires excellent/good ranking for fallbacks

[heady smelt]

look into doing it manually

[worthy bridge]

https://docs.vrchat.com/docs/avatar-performance-ranking-system

https://docs.vrchat.com/docs/avatar-fallback-system

[summer jacinth]

Is there any way I can merge audio sources together (i.e. sound packs for variants of the same sound)? I have an avatar that has good performance, but its rating is tanked to very poor because it has 13 audio sources.

I don't think its much of anything to worry about because the most complicated possible combination under the avatar's animations would allow only 3 audio files to be played at once, and that requires a very specific setup (the player must be hit by a certain dynamics collider at the same time that a specific alarm sound is playing all while they are firing the weapon of the avatar). My outlook right now is to just not worry about it fwiw. Most of the SFX are dedicated to effects for specific actions, very short sounds.

Which on that note, why are audio sources so impactful to avatar performance rank? I assume it's a "worst possible case scenario" type outlook (i.e. all sounds playing at once type deal), unless it's related to the spatialization or something like that. It's not very clear what defines that limit in the first place, so I don't know how to work alongside it.

[worthy bridge]

christ finally, for some reason some of the visible areas appeared as complete black/transparent even though they were visible on the traxture map, either way I'm happy that this is finished

for some reason the hands were unscathed which I'm grateful for

[latent sonnet]

It’s not worth it to have multiple variations of a sound on Avatars right now. And yeah audio sources are worst case scenarios for Avatars. Audio processing takes up more computational resources than most people think and in general it’s just better minimising the amount of actively playing audio sources on avatars and in worlds

[worthy bridge]

only reason why I won't put boosting and walking sfx on my avatar 😔

cut down on the excess empty space in gimp but now a majority of the textures are just black even though the map itself looks fine

[worthy bridge]

oh oddammit it was because the alpha in aterial properties has to be changed to match the new material

alternaitvely coudlve just set it to attribute | alpha

I cant believe I spent 2 hours figuring this out

atleast my texture map is 2048x2048 now

I wonder why the alpha can have an image that's completely separate from the material now

didn't need to spend all that extra time rearranigng the uv maps either afjklaa

[worthy bridge]

holy shit finally

[slate marsh]

any way to make a vroid avatar only have 1 material?

trying to make a backup for the quest users but like

vroid is only willing to cut materials down to 2 (face+everything else) which places my avatar in medium instead of good, meaning i cant use it as a backup

this is literally the only thing stopping me

i could fix it in blender if i could find the mesh in unity

which i think i have but instead of an .fbx or any other normal people format, it was like a .asset or whatever

like what am i supposed to do with that

to elaborate:

i have made a combined atlas:

now i just need to edit the face's UV mapping to work with this

which i cant because i cant find the mesh

i mean i can, its here

but what do i even do with a .asset

[slate marsh]

YES

YES

WOOOOOOOO

there is absolutely need to yell my dear friend @hybrid mulch i have made a breakthrough

[buoyant holly]

so did you get it to work

[slate marsh]

i did

just needed some janky unity plugin to convert asset to fbx

and hoped for the best and somehow it worked

[buoyant holly]

cool

[ruby isle]

This is a ripped avatar. I suggest you delete it as it's against ToS

[slate marsh]

no its not, the reason it was a .asset file is because i made it in vroid which saves as a .vrm and i used an unity plugin to make it usable with vrchat, which converted the .vrm to a .asset

but i like did make the avatar in vroid as thats literally its purpose

[ruby isle]

Ah okay then. 9 times out of 10 a .asset indicates a ripped asset

[buoyant holly]

next time would probably be a better idea to do the modifications on the vrm itself

[radiant shadow]

if you need an fbx from a vrm, just import it to blender instead 🔨

[ancient crystal]

this should go on the canny 🥇

[buoyant holly]

good luck with that haven't seen much traction for any modify the Avatar ranking system on canny

[ruby isle]

If you have 2 pens, it should be 4 materials. Not at all equivalent to 500 mats in perf rank

[ancient crystal]

2 pens is 2 materials

but also im talking on quest as well

[ruby isle]

2 for the pen model, 2 for the trail/particle system

[ancient crystal]

u dont need a pen model

[ruby isle]

True

[ancient crystal]

can just put it on the finger

[ruby isle]

So yeah 2 materials

How is it even close to the 500 mat people

[ancient crystal]

even though this is beyond the point of what that message is even saying

reminder that trail renderers and particles are also in their own performance rank categories

[buoyant holly]

because Quest gets grumpy if you have more than two materials

[ancient crystal]

and on quest having more than 1 is already very poor

[ruby isle]

For good reason

[ancient crystal]

and on PC you drop many performance ranks

[ruby isle]

Well yeah. they're heavily scrutinized for a reason

[ancient crystal]

i dont care

[ruby isle]

Well then you also shouldn't care about the rank

[ancient crystal]

also still completely missing the point

that the current performance ranking system does a horrible job of actually computing "performance cost"

[ruby isle]

I think the OP was missing the point. The performance system isn't supposed to be perfect. It's a rough guideline

[ancient crystal]

and treats a barely Very Poor avatar the same as a clearly Very Poor avatar

its not a "rough guideline"

it literally determines whether or not your avatar will be seen

and shows a funny icon in game

[ruby isle]

Remember, the Perf Rank system isn't perfect-- it is just a surface-level recommendation system.

