I’m fairly new to UE5, but it is widely known apparently that Metahumans are performance heavy for many people, even on high LODs. This is mainly due to, when using higher LODs, while polygons get reduced as (LODs normally do), the bones don’t seem to disappear for real.
It’s really weird that even after checking the MetaHumans on run with “show bones” in console, and clearly seeing the difference on the skeletons (I put multiple BP copies of the metahumans with different LODSync settings), the fps were down to 20fps most of the time, even at LOD7 (about 15 LOD7 MHs)!!
So, basically…I ended up exporting the MH Mesh to blender, manually removed some of the bones only shown at lowest LODs, almost every 12IPV bones, and then re-imported to UE.
The result is fantastic, reducing from 700+ bones to ~260, it runs at smooth 60 fps with 12 MHs at LOD0 (on a NVIDIA1660Ti).
The problem is…why are the bones not getting automatically removed as it is supossed to? I have read the following thread:
Same issue, no solution.
I want to report this bug but maybe it’s been already solved or has to with with some config I forgot to tweak or something.
Hey I notice when, inside the metahuman blueprint, I click on Face, then on the right under Skeletal Mesh, I untick “Allow Anim Curve Evaluation” the fps goes way up. I’m not sure what this does tho.
Metahumans are trash. Or well, cinematic only trash. This should be known by all attempting to use them.
Skeletal meshes should be limited to 255 bones like they used to be in ue3.
They raised that to 2 byte (65535) total in ue4 and made this trash possible.
(Bone influecnce is still 4 to 1(vert), so having 65535 bones is only detrimental really).
the computation cost per bone per skeletal mesh increases exponentially with every bone you have to calculate.
The performance differential between a normal amount of bones (around 18 for a human) and something with 700 bones is not just around 700-18times more expensive, but closer to 1800 times more expensive.
LOD bone reduction works OK in ue4 IF you manaully select the new skeleton for the reduction / add the bones to the removal list (good luck with 700+).
Metahumans should use face morph ser up for ARKit out the door.
The computation cost of morphs is high, but less than keeping the face bones avaliable.
Thus, anyone should be able to remove the face bones (leave the head bone) and use the morph targets to animate expressions.
LOD reduction on the skeletal mesh should prevent / remove the morph targets.
Having 52 morphs per mesh is expensive as hell when you have 100 mesh on screen. Even if they don’t animate they have a large/useless memory footprint.
Animation bluprint for LODs should probably be different / not contain morph alterations.
ABP can be swapped at runtime. I suggest writing a bit of code around the LOD in use and swapping ABP based on the LOD.
Less variables = less memory footprint = less computation = smoother framerate/performance. Sure its a chore, but you end up doing it even without using metahumans.
always report all bugs.
The epic team goes out of their way to ignore your reports anyway.
In the rare event that someone responds to your report (id say a 1 in 1000 chance currently), they usually just tell you they cannot replicate, Or, that the engine is perfect and working as intended even when its not.
RARELY, they used to scramble and actually find a workaround 4+ years ago. Don’t expect the same now. In fact, expect to be told to suk it…
metahumans should have a dedicated team which wasn’t Epic at least originally. Potentially they’d be more receptive to any bug report.
Keep in mind, epic will ignore their bug reports even more than they ignore yours
To have an easier time - if licensing permits, which it should but you need to check for yourself - export the metahuman out.
Mod a mannequin skeleton bone positions to fit the mesh.
Parent/assign the skeleton to the mesh.
Update the vertex group names to match the bones if they need it.
Export to fbx.
Import in engine.
Retarget the ABP of the mannequin to the new skeleton and test it (Assuming you didnt accidentally mess with bone rotations when you placed the skeleton it will just work).
You’ll have a custom character that performs near the same as the mannequin would.
(And don’t forget that the hands have to be LOD out too. That’s 30 useless bones at a distance).
Generally I would agree, but In this case, not so much.
As a whole they are just contributing to knowledge dissolution.
People use them without knowing the first thing about character creation.
They end up having problems.
Then they end up not learning anything anyway because replacing a metahuman after learning how to make characters will take you roughly 3 months of learning.
Thus, metahumans are left to just damage their project and make it even more unplayable .
At least, going by all the threads on here and on more reputable forums, thats the way things are going. When you give people who use metahumans an answer that requires them learning something you get an almost immidiate “thanks but no thanks” response…
Obviously not everyone is that way, but in a general sense they are causing the dev community to be even more of leech than usual…
MostHost_LA is absolutely right, but I’m bumping this thread anyway to keep Epic on notice about this issue.
Do they not want to make money with their creations? Cinematic is fine, but for live gaming, create performant creatures so we can propagate your properties.
This is an example of creators making content for themselves and not considering the needs of others.
Somewhat to their defense - not really, but still - knowing what bones a biped will need for a game is akeen to having knoledge of the future.
During development you always tweak/modify improve your rig and character
So much so that you often start with plastic low poly stuff and end up with a better version of whatever top end cinematic actor you may be working with (idk, say for inatance cyberpunk’s jonny).
Its essentially not possible to make an asset that will work for everyone how they like it to work - no matter who you try and please, you always f* someone over.
