Paige Two, Learning Character Rigging, Animation, and more

Frustrated with weight painting. With my ignorance of the details of how to do it really.

Auto-Rig Pro add on does a lot of the work but I’m finding things needing much modification of the weights. The facial rig will surely need much fine tuning. I’ll have to learn the how to’s of weight painting to get useful results. Just making a mess now. Not complaining about having to learn the tools though. That is an obvious necessity for CG.

I intend to learn rigging with the native Blender tools too and I’m starting to think that will be better than add on’s. There may be no gain over doing it “manually”.

Auto-Rig Pro’s rig allows unnatural stretching of the figure. That may be useful for cartoon characters but is harmful for a realistic character. Hopefully there is way to disable all of the “cartoon” features with a few clicks.

Blender (2.90) stinks for user interface regarding weight painting. Again, my ignorance is at work but I have to select each of thirty trillion vertex groups to determine what bones are weight mapped to what area? Insanity. The weight paint brushes move in a jerky strobe effect removing any useful feedback. Paint, wait, see result, repeat. Maybe my AMD FX 8320 and two 1070 ti’s is not enough power for smooth live feedback in this. Changing some settings could improve it.

Like many things in Blender, the learning curve is steep, the UI not great, but CG power awaits beyond these obstacles.

When I’ve learned enough about animation I’ll finish an animated short I’ve wanted to make since 2018. Although the action figure toys come to life for that don’t need the realism and detail of Paige I tested how many Paige’s I could append into that scene. The answer is 2. Nearly all of 16gb ram was used during bvh prep and the gpu with the screen plugged in stayed just under its 8gb.

About 3 minutes render time at 1440p. This might need a render farm to complete.

The hero, Jedi Tevu (here), has no hair and only 20K polygons. His nemesis will also be less resource consuming that a Paige. So my machine can do this.

Good tutorial about the NLA. https://www.youtube.com/watch?v=Hz1TwvSNsrA

Not much time lately to learn about animation. Took a break and played with this. Decimated to about 15K to 20K polygons, tissue add on for dual mesh, extrude faces, scale with individual origins on, smooth modifier, apply.

I want to get back to this but work, work, work and practical things. No time for fun.

I really like your work, and I hope you make more really soon.

Blender 3.0 speed tests and improved Paige image.

NEW IMAGE

How do I make this image the thumbnail for my portfolio?

I changed the skin material’s subsurface scattering method to random walk and the micro scattering to multi-scatter ggx. I stopped using the Filmic color look up table and used the Blender standard one. This improved the skin appearance. Her eyes have more realistic color. The hair got slight random color variations and is more shiny. The lighting is a mixture of an HDRI environment and the studio lighting of the earlier images. I also post processed the image in Paint Shop Pro with a filter called Fill Light and Clarity to make it crisper.

SPEED TEST

Blender 3.0 cycles vs B2.92

Paige:
– 100 samples, no denoise, no adaptive sampling: B2.92 1m11s B3. 57s
– 469 samples, no denoise, no adaptive sampling: B2.92 5m6s B3.0 2m47s
– 469 samples, no denoise, with adaptive sampling: B2.92 4m43 B3.0 2m28s

It is sometimes inconsistent though. Sometimes 3.0 is only slightly faster. I’m sure it will be better when I learn more about 3.0. No denoise for tests since optix denoise is not available in 3.0.

Torus:
– About 150K polygons, 1000 samples: B2.92 18.9sec B3.0 17.5sec, not much difference

Ice Moon:
– For this image B3.0 was around 2x as fast.

Big disappointment: No optix denoising. I could select optix, nlm, or oiden with B2.92. In B3.0 I only have oiden. Oiden removes almost all detail (yes, in 3.0 too). Only optix denoise works well in B2.92 for Paige.

Maybe my gtx1070ti’s are to old for optix? In B2.92 perferences I can select optix instead of cuda but that rendered very slowly. Optix denoiser is good though. In 3.0 optix is not available but a warning babbles about drivers so I may just need to update.

UPDATE Dec 14, 2021
Updated the gpu driver and optix is back. It is different though. Denoise quality, loss of detail, is about the same with B3.0 optix vs oiden but optix is faster. B3.0 is almost 2x render speed though so I can throw so many samples at it that it doesn’t need denoise but renders in about the same time.

Shoulder Trouble

Still can’t get the shoulders to work. Auto Rig Pro. No matter where I put the shoulder dots the skin deforms in strange ways. I even made a reference image in Daz (This is a Genesis 2 figure) showing the Daz bones and used it in Blender to place the dots.

Render Farm?

This is a project for me to learn about CG characters, animation, and related things. Ultimately I hope to have Paige move with mocap and talk. I’d like to do realistic rendering for a talking video. A quality close up head view in cycles B3.1 takes from 2 to 5 minutes to render with my computer.

My Computer
120 hours per minute of video at four render minutes per frame. 30fps.

Render Street

On Demand GPU: $600 I could never afford that for a hobby.

Monthly: It would not be right to sign up for monthly with the intent to stop asap. For some people monthly could be a bargain but I would only rarely use a farm. 12 months x $50 = $600 is the same as On Demand. Never happen for a hobby.

This is a check of feasibility for me to use a RF not a belief that they are meant for hobbyists. These prices may be good for professionals.

There are free render farms

1 Like

Mostly a dream. Real life will probably allow little time for fun projects in the foreseeable future.

Thanks. I looked at sheep it before. I’ll look again. Maybe I misunderstood but don’t you have to provide your computer to work in the sheep it farm in order to use the farm? Such access does not meet my “foil hat level” privacy and security requirements. Otherwise that seems like a fair way to pay for using it.

Concierge render farm looks less expensive but I can’t imagine it being $50 when RenderStreet is $600/min.

