Human Progress

it looks very real nice

I’m still trapped on 2008 hardware, so all of what you just said sounds like crazy science fiction talk from the future to me. :upside_down_face:

If it’s a CUDA error though, then that seems to suggest something to do with the GPU… and since my GPU is (apparently) not CUDA compatible, I can only do CPU rendering, so I haven’t seen that message. Maybe a post to the technical support section would yield some results?

Regarding memory management, my project is pretty lightweight by today’s standards, however exceeding my 8GB of RAM usually results in hanging followed by a hard reboot, so I have to keep an eye on memory usage. Anything more than 4 subsurf levels for the body usually pushes it over (subdivisions eat the most memory), and one 8K texture, a few 4K ones and a handful of 1-2K ones plus a face full of hair has it teetering on the brink sometimes.

3 Likes

wow, I am blown away that you can do amazing stuff with that kind of hardware!
because of your answer, I am now curious how much time does it takes to render a frame of your human projects with the details and Displacement that you are doing. and can you even possibly see what you are doing without really rendering medium or high quality constantly everytime you changes or adjust something? how do you even do it with a hardware from 2008!??

I’m also curious about that, and in addition, your recent screenshots look like new blender, but you said in the past that your hardware cannot handle newer versions. Has that changed?

Usually an hour or two per frame. I spend a lot of time rendering small regions and making adjustments based on what I think the grainy/blurry mess might resolve to.

It was touch and go during the 2.8x series, where it wasn’t working, then a build came out where it was, then it stopped again… but it’s been ok since then. My graphics card just barely scrapes in on the compatibility list; I can’t use EEVEE, and the UI is slightly lethargic, so I use it mainly for rendering and demos.

Don’t want to derail the thread but, are you using adaptive subdivision? That feature is a RAM black hole… If that’s the case make sure you change the Subdivision settings on the Render Properties tab to something like this (the default values are way too much for most things):
image

Also, on the subdivision surface modifier turn the levels viewport to 0, there’s a bug with adaptive subdivision that adds extra subdivisions if the levels viewport is set to something other than 0.

If you’re not using adaptive subdiv, then I’d try keeping all the textures at a max of 4k resolution; unless you’re rendering detailed close-ups at 4k there’s probably no need to use 8k or bigger textures.

2 Likes

oh wow thank you very much! I will try to use these settings from now on
and yes Iam using adaptive resolution it is pretty heavy on ram usage

Hi! Anyone knows what would be the best workflow if i want to add clothes for the character in Blender, using Chris’ rig? Shall I just model the clothes around the character’s body, and then rig it myself, or can I duplicate and edit the base mesh to make it the clothes? I tried to do the latter, but it seems that it’s very difficult to do if I don’t really know how the rig works, and I will mess up the rig very quickly…

I don’t want to add realistic clothes, rather just simplified and stylised clothes, so using cloth sim seems like an overkill. Or is that actually the best way to do it after all?

I haven’t started with clothing yet so I’m no expert, but I was planning to go the cloth sim route, and someone has confirmed that this approach works. If you’re not after realism though and want to avoid cloth sims, then duplicating the mesh might be the way to go, since you wouldn’t have to start from scratch with the weights and shape keys.

Otherwise you could try sculpting them in…

2 Likes

Thank you Chris for such a quick reply!
I tried duplicating the mesh, and editing it, but I’m not familiar with the inner workings of the rig so I just ended up messing up everything… seems the cloths sim + modifying the sim result would be indeed an easier way for me for now.

If you plan on having clothes that fit the body, then you can model the clothes normally and use the Surface Deform modifier to follow the body deformation, without having to bind the cloth model to the rig or run a sim. The process is this:

  • Make a copy of the body model and rename it to something like “deform_mesh” or similar, just so you know that’s just a reference for the modifier. Make this object not visible in render.