• From the official VRC Documentation at https://docs.vrchat.com/docs/avatar-optimizing-tips#-except-when-youre-using-shapekeys

[ancient crystal]

just because "its not perfect" doesnt mean it couldnt be better

that is such a stupid excuse

say that to SDK2 and we would never have Udon

because clearly we could make games just fine with triggers and animators

[ruby isle]

If you actually understood the impact of the systems it ranks at Very Poor, you'd understand it's not a stupid excuse

[ancient crystal]

it is a stupid excuse because it doesnt actually take into account the difference between a balanced avatar, an unoptimized avatar, and someone who didnt even try

they are all lumped into the same category

at least with the points system you can actually see how bad it is

[ruby isle]

Because it doesn't need to. If you are Very Poor, you did a poor job at optimizing

[ancient crystal]

instead of having just some useless icon

[ancient crystal]

even if you have like 4 meshes but 0 of anything else

or 4 pens

and 0 of anything else

yes very poor

makes sense

[ruby isle]

Don't have 4 pens

4 pens is not optimized

[ancient crystal]

what if you want desktop users to be able to draw too

thats 3 pens

also "just use a parent constraint" but it doesnt work on quest

also also the impact of 4 pens relative to anything else that ruins your performance rank

really cant compare

[ruby isle]

It can compare

[ancient crystal]

but again with a point based system

everything is weighted

and an average is computed, not maximum

[ruby isle]

Do some benchmarking

See the impacts

[ancient crystal]

you do some benchmarking and prove that it makes an equal difference

because it doesnt

[ruby isle]

Wants to change the performance system
Has no proof of what impacts what
Doesn't want to put in the effort to prove it
wut.png

[ancient crystal]

the point still stands that having a point-based approach not only lets people define their own "very poor", but also lets the impact of other things have a much larger scope

for example blocking people with 50 materials rather than people with 7-10

it would be similar to how file size blocking works

except the points come from ranking rather than file size

also its about thinking about actual performance

someone who has everything at Poor is Poor rank

someone who has only one thing that even reaches Poor

is also in the same category

you cant tell me that they have the same performance

it makes 0 sense

the average points for the 2nd avatar would be much lower than the other

it would also help with performance ranking of blend shapes

since blend shape performance is also based on the vertices of the mesh

so you can scale the points of each blend shape by the vertex count

with a more accurate point ranking than just saying "no more than 10 blend shapes"

[buoyant holly]

that would probably get complicated with the quest as like vrchat doesn't want any very poor avatars to exist on it so they probably wouldn't want people to be able to Auto enable avatars in the low very poor Point totals on quest

[ancient crystal]

there will probably just be a limit to how many points you can allow

same difference, but with a more accurate performance measurement

[ruby isle]

Who has everything at poor. Give an example

[ancient crystal]

someone who has all of their stats in the Poor ranking?

just look at the poor rank

[ruby isle]

Vs someone with a single poor setting?

[ancient crystal]

yes

[ruby isle]

Regardless, it's poor in something and should be treated as such

I see so many ignorant people who say their avatar isn't actually a bad very poor if their Poly count is the only thing in VP.

If you have a poor rating, you are contributing to a big loss of FPS regardless of what you have as poor

[calm spade]

Big loss, no, it's a lot more finer than that, but the system can't treat every single parameter that way. It's a conscious choice, and will be improved in the future

[ruby isle]

It's a bit of an exaggeration but it still contributes to FPS loss of some kind

[late quiver]

Thing is its not just that people use a lot of textures, uncompressed and such, its that they also use BIG textures, that eat a lot of VRAM

[ancient crystal]

sure but there is no argument that one Poor case can contribute more loss of performance than an avatar with only Poor case

unless it happens to use something that is immeasurable like unoptimized shaders

[ruby isle]

Again, the performance ranking system is not meant to be perfect. It's meant as a general guide for new users to easily understand

Also, one huge flaw with a points system is that you could have every other stat have little-to-no performance points and then jack up the draw calls to huge amounts and still be poor

[buoyant holly]

sounds like if you were to try that you would want to have draw calls be worth more points than like other stuff?

also would probably kind of be tricky to go okay like x amount of bones are equivalent to a draw called in performance cost to make it that granular

[ruby isle]

Then you'd need to increase the overall points to accommodate. Otherwise the limits for Poor would be as strict as Good

[buoyant holly]

and all of this point stuff seems like it would be a lot more confusing for new creators

[ruby isle]

I agree. It would be confusing for both new creators and players

[buoyant holly]

like how would you know what would be a good point total to do in your safety settings

[ruby isle]

I assume the points would contribute to an overall perf rank like we have now

But there's no way to balance it

[buoyant holly]

then again I probably wouldn't be good for the suggestion as I'm in Camp stop stuffing so much in one Avatar

[ruby isle]

Same here

I'm in Camp Good perf or bust

[buoyant holly]

personally medium rank would be fine but yeah nobody has any business being very poor

[ruby isle]

Medium is fine but me personally I will only settle with Good or better

[ancient crystal]

there will be tiers like good very poor still

or you can set your own tier

plus could make it hide things that go over similar to particle limiter

and the sdk will calculate the points for each category and give overall ranking like normal

and it will probably show the weight of the category as Light or Heavy

to show which ones are most affecting performance

so you know if you have Poor particles and Poor materials, probably the materials will be Heavy from all the draw calls

so you know which one is more worth optimizing

the only real difference is going to be how the overall performance rank is computed

using points rather than a maximum

and it could be more complicated than an average, penalizing someone with all Poor as Very Poor

and someone who is Excellent with one Poor will be Medium or Good

depending on weight

this weight is also based on the points, not ranking, so having something that is way beyond Very Poor will still have a significant impact in the lowering of the overall rank

while a system like this still won’t be 100% accurate like when dealing with toggles for example

but it will make more sense than the current system in terms of performance judgment, and more accurately describe avatars with too much separately from avatars with one or two things going over that don’t necessarily impact the performance equally

plus it might also mean tighter limits can be set on things since now it is considering combined performance rather than worst category

[pallid rain]

I'm sorta new here and i'm thinking about making an avatar in the future. I have certain preferences (like physics on things that should be bouncy) can models with those kind of things be optimized?

because i've never seen a jiggly model be anything but poor

Also, i've never seen flowy skirts or dresses, is that just not possible or is that something only possible on PC models

i'm an quest user btw

[tawdry drift]

only pc because the quest has a hard limit on physbones

[pallid rain]

damn

but understandable

[heady smelt]

flowy skirts and stuff are really hard to do whilst having something that actually looks good

[crisp monolith]

you have no idea how many bones are in a flowy skirt

[ruby isle]

#avatar-optimization message

[rose sequoia]

Skirts can also be done using cloth components. They can look really nice but are usually never going to be performant.
Also you cant use cloth on quest.