In the metahuman case however, they are doing more damage to the indy community by doing even less…
Making game ready characters wouldn’t be a solution, it would mean further deterioration of knowledge.
What they should do, if they had any interest what so ever in this and the indy community - which mind you, they do not - would be to hire professionals and set up tutorials and pipelines on how to create custom characters for the engine properly.
Meaning with hair cards and phat assets and proper animation bluperints without code… etc.
So in short, my thing is never going to happen as they really don’t seem to care a single iota.
(The shoulder on ue4’s mannequin is still broken in all animations, since inception, for instance).
Your thing - creating game ready characters - is really only going to make things worse in the long run…
The issue I have is it’s so hard to learn things sometimes. I’m all for learning, but not if I have to scour 13 different sources for the simple basics of what I need.
Just keep in mind that my goals are different from yours. My journey is not yours.
I’m not going to go into it here, but I’m approaching game dev from my own place, and if successful during my first release, will hire a team.
Nothing worse than a leader who knows nothing about their product or the work that goes into it.
Unfortunately, to truly learn you have to either pay - usually by working for free/as an intern to a company that will teach you - or try until you get results.
The sources you touch while learning - any subject - are near infinite. There’s always something to learn.
Ofc, there’s also those who teach who have no business teaching wrong things. And unless they get flagged by others with knowledge, one would have no real way to know unless direct experience is involved.
There really isnt a one and all for character design / videogame character making. Each video/article you find can be great or horrible.
Geberally speaking, FlippedNormals content is decent.
Also, anything where you hear “without further ado” is something you can in all likelyhood thumb down and forget the existence of. Idk why exactly, but I think I’d remeber otherwise if any of those videos were ever worth the time to watch.
If you truly are interested in teaching yourself, anyway, the first step is always to crack an anatomy book if you don’t know your nomenclature and basic functionality.
This keeps you from doing rtard stuff like rotating a palm and complaining to people about the candy wrap effect it generates.
Of special note on youtubers, most of them will try to lick their own elbow for an hour after you tell them it is impossible to do so.
Learning from people like that usually means you aren’t learning anytning correct in any way shape or form.
You come to that realization with experience, mostly in other fields. And trial/error - after what you put together from 20 different sources fails to perform like you need it to miserably.
There are always reasons for it - the main one being nothing you ended up implementing was thought out in the least… but to think something out, you need enough knowledge. Thats why DIY with no team is a step process.
Make, test, trash, make again, etc. (Goes more for coding than for characters, though I myself am still playing with weightpaint 2 years down the line, so maybe it goes for that too )
Unfortunately, I just discovered how bad the Metahuman skeletons are. Not very compatible with setting up a pipeline to get new motion capture data on the skeletons.
For example, left thigh bone you can zero out the rotational values and it goes back to the normal stance pose. If you zero out the right leg, the leg ends up through the body kicking the shoulder. Why?
The idea is interesting however it’s not a very good deployment.
I understand where you come from. I managed to find a way to make MetaHumans usable but after changes in my project it’s kinda useless now because I’m going all into a pretty stylized comic-like design and Metahumans since its conception it’s just too realistic to what I want to achieve. So instead of using pre-made characters, it’s better to learn how to design one from a scratch (or a base) to meet the style one wants to get.
The advantage I find in pre-made stuff (that counts character creators, marketplace assets, and stuff like that) is that you save time, ONLY if that fits your project. But what I see in Epic is that most of their prioritization over “free stuff” and loads of assets circles around a very specific, rather realistic style of models and characters, and forgets about things like what you said, tutorials on how to achieve for their Engine X custom model/stylization/whatever. There are some good tutorials like the outlines and celshading ones (back when UE4 was the main engine), but not much for the new features there are appearing in UE5.
For example, the new video of MetaHuman animator features Hellblade actress testing it. I see the model, and I see a highly plausible recreation of the actress. But in my experience, I was incapable of making a MetaHuman with the eye size of that actress (which is to say, she has pretty big eyes heh). Literally one of the main things people complain about MHs are the small eyes issues (and lack of hair cards). So, how can it be that there are not tutorials around this? Even if it is one that mixes other software with theirs (e.g. “How to MH edit eyes in Blender”), but nope.
Must have missed this message, but look it doesnt really matter.
First of, the hellblade actress has a real 3d model that’s been used in game which was likely created by some artist painstakingly working, and aligning the required photoshot takes or 3d scans to get skin markings etc.
Chances it was generated on the fly for the video are less than 0.
Second of, meta humans have god awful licensing.
Third of, they have horrible performance even in non cinematic mode.
Realistically, no one should be using them for anything but rendering.
If attempting to make videogames, you nedd to learn how to make videogame characters.
Thats the long and short of it, 1 year from topic as it was 10 years before the topic… as it will always be?
Afterall, you wouldn’t go to a butcher and ask him to make you a table for the living room would you?
No, you’d go to a wood worker…
The only difference here is that the meta humans make it so that the butcher can provide you with a table of some sort, but the burcher does not understand or care what is involved in making the table.
Now you went to the butcher, got some table, you take it home, and while you put sruff on it it falls apart misarably damaging all your expensive china and crystal in the process. that is what Meta Humans are.
.