  • Remove any subdivision or multires modifier from this mesh, but not the armature modifier! The idea here is to remove anything that modifies the polygon count. (This step is important because the Surface Deform modifier uses the original polycount of the target mesh, so if you bind it to a model that has subdivisions, it will not render properly because the polycount in render time will be different to the polycount used to bind your objects)

  • Now select your cloth model and add the Surface Deform modifier, on the Target option select the duplicate of the body mesh, Interpolation Fallof and Strenght can be left unchanged, those defaults work fine most of the time.

  • Now click on Bind, and your clothes will deform exactly as the body without doing any extra rigging or simulation.

  • Keep in mind that this modifier will work better in lowish polycounts, so if you’re going to use Solidify, Subdivision Surface, or Multires, add it AFTER the Surface Deform modifier.

Hope this helps :slight_smile:

EDIT:
This process can also be used along with a cloth sim; on a dress or skirt for example, you could bind the mesh only on the waist area using a vertex group, and then run the cloth sim for the rest of the mesh, so it will follow the body deformation and the cloth sim will react to that.

6 Likes

Thank you so much for taking the time to write down such a detailed instruction! I never thought of this method, it sounds fantastic!
I’ll definitely try it out, thank you!

2 Likes

The Plica Semilunaris (or “third eyelid”):

Per millimetre, this little triangle of mesh represents the most time I’ve spent on anything to date. It had to be thin enough to be wedged in-between the eyeball and eyelids, and appear to be touching them without actually touching them (otherwise there’s a conspicuous black artefact at the point of contact), while being thick enough to withstand all the forces tugging it around without intersecting itself or the tear line, which is sliding over it and contorting with the eyeball and eyelids.

This was initially facilitated by extending the plica geometry so that it travels more evenly over the eyeball, the upshot of which is that it alleviated the darkness in the earlier version, which it seems was caused by the Random Walk SSS exceeding the geometry thickness. I bolstered this by adding a greyscale map to reduce the SSS depth in the thinnest areas, and softened the sharp leading edge by adding a transparency gradient and mixing in a Glossy BSDF to retain highlights on the now invisible edge.

While I was at it I reshaped the tear line to cover the caruncle, to stop it from tangling up as it follows the plica. Since I had extended the plica, the bump on the tear line (which is tracking the eyeball) needed to slow down where it’s overlapping the plica, so it looks like the meniscus is moving with it instead of streaming over it. I faked that by dissolving it into a stationary bump at the caruncle.

Other improvements include refinements to the skin material, reshaping of the eyelids, a makeshift bump on the eyelashes (more about that here), a fix to a perplexing eyelid driver glitch, and more UV revisions than I care to remember. And I finally got to make use of that Iris sculpt I made back in 2013!

Incidentally I had to re-render the upper eyelid in 2.92 because of too much splotchy flickering in the SSS, which neither disabling denoising nor increasing the samples would fix. It’s still visible, but not as pronounced as in later versions, which is a tad concerning…

22 Likes

Amazing work again Chris! I just have a question. in the facial rig that you sell in gumroad does it also includes tearline in the eye? also how does it work if we want to reshape the basehead? since I believe that the rig is shape key driven? or is it? Thanks in advance

Wow… I really don’t know what to say… This is mind blowing, such a detailed and impeccable work! :clap: :clap: :clap:

You continue to amaze with this work. Wow!

Thanks :slight_smile:

It doesn’t at the moment - I haven’t decided whether it should be part of the model, part of the textures or a “minutia” upgrade.

It is mainly shape keys, and reshaping it would be done via the proportion controls, sculpting, and/or adding a shape key. The eyes and mouth are particularly sensitive to sculpting and mesh editing, so it might be necessary to edit the individual facial shape keys to compensate for significant changes in those areas.

3 Likes

Astonishing work, @ChrisJones! Respect… :clap:

1 Like

Love the lanky Giacometti quality of your human. I’m still Maya minded, and its taking me a moment to get going with Blender and the rig.

Mocap is DIY for the moment, but I’m doing some work in preparation.

Is there a way I can get the bone coordinates of the head bone? I want to place the head directly, having the pivot line up with the heads eventual position/rotation would be very helpful.

Amazing, but is it available for obj files? I could use it for mobile modeling.