[pallid rain]

yeh. i really dont. I'm just an excited noob who wants to make cute stuff

[tropic valley]

lmao, was literally arguing for this same system a week ago in this very chat

[heady smelt]

Skirt physics is my nightmare

[heady smelt]

i just weight everything to the legs

even if its like some huge dress

i like how it looks it has this specific aura of scuffed-ness that i love

[median wharf]

Anyone has retopology plugin recommandation for blender?

[supple dirge]

So when trying to decrease the VRAM usage of my avatar I've been decreasing the size of many of the textures but I also have "texture crunch" enabled. I heard it will uncompress the textures at runtime so does that mean lowering the quality is undone by having this checked?

[buoyant holly]

basically crunch compression only affects download size not how much vram a texture takes up like a 2K textures always going to take up the same amount of vram

[supple dirge]

but it doesn't undo the benefits of lowering the texture quality right?

Like going from 2048 to 1024

[buoyant holly]

yes doing that kind of thing would actually lower vram

[supple dirge]

Nice so is there any reasons not to crunch compress everything?

[languid quartz]

There are rare cases where crunch compression looks really bad, but otherwise it's fine.

[supple dirge]

Alright thanks a lot!

[buoyant holly]

and not sure if it'll be relevant but you can use detail maps to have the fine-grain repeating detail on a texture being tiled so it looks more high-resolution than it is https://docs.unity3d.com/2019.4/Documentation/Manual/StandardShaderMaterialParameterDetail.html

https://twitter.com/VIANVOLAEUS/status/1517951160740687873?s=20&t=SUVdmKmswCBmCJtEqGiJ5Q

𝗩𝗜𝗔𝗡𝗩𝗢𝗟𝗔𝗘𝗨𝗦 (@VIANVOLAEUS)

Back to shader: Can use seamless tiled normal maps per RGB area - tiny (128/256px) normal maps, but tiled around 50-100 times in the shader, giving you a higher effective TD.

You can combine these with an base nrm map for creases and baked normal details etc: (Blend with Base)

▶ Play video

Twitter · Apr 23, 2022, 7:38 PM

[twilit fossil]

Can someone tell me please why when i atlas my textures that it uses another texture that isnt even asigned to the model instead of the one i want..

[agile plover]

How do i fix this? Should I use CATS for it? Thanks in advance

[glass ibex]

yes for easier method to do it.

[agile plover]

I assume I have to use the atlas feature?

[dawn hull]

https://youtu.be/PyRmBKBCq9k?t=815

Take a look at this tutorial

[agile plover]

Thanks, will do

[south crest]

ive got a weird one. my model is 9.2k faces in blender. and when i upload in windows SDK< i get an excellent rating. when i go to publish in android it claims i have almost double that!

am i missing a setting?

oh, and in windows mode im also getting a 17886 polygon value. just noticed that

[tawdry drift]

triangle is poly

[south crest]

o

im blind. thank you

[tawdry venture]

Yeah, Unity uses triangles for rendering. Quads get triangulated.

[past sable]

hi, im having rly big problems cutting my model down to size

its at a strong 100k faces , and i just cannot find a way to re-topo it without spending like 20 years doing it manually

[buoyant holly]

have you been dissolving Edge loops

[heady smelt]

ok and

guys protip don't put something with 9 million poly into unity my screen just went black and everything DIED

killed my dog too

actually i think the file corrupted and that was why i died

is there a way i can force blender to commit suicide when i do something really stupid so i don't have to restart my pc for the 4th time this hour

[stray tangle]

why the hell do you have 9 million tris in the first place

[heady smelt]

its funny

that is the only reason i can give you

[ruby isle]

100% unfunny

[heady smelt]

rong

[tawdry drift]

don't feed into it lol

[heady smelt]

riley do you actually contribute to the server or do you just complain about me

whenever i bother to expand your messages its just complaints about me

[ruby isle]

Well with your about me bio being a first impression, I don't blame them

[heady smelt]

how do you even believe yourself in that being a genuine argument

youre the 684th now

[ruby isle]

The same way you believe its funny is a good reason to have 9 million polygons

[heady smelt]

stop whining lmao

it's honestly just confusing

[ruby isle]

Then why question it

[heady smelt]

because i only came to that realization just now

[ruby isle]

Yikes

[glass ibex]

[past sable]

im trying it right now but im not seeing much improvement... im also really new to blender, could you maybe show me how its done? internet tutorials always use basic shapes

[buoyant holly]

https://www.youtube.com/watch?v=8MkgXcidxg4

[worldly hull]

how to lower the size?

[heady smelt]

how did you get so high in the first place with something so small everywhere else

[worldly hull]

already fixed, never mind.

[thin steeple]

Trying to use CATS to merge my meshes but when I do, This happens... any ideas how to stop this from happening?

I need to merge the tail, ears and hair onto the body

CTRL+J on the meshes makes them loose their weights...

[calm spade]

Apply your transforms first with Ctrl + A

[slender blade]

im so close but this is what shows and idk how to fix

[silent bone]

I need help

with fixing this

[calm spade]

Disable other avatars in your hierarchy

[onyx harness]

Scroll up so we can see the errors

[heady smelt]

???

[heady smelt]

is it good enough for fallback

[autumn drift]

Is there a way to have a bunch of physbones in my hair and the avatar be good, it says I need 4 to be good

I want 14

So

Ya

[late quiver]

Best probably not to use cats for that. Control-J should join your meshes correctly, if not, make sure that all your transforms are correct, you can apply transforms by pression A then Control+A and clicking all transforms, then try joining the meshes.

