Next gen real time hair workflow?

System · Apr 2020

[Moderator note: original author requested their account be deleted. We're preserving this topic for it's overall usefulness, and scrubbing their name per their request.]

So, I've been spending the last week trying to learn a workflow that won't get obsolete in two years time when the new consoles are well stablished and kicking. Trying to figure out the best approach. I know there's no fixed path towards achieving your goals, you can do your thing the way you want if that gives you the best results, but still, I want to learn the "most correct" workflow.

So far, the most popular and most used way in the current gen seems to be to set your Xgen hairs in planes and render them to get high quality textures to be applied in your manually placed hair cards.

Is this going to be a thing in the next gen too? Is it still going to be the best most efficient way?

Thanks!

Obscura · Apr 2020

Yes. Though the placement doesn't need to be fully manual.
Some engine has own solutions (see the new Unreal Engine hair tech, and Frostbite has something similar too ) but it doesn't really look like those are targeted towards games that runs on low spec systems. I doubt that those tech can be used in "mass". Like open world games wont have those for sure. The Unreal Engine one doesn't have game ready performance yet even on srtong systems. So I'd say yes cards will be still a thing for long. Its much cheaper and looks ok if its made nicely.

thomasp · Apr 2020

What's desperately needed to open up new possibilities for hair in my view is mainly to make the whole process much faster and hairstyles more adaptable/reusable.

For now like Obscura I'm sceptical that the groom approach - tasty as it may look - can really substitute hair cards in this upcoming 'gen'.

I think it may end up like tressfx and all these solutions - something that drags down the framerate considerably and doesn't look as subtle as hair cards can (hairlines, facial hair, etc).

Done at high enoguh a resolution I do find it likely however that it will end up being the preferred solution for games like Detroit or the Supermassive titles. Titles where all the focus is on dialogue basically.

I'd definitely look into moving away from all manual placement. Use that only where you absolutely must and for final tweaking. Horrible slow way of working IMO.

Also don't do the texture first - it will lock you in and determine the overall look way too early. Use texture mockup doodles instead. The focus should be on geometry/placement, not on how you generate the texture. You can do that in a myriad of ways. Again, IMO.

RS7 · Apr 2020

Sooner or later there will be something they call alembic hair.. something similar you see on offline renders today. This tech is under development and it is not ready for in-game use to this date, tho.
Play with it if you are looking for next-next gen

Until then - hair cards has to do the trick. For example Maya Xgen has some nice tools to help you generate hair cards based on hair system you blocked in. This method is widely used.

thomasp · Apr 2020

Nah, I wouldn't go with fibermesh. That's inflexible (can't change shape of fiber geo nor number of strands generated once created). I'd look into generating hair geometry from a hair system or grooming it based on geometry. It all depends on your chosen software. Ornatrix, possibly Hairfarm, maybe xGen, definitely several solutions available for Blender.

You can use fibermesh to help define a hair system (via export as curves to be used as guide curves in your target software) if you are familiar with it. Else spending time with the far more flexible hair styling and grooming tools in regular DCC apps will be preferable.

Alex_J · Apr 2020

fibermesh is simpler to use, so if you are just starting out and need to understand the overall pipeline it will do to get started with learning.

But if you are going to be doing this very much at all, listen to the experts here and go for a more flexible tool.

Obscura · Apr 2020

You will know it only once you applied textures and shading. Then you'll see how is the density, overlappings, clippings etc. The 2 divisions along the planes seems unnecessary and unproportional compared to the general density of the geometry. I mean there are much less silhouette geo than this so the planes are visibly angular in the silhouette.

thomasp · Apr 2020

Hey when I said don't bother with textures first what I meant was do mockups for the shapes you reckon will be needed to pull off the hairstyle - and slap them onto your geometry first thing. Should have been clearer about that I know.

I'd never recommend just doing geometry and adding textures much later. You'll need to see right away how things fit together.

Final textures on the other hand - the stuff you see people mentioning first thing in those xgen tutorials - IMO those are not needed in the beginning and are more a hindrance than a help since you'll want to try out different strand shapes, see just how many variations are really required and what kind of density/strand thickness you should be aiming for.

Mockups that help determine this can be done by just painting a few strands, all you need is alpha anyway to figure it out.

Anyway, good to see you went with the geometry from hair system route, keep it up!

Oh and to stay in control of the polycount I'd go with low res output from your hair system and slap on a subdivision to keep it round. Easier to uprez/reduce later in my opinion. I've always worked with software that has a modifier stack so there it's non-destructive, not sure how Maya handles that - I trust there is a way without commiting to a high polygon count from the start.

Next gen real time hair workflow?

Replies