[tulip rose]

Combine scripts together, unless you need different values or the bones would have to cross humanoid bones (ie, hair and tail), it's always best to combine them down.
Hair example: right click the head, new empty game object, name it, drag and drop all the hair into that empty, add a physbone script to that empty/pointing at that empty. Copy-paste values from one of the hair scripts. Remove redundant scripts.

[thin steeple]

We found the issue, the hair wasn't weight painted nor had bones, or vertex groups, we got it all working <3

[late quiver]

Yay!

Im glad you guys found a solution!

[heady smelt]

does anyone know how to remove physbone components on hidden models from the one model? i have an avatar with mutiple models in it for the toggles and it has too many physbones for quest, but it wont let me delete the physbones for it

[thin steeple]

In your Hierarchy, search Script and it will give you a list of alll the scripts in your Hierarchy. Including hidden ones.

[daring summit]

Texture atlasing

[worldly hull]

atlasing? - never heared

[buoyant holly]

https://en.wikipedia.org/wiki/Texture_atlas

[heady smelt]

Uh. is avatar quality in polygons means triangles? or polygons? because polygons can be n-gons

and i can have 8 triangles in an n-gon

[sick timber]

It’s in triangles, as unity triangulates everything

[heady smelt]

I see. Alright.

what's the texture resolution?

is 4k allowed?

looks like it's about filesize, so light crunch for 4k i think

good that means i can stich 10 materials i use down to 3

[ruby isle]

4K is allowed but not recommended. 4K textures will fill VRAM super fast

[heady smelt]

😩

[sick timber]

And in quest builds for avatars, you get 10mb to use

[heady smelt]

What about crunch compression?

it unpacks to vram all the same? @sick timber

[sick timber]

Yes

[heady smelt]

🤔

[sick timber]

Crunch compression lowers storage

It’s still decompressed when needed

[glass ibex]

Uses the compressed texture, but file size stays the same. so low quality in same image size.

[lone tiger]

Crunch compression lowers the size of avatar asset only by a few kilobytes at most compared to using Normal quality compression while also reducing texture quality even further usually, so I don't recommend using it on avatars.

Also Crunch compression usually causes load-in freeze of the avatar if you use it on more higher-res textures.

[mellow hare]

is the highest polycount or tris 10k for optimized?

[heady smelt]

sigh welp. guess i'm doing 90s texture atlassing then. there is no other way around it. https://i.imgur.com/EcgZQtr.png

[heady smelt]

oh....nevermind. this is misery

[heady smelt]

actually...nevermind. some edgesliding and recutting started doing the trick

[mellow hare]

is the highest polycount or tris 10k for optimized?

[heady smelt]

for props i think? because overall says it's 32k https://docs.vrchat.com/docs/avatar-performance-ranking-system

that's for excellent rating

[tawdry venture]

10k is the max for an avatar to be fallback-compatible.

[heady smelt]

Ew.

where are the lod levels?

[oblique patio]

@mellow hare it depends on your usecase. For a good PC avatar it's 70k. That's what most people see and aim for. Excellent rating is 32k on PC. Then there is the quest platform with 7.5k/10k. Fallback avatars have to be lower then 70k for the PC Version and Quest Fallback lower then 10k

[sullen junco]

Hey, on the topic of optimisation. I need to know how to properly atlas textures in blender without them breaking down, can anyone help

[oblique patio]

You can use cats, @sullen junco https://m.youtube.com/watch?v=0gu0kEj2xwA

Or atlas manually, but it takes longer and is a pain https://m.youtube.com/watch?v=0cSesu_zkQQ

[mellow hare]

Oh I made it 7.5k-

9.5**

so |I can make the pc 70k?

for it to be excellent

[oblique patio]

https://docs.vrchat.com/docs/avatar-performance-ranking-system here is the source 70k for good, 32k for excellent @mellow hare

[heady smelt]

Ok so it is working but fuck is it taking forever.

[mellow hare]

How do I fix this?

and I dont have 15

[calm spade]

Since you have multiple meshes, those materials are counted as unique on each meshes

[static mortar]

is there a tutorial for taking a bunch of individual physbones and grouping them together so that it can be uploaded for quest?

[stray tangle]

@static mortar you need to create an extra bone and have all of the bone chains parented to it. especially useful for things like hair

that way you can just add a single physbone component to the root bone, and all of the hair will move

[heady smelt]

Jezuz this was annoying https://cdn.discordapp.com/attachments/360718065626775555/988152674476261386/unknown.png

[heady smelt]

as for triangles....ech

down to 41,790

is that good enough, if i only have one material?

[ruby isle]

For PC that's not too bad. For Quest, it needs to be 20k or less

[heady smelt]

How the hell do i even do that on this avatar.

[ruby isle]

You can probably dissolve a lot of those edge loops

[heady smelt]

i can't

they carrying the panelwork

i already dissolved every loop which wasn't

[ruby isle]

That's just a texture though

[heady smelt]

Can you...scroll up a bit?

that's not a texture, that's a 90s texture atlas work

[ruby isle]

What does 90s texture atlas even mean?

[heady smelt]

That means using distinct texture blocks on a texture atlas and putting polygons around them

like total annihilation did

modern texture atlas means just baking uv islands together

[ruby isle]

Um texture atlases haven't changed ever

There is no modern vs. 90s atlasing

[heady smelt]

yes there is

the methods are in reverse to each other

[ruby isle]

So then why is there a single wiki page for texture atlas and not one for "90s" type

[heady smelt]

there is no wiki page for fucking tensorfield cityscape generation and yet it's here and working

wikis are written by laymen. for juicy stuff you have to dig for yourself in papers

[ruby isle]

There is a page for tensor fields

[heady smelt]

generalistic knowledge, no specific methodology

[ruby isle]

Okay, can you provide an example of the difference to a 90s atlas and a modern one?

[heady smelt]

Let me sum up again.
90s texture atlas - you build a texture atlas first and then uv map around it. carefully
modern texture atlas - you do uvs first and then let software stich the atlas for you automatically

these are diametrically opposite approaches

[ruby isle]

They're both the same atlas. The way you make it doesn't change the outcome

CATS auto generates an atlas for you. Is that modern or 90s?

[heady smelt]

This is not just about outcome its about methodology + outcome.

[ruby isle]

Then how does that even matter in the end

What difference will that make with your avatar

[heady smelt]

In one term i described both methodology and the result. you are the one who's having to think that he has to be right in the end. stop it. get some help.

[ruby isle]

YIkes

You know, blender automatically fills in dissolved faces

So regardless of your UV methodology, it won't affect it

[heady smelt]

just shows how little you understand. all these uv patches are separate polygon pieces, 4-8 polygons in strips or less

if i dissolve any of the carrying edges - it will autoglue loose edges and go ACROSS the atlas

[ruby isle]

Why would you make a UV like that. You're then creating so much more work for yourself

[heady smelt]

Because this way i have 3401 texel resolution on a single 2048x material

as opposed to having 10 1024x materials

it looks highres but uses poultry amount of vram

[ruby isle]

And then you spend 95% of your time trying to get around the limits you've imposed on yourself

Diminishing returns

[heady smelt]

Not me

Vrchat

[ruby isle]

VRChat doesn't force you to split UV polys

[heady smelt]

had vrchat had more civilized tools - like autolod, like secondlife has - we would've not had this discussion and i would've not spent 2 days making this atlassing

[ruby isle]

LOD isn't even going to help performance in most VRC cases

VRC is a CPU hog. GPU very rarely gets worked

Also SL and VRC have almost nothing in common

[heady smelt]

they are both metaverses. they both allow user content. tho secondlife is much better in that regard because its monitezation system actually works and works well. first release in 2003- it's still going.

[ruby isle]

Yeah because that's what everyone wants, monetization in their social games

[heady smelt]

Yes. Because social aspect means vanity aspect. meaning you can make money off of it

I know several people which make living off secondlife by making and selling clothing for various avatars. and people buy it just fine

[ruby isle]

Yeah don't worry about the social part of it, just consider it a business.

Why make meaningful friendships and social connections when you can pad your pockets!

[heady smelt]

#general-media

[void grove]

how do I fix this

[oblique patio]

Hey @void grove Do you want to upload an avatar for quest? The sdk tells you to change your shaders to VRChat/Mobile/Toon Lit. You have to find all your materials in the "hierarchy" window and change the shader on the right side on the "inspector" window.

[void grove]

I figured it out I just cut out all the materials in blender

[worldly sapphire]

Any thing I can do to make my performance go from very poor to medium for quest?

[tawdry drift]

remove excess bones, atlas your textures

[worldly sapphire]

Which excess bones are you talking about, and how do I atlas my textures?

Here are my errors

Found a video on the texture one

[tawdry drift]

the physbones are whats making it very poor but you can just reexport from vroid because theres an option to atlas

any bone that doesn't have any weight is excess

[worldly sapphire]

So let me check on weight paint mode

[tawdry drift]

males don't need breast bones

[worldly sapphire]

Ok

Did that

[worldly sapphire]

It's outdated

[tawdry drift]

atlasing doesn't change

but you can use cats for the quick fix or reexport your vroid character from vroid

i can't see the full stats so im guessing its just the physbones making it poor/very poor

[worldly sapphire]

I know I can't delete the hair bones

[worldly sapphire]

How do I do that?

It still says very poor

[worldly sapphire]

😡

I fracked it up

It keeps fracking it up

It was at "The Power of 2", the lowest number, and 0.

No ones gonna help me

[ruby isle]

Dude how do you have 26 physbone components with short hair

[worldly sapphire]

IDK

[ruby isle]

Don't put a physbone component on every hair bone

[heady smelt]

Is it better to use shapekeys to control props/clothing toggles, rather than have a mesh for each one?
The Documentation simply states that shapekeys can be laggy and it's best to use different meshes to split those, but using several meshes is also bad. So since both are bad, and the SDK/documentation warns just the same for both methods, which is better to use?
Just make a mesh for each prop/clothing, or use shapekeys to control those?

[stray tangle]

if you're trying to hide a bunch of props, use bones and scale those up and down

or use Poiyomi's UDIM Discard feature

that way, they can all be on the same mesh, and you don't need to use shape keys

[heady smelt]

Shaders won't help with Quest compatability, so it's a pass on that. But are bones better than meshes/shapekeys, because the SDK complains when you get 100 of those too, and your avi ranking immedietly goes to VeryPoor

[stray tangle]

bones are the most performant options

the SDK has no way to check for shapekeys

which is a double edged sword

[heady smelt]

Therefore the best way to sneak things in, eh
Then know of any tutorial to do bone transformations? Because every time I've tried that, nothing works. I'm not certain how to set it up in ActionLayer+FX layer to make it work, and can't really find any good tutorial on it

[ruby isle]

Transform animations/bone toggles need to be done in the Gesture layer.

[heady smelt]

Well that explains it. The documentation says "Action layer is for bone animations that will override all other layers, when you need to take over total control of the character."
But I guess it's not as worded, bone animations/bone transformations

[ruby isle]

What they mean by that is entire control, like a full AFK animation or sitting animations. Gesture is for transform animations that would be smaller parts such as toggles and finger movements

[heady smelt]

Got it, thank you two for your help/

[heady smelt]

So, new question. Really torn about the worth of optimizing.
Most avis that can be purchased, most tutorials, just recommends creating meshes to add stuff like clothes, toggles. If I do optimize, to use one mesh and bone transformations to control toggles, others won't easily understand that avi because all the tutorials suggests doing it the unoptimized way (i.e. like purchasable avis).

So really, how badly can an avi with, say, two skinned meshes and 15 mesh renderes really lag others? Because that is maybe an average number of what most avis out there, with clothing and prop toggles have. Blender CATS plugin and the VRchat SDK warns that this is horribly unoptimized, but is it really that bad?

My Quest2, both standalone and linked (GTX 1080) can handle a typical Murder4 lobby, filled with verypoor avis with playable frames and lag. Even if I do optimize my avis, will that one splash in an ocean of other unoptimized avis even make a difference? ...Maybe there's an optimization chart, that shows how unoptimized avis impact others or?

[oblique patio]

I think it's totally up to you what avatar you're using. I don't think the devs are forcing you to use a green avatar? Most people optimize their personal model and don't sell it to other people.. nobody other then yourself needs to know how it works. I believe the average avatar has a lot more components then you're saying and is not optimized at all. It's not bad for us rtx 30xx guys, but it's definitally a problem for people that have a vr headset and own a pc that fits the min requirements. Steam has a very interesting survey about what hardware the average gamer is using a gtx 1060... I think content creators should keep that in mind. Idk if you show a lot of avatars on quest, or what is "playable" for you, but my frames are really bad with 10-20 very poor avatars....

[languid quartz]

For meshes, if you don't show all of them at a time and use them as toggles for different outfits it's not that bad for performance.

However, thinking of it as how badly can an avatar alone can cause performance issues is kind of asking the wrong question. Most avatars even the worst ones just all by themselves don't appear to affect performance all that much when you test them.

The real issue comes from when you have to share system resources with multiple different avatars, if everyone's avatar in a lobby has terrible optimization than you won't be playing a game you'll be watching a PowerPoint presentation, and just because your PC can run that just fine doesn't mean everyone else's can do the same.

[leaden birch]

I need help with quest optimization

[worldly hull]

Would it make a difference, if I use a 8bit png to save space? to get under the 10MB upload limitation?

I mean, my texture would bein 8bit only 844kb instead of the 2,1MB with 24bit

[solid smelt]

Changing any of the texture settings will show up in editor, if you changed it to 8bit, and the visual material didn't change but the file size lowered then you should be fine

[heady smelt]

So is something like this considered optimized?
*1 body mesh, using a skinned mesh renderer. Default number of shapekeys for face. All props/items get a single bone at a hand, bone transformation used to hide/show. Later in Unity, Gesture Layer could move these to the other hand via anchorpoints to dual-wield.
*1 clothing mesh, using a skinned mesh renderer. Several outfits made as different meshes and shaped to body before being merged. This mesh using shapekeys to hide/reveal different outfits. 10~ shapekeys would be used.
*Total polygon count 50-100k.

How much higher could the figures be above, before this is no longer considered optimized?
The VRCDocumentation state that any mesh above 34k polygons using shapekeys should be made into its own mesh. But if I were to do that, I'd instead be making more and more meshes, which is also bad. I'd only have used bones if I could, but Unity doesn't support Blender's object contraints.

[onyx harness]

These outfits, are they all using 1 material?
Are the shapekeys being used to Hide or Show outfits? / How many would be >0 at any given time?

[heady smelt]

Yes. But both the body mesh and clothing mesh would have their own atlas material, for sake of different shaders. About 5 shapekeys would be on at any given time in the clothing mesh, and maybe the rest 5/10 would be at 0. Those 5 shapekeys on, would be maybe outfit1's shoes, dress, hat, so on.

[onyx harness]

Sorry for the message delay
The materials sound good, a lot better than the usual in fact.
I would try to limit the use of blendshapes as much as possible. They do like to hog the fps as more are enabled. For clothing pieces such as a hat you could have it weighted to its own bone and scale said bone between 0.01 and 1 when toggled off and on respectively. To continue that pattern it gets a bit harder with things that are weighted to multiple bones but is equally possible. The last avatar that I worked on had ~40 toggles all that used bone scaling for clothing and accessories.
Here, the weighting for the shoes is moved to specific "Shoes" bones of which the toggle is using the method described above.

Else I would check over the Vram usage of the model. It's not checked by the sdk so it's not shown on the Performance Rating but if you're aiming for optimised then I'd take a peek at it. Thry's Vram Calculator is a great unity tool to get a breakdown on what's inflating the size of your model on people's machines.

[heady smelt]

Thank you for the post, and yes I would make bones instead for solid stuff like shoes/hats. But you don't happen to know if this could be done for something larger, like a dress covering several bones? I added a post about that in #avatar-rigging, and how I can't find a way to make a bone scale toggle for a mesh that's rigged to several bones. Incase you wanted to take a look, and try to answer

[onyx harness]

In the example above the Shoes are weighted to the knee and ankle bones. Using a bit more complex example I can show this (image). It's an Armour piece which is weighted to 6 different bones and thus has 6 bones each parented to the necessary respective Armature bone. This is all set up in blender as 1 fbx the same way. I am targeting these bones with an animation and animating their scale values directly.

[heady smelt]

Yes, I had this idea before. It seems to be the only way.
But wonder this, which is my question in rigging too:
If the armor piece was weight painted to the armature like any other outfit, but instead an extra bone is weight painted to all of the armor piece. For example, a control bone, attached to the spine.

The problem is that the control bone, will move when the spine does. Which will move the mesh. When only the armature should be able to move the mesh. In Blender, Object Constraints can be used to force this bone, to ignore the armature's movement/rotation except for scaling. Therefore, the armature retains full control of the mesh. The control bone is only used to scale it.

Object contraints doesn't work in Unity. So the control bone will move along the armature, which means the mesh will be all over the place. Do you know of any way to bypass this? Or is the method in the screenshot the only way, it seems to be.

[onyx harness]

No, I haven't found any way in Unity to tell a mesh to ignore the position/rotation of a bone. It'd be nice, would massively cut down on bone count but what I demonstrated above is the only method I've come across so far for bone scaling toggles.

[heady smelt]

Oh well, maybe in a future update for sake of optimization.
Thank you for answering all the questions, even the one in Rigging. Will edit it to mark it answered, and link here.

[onyx harness]

I should bring it up, above shapekey or bone scale for toggling, there are shaders built with inventory toggling systems in mind and iirc are better than really anything else. I haven't personally used the shader before, which is why I didn't mention it sooner, so it's all hearsay and reading documentation but it has been spoken quite highly by those that are more knowledgeable than I.
I am just unsure if it'll have the same options that you require for your clothing material.
https://gitlab.com/s-ilent/SCSS/-/wikis/Manual/Inventory-System

[heady smelt]

This wouldn't be Quest compatible, though? Unless the shaders could just be changed to one of the VRchat/mobile ones?

[onyx harness]

ah, if it's for Quest then no, this shader wouldn't work.

[heady smelt]

Too bad. Quest hardware can run shaders just fine. Evidenced by legacy avis using legacy shaders still available out there. Maybe one day, VRchat would make a shader such as this but for Quest. I'd still take a look at it, see if it's worth setting up for a PC avi and using bones for the Quest. Altough, at some point, optimization might go to far due to the amount of work. Thank you/

[buoyant holly]

https://twitter.com/Silent0264/status/1365127506924249094?s=20&t=cl-K5QLnaS81VPulQ1usCw

Silent (@Silent0264)

Tip!

Though VRC doesn't count or penalize them like bones, shape keys are expensive. The cost scales with poly and shape count.

If your model is 50,000+ polygons, splitting face and body and removing all keys from body can improve performance for those around you.

#vrchat

Likes

113

Twitter · Feb 26, 2021, 2:32 AM

[ruby isle]

Yeah Quest can run shaders, but try getting a lobby of 40+ avatars with transparency and crazy shader effects. Your quest would detonate

[radiant shadow]

nope

chest/head are for interaction from others/self , fingers collide with hair/other people physbones

[normal nova]

I added a thing to CATS for that

[buoyant holly]

very nice

[calm spade]

I didn't have much success with it

[radiant shadow]

fbx size get tiny bit smaller if you seperate the mesh that has blendshapes too

[normal nova]

The splitty thing? hmmmm, what went wrong?

[calm spade]

That kind of stuff

[normal nova]

Well that's correct... it's supposed to fix normals in place using custom normals though

have you tried it recently? I made some updates a little while ago

[calm spade]

I don't recall using custom normals, but that was a year ago, have not tried since because those results scared me enough

[quartz geode]

I have a basic question if I have a avatar with both a quest(medium performance) and pc version, when I use it will quest users see me as that avatar or my fall back avatar by default?

[oblique patio]

@quartz geode u need to upload for pc then switch to the android platform and use the same Blueprint ID

[sharp galleon]

They will see the avatar by default unless their safety settings are set otherwise. If the avatar is ranked as poor or higher then it will be replaced by your fallback unless they show avatar

[mossy shore]

hey, so im trying to atlas a texture with cats and i have material combiner installed, but every time i try to use it it says install pillow, so i hit the button, and it says to restart blender. i close blender, reopen it, and the cycle repeats. what am i doing wrong? i know theres something about running as admin but im on a mac

[heady smelt]

I think you have to install it on your own. I didn't have to at first, but I had to the last time I did it. You should find it on github.

[crisp monolith]

Besides decimating polygons and sizing down textures, how else can I reduce avi size?
removing bones?

[buoyant holly]

bones aren't going to have a meaningful effect on your file size

[heady smelt]

Remove any large animations. Some can be mb in size.
Crunch compress the textures, unless this is what you meant by sizing.
Atlas texturing should also decrease the size.
Blendshapes/shapekeys also make the FBX larger, and removing those in Blender will decrease the size.
What you also can try, is the VRCQuestTools. It promises to lower the size to 10mb, but will likely do it via texture alteration.

[normal nova]

Good news for uh, both of the people using CATS Bake and height maps, height maps are now bakeable x3 (once I push these latest changes)

so you get real, stereoscopic surface detail on your model, but still just stuck into the one material

[twilit fossil]

I'm trying to atlas my textures but i get this error? any ideas?

[heady smelt]

how do I combine skinned meshes

[tulip rose]

Generally, the best way is through blender. Selecting two meshes and ctrl J to join them, (select the avatar base first so it doesn't mess with the origin point)
Using the cats plugin can also auto do that for you. If you want something to be toggleable or move on its own, usually recommended to just keep them unconvinced
Optional, while you're in blender it's recommended to atlas all your textures together as well.

[twilit fossil]

no one?

[quaint venture]

If your textures are not pngs try changing them to pngs

[thin steeple]

Do empty blendshapes still count towards avatar size and performance?

For example: if you made the body of the model.. then made clothes off of it's geometry or what ever, and all of it's blendshapes got copied to the clothes but don't do anything (or at least shouldn't)

[calm spade]

Yes, blendshapes have an inherent cost, which is based on the amount of vertices that are on the mesh that those blendshapes are on

[calm spade]

So it increases its filesize, but doesn't cause performance as long as those blendshapes aren't active

[thin steeple]

Even if they're not assigned to any vertices or vertex groups? 🤔

Example: You create a new blendshape but don't do anything with it.

[carmine plank]

Just wanted to show my compressed UV map, just that i can use low res textures for high detailed result

[calm spade]

By being on a mesh, the vertices are assigned automatically to a shapekey

That's why people are often told to separate their face and body mesh, because that way the shapekeys only affect a lot less vertices

[rose sequoia]

Does "active" mean "are set to a value greater than 0"? Or something else?

[calm spade]

Yes, values greater than 0

[heady smelt]

How do I fix my avatar when I do atlas because it doesn't send me to file folder

[dawn hull]

Try this BEFORE doing the atlas: Files, External Data, Unpack All into Files, then choose first option

[heady smelt]

That did it thank you so much!

[hasty sedge]

So yeah. This happens when trying to use the atlas in cats. Everything is greyed out, except for the jacket, X, and tie, and even then the jacket and tie lost a lot of their color. Is there a way I can fix this?

[hasty sedge]

Alright I somewhat fixed the issue. Saw people saying to go into the file menu and do the whole Unpack all Into Files thing, and that worked for a second. But now whenever I do it, it just brings up this error saying that a shit ton of files failed to create. How can I fix this? It worked before, and now it just decided it doesn't want to.

[stray tangle]

do the files already exist?

[hasty sedge]

Hell if I know. How/Where would I check?

[robust lintel]

My avatar has a lot of bones on the tail. I want to have physbones on them, but then there would be too many. Can I skip half of them?

[hazy tree]

Any idea why my materials/colors got darkened? First New to the image map for less materials, but the first time I did it, it worked. Even the icons are dark, but the tail tip is fine

[onyx harness]

Because blender and unity are different programs that don’t share shader information past basic textures. You’ll need to use a shader in unity that supports the lighting you want.

[hazy tree]

any recommendations?

[onyx harness]

Which platform are you building to?

[hazy tree]

PC to fix the version I already uploaded, but I was to try to put him in VR right after

[onyx harness]

Yeah so vr is both pc and android so that’s not the best term to use

[hazy tree]

sorry, Quest

[onyx harness]

Okay, well the shader a lot of people default to on pc is Poiyomi Toon. Though it you want it to look the same on pc and quest you could try and set it up using the VrChat/Mobile shaders

[hazy tree]

Will it look noticeably different on PC with Poiyomi compared to Quest, or mostly the same? Also, it looks like most people use it for things like shiny metal and rippling water. How do I just get the colors to look like Blender?? never used one and the tutorials I'm finding are over my idiot head.

[hazy tree]

also, stupid question but I gotta make sure - after I upload the avatar and everything's working, I can just go ahead and get rid of everything on my side? The avatar will be stored on their end?

[tawdry drift]

they wont look the same but mobile shaders can still look fine if you mess around with it enough, but once you upload your project you cant obtain the files from the vrchat servers but you can use it inside of the game

if you need to edit something in the future you should keep your files

[hazy tree]

Yeah I'll keep a backup, but I'm planning on getting rid of the programs at the very least. Give a good look over to my model, make sure I'm happy, then uninstall everything and hopefully never have to fix it lol

[buoyant holly]

the unity and blender lighting aren't going to match up as their lights are probably in different directions with different intensity values

[hazy tree]

Just in the default test home world, I actually can't tell any difference between the non-shader 20 material model and the Poiyomi single material updated one, so I guess the question is irrelevant now, thank you to all who answered though!

only thing left to do is figure out how to upload a VR version and fix the fluff on the shoulder from disconnecting when I move

[wanton quartz]

hmm he looks familiar, i'd try looking for a folder called textures where your .blend is saved and messing around with that or, forget the atlas and manually merge all your materials.

[hasty sedge]

Damn, didn't think anyone would ever recognize him ngl
But anyways, I managed to get past the issue of the textures graying out after atlasing, but now hes gray when I import him to unity, and I cant export textures. Any clue on how to fix that one?

[wanton quartz]

lmao i'm actually trying to catch up with the comic rn

what i ended up doing was switching to external materials in import settings and then assigning the textures to the materials albedos', but that option says "legacy" so i'm not sure how long that'll work for

[hasty sedge]

Alright Ill try it. Where can I even find import settings? Im a little dumb I cant find anything

[wanton quartz]

ah when you click on the model in assets it'll be on the right under inspector

[hearty coyote]

What's up guys, I'm not sure whether this channel is the one appropriate for this question, but I've been trying to create excellent performance rank avatars and I have a couple questions about specific features I've seen on excellent performance rank models

[ruby isle]

Sure

[hearty coyote]

Alright, I've seen color wheels with hue and saturation in a quest model, taking a single slot material

[ruby isle]

The only way to do that is by using material swaps (which won't transition between colors smoothly) or by using emission maps or something and changing that color

[hearty coyote]

I'm guessing they used a diffuse and a emission map, but I've no idea how they set the parameters

[ruby isle]

The parameter is just a float.

And it transitions between colors as any wheel would. There's plenty of tutorials online for color wheels

[hearty coyote]

I couldn't find one tho

[buoyant holly]

https://www.youtube.com/watch?v=EkVlkrQ6ypE

[ruby isle]

Hmm first link

Might wanna brush up on your googling skills as well 😉

[hearty coyote]

I suck at researching, I know that

Imma check those out, thank u guys

[ruby isle]

Google skills are the #1 most important skill to have if you use a computer

[hearty coyote]

oh, I've seen this, that's my problem, vrchat shaders don't have hue shift, but I've seen that on quest models

that's why I'm here

[radiant shadow]

you can material swap using 1 animation then get similar effect (also aweful) , ive never used a blendtree for hue stuff (get the same effect with 1 animation 0 <> 100 on hue and float)

old avatar exsample

[hearty coyote]

yeah, but I saw it taking a single material slot, maybe adding a hue shift slide to the shader's code, I've no idea what the code lines would be tho

[buoyant holly]

Quest avatars are unable to use custom shaders

[hearty coyote]

alright, I don't think Imma get further in here, thank u guys

[lone tiger]

Swapping materials via animations doesn't increase material slot count.

[pseudo obsidian]

Yeah, that's something that even I didn't realize going in. I'd imagine having all those packed in will get to file size eventually though, if you have a ton that use large assets.

[dusty otter]

when i use the decimation modifier on my model and then export using catts the export doesnt have any of the decimation what am i doing wrong on export

1 in blender and other in unity but its the same fbx i exported

[onyx harness]

apply the modifier

[dusty otter]

?? example

[onyx harness]

apply the decimate modifier before you export

[dusty otter]

i do not see a apply button

nvm i found it

how else can i go about reducing size of the avatar ? as just decimating doesnt lower me enough to 10mb

i went from 22mb to 17mb

[slate crypt]

Delete unused blendshapes

Decimate in general is a pretty bad idea if you want your avatar to look good afterwards, but it will work if you don’t care about topology

[calm wave]

help

[ruby isle]

There is no helping that

[stray tangle]

that can't possibly be the only thing bad about that avatar

sounds like a very poor PC avatar