I can look at it as an opportunity to learn more about eevee and do some or all of the animation with that. Or cycles with patience and reduced size and quality but still good enough.

You probably need to adjust the weight painting. I’ve never gotten perfect results from auto-weights, especially on the shoulders and hips. I know autorig pro probably has special auto weight algorithms, but still, it may require adjustment.

The daz figures have decent weight painting and rigging as is. Maybe it would be worthwhile to use the Daz to Blender plugin? I think there’s a button to convert it to rigify. I’m not sure I’ve tested that rigify part since I build my own control rigs, but I know you can get the G2F from Daz to Blender with the rig intact through the plugin, and I believe the deforms were fine. Been a while though.

If nothing else, maybe seeing how daz weight paints the G2F and where they’re placing the bones will help your custom weight paints.

Another important tip: Did you export your figure at base resolution (0 subD)? I believe Daz defaults the mesh at one or two levels of SubD that have to be removed before export else your mesh will have quite a lot of extra vertices which will make weight painting harder.

2 Likes

Thanks. Daz bone placement is why I made the back ground image for reference with Auto Rig Pro. This figure has 82,000 polygons. Maybe level zero subd is less. I’ll try exporting one with fewer polys.

I want to learn about rigging characters so rigging meshes in Blender is important for that education. Otherwise I might use an importer to preserve the Daz rig and weights. Also, Auto Rig can do mapping of mocap data to it’s rig. That is also a factor.

Yeah, go ahead and re-export that and use Blender to give it a new SubD modifier. That’s really the workflow you should be using anyway (I learned that lesson the hard way too).

That 82,000 polycount definitely is not helping you. Export it at 0 SubD/Base resolution and you’ll be at 21,098 polys. I can’t promise that’ll magically make weight painting easy (it’ll still be brutal) but it’ll make it less horrible. And you’ll get better performance when posing. Might even be worthwhile to decimate the 0 subd model and give it the same rig and weight paint that and transfer the weights.

Another note: when posing the arm in a raised arm position, anything higher than t-pose will need the collar bone raised too a little to look natural, no matter the weight paint (if you raise your arm and pay attention to your collar bone, you’ll notice why).

To actually learn rigging though, I would suggest less automation. Learn rigging by making rigs. Auto Rig does nearly all the rigging for you, doesn’t it?

1 Like

Just the basic automatic rig generation, not complex things…

Yes I was moving the shoulder in an unnatural way. Using the collar bone too looks better.

The Optimize High Res option works and it rigs with all the polygons. I used high res for export since I made her more athletic in Daz and the abs would not appear much in low res. You are right that I don’t need so many polys. The abs will be covered with clothes and aren’t important and I’ve seen methods to use normal maps etc. to add detail without it all being in the mesh.

I do wish it would rig a high poly mesh without the Optimize option. A no compromise high res binding should be possible if the user has patience.

HELP? I need to learn to fix the weights of the lips: Help weight paint Auto Rig Pro lips
So far I have used hide to reach the inner parts but have not succeeded in removing the weights from the incorrect bones. I am also making the gums+teeth and probably tongue separate objects.

HELP? Parts of the skin or body intersect sometimes. For instance, bending the elbow causes part of the forearm to move inside of the upper arm. A Corrective Smooth modifier helps. I checked Preserve Volume in Auto Rig’s options. It looks ok but I want to know a way to have the body deform more realistically with no parts going inside of other parts if that is possible. Self collision for the mesh I guess.

I agree about rigging from scratch. Knowing all the steps and tools, a more in-depth knowledge, of rigging allows more complex things like MichaelBenDavid noted and greater ability to fix generated models and automatic rigs. I want to learn it from scratch if I ever have time to.

1 Like

I placed the lips markers (Auto Rig Pro) more on the edges of the lips and they don’t have weights from the wrong bones now. I would still like advice about correcting that (Help weight paint Auto Rig Pro lips) with weight painting or other methods. Also with the intersection noted in the previous post.

The upper and lower gums+teeth are separate parts. What Auto Rig Pro bones should I parent them to? Or should they be 100% weighted to the bones? Both?

Also, does ARP do anything like its auto eye lid thing with lips? Teeth and gums affecting lips?

If you’re using the HD morphs, you can use the multires modifier to copy the mesh details from the high res to the base res. If they’re not HD morphs, you probably just need to use subdivision on the base res import to get the same effect but you can still use multires on the base resolution mesh and transfer the high res details to it if it looks off. I think it should look roughly the same though unless they’re HD morphs.

You need to check the weight paints. You may have weights on vertices you don’t want. It’s a slow and tedious process. First time you do it, you’ll probably hate it. Worth it though.

Second time you do it, you’ll still hate it :wink:still worth it though

1 Like

Some success!

This is straight ARP rigging. No adjustment prior to skinning. No weight paint corrections. Keyframed by hand. I posed the face for 6 or 7 expressions/movements and keyframed the entire rig at each for convenience. I guess it would be more efficient for space and speed to keyframe only the effected controls.

In HitFilm Express the workspace wireframe part didn’t look good until I set the video bitrate up to 30mbps.

I know to use weight painting to correct some things. The question was how to do that for the spikes. I realize now that my question in the Help Weight Paint thread linked in an earlier post lacks detail about the problem. I’ll update that in the future. For now I discovered the Hide feature which gave access to the vertices. Placing the ARP lip markers near the outer edges of the lips, away from the mouth opening, produced good results but is is important for me to learn weight painting since sometimes there will be no alternative.

Multires. An alternative to normal maps. It is good to know several ways to do a thing. Versatility lets a CG artist choose an appropriate level of compromise. Ultra-res no compromise, multi-res some detail in the mesh, normal/bump non-mesh details, rigged low res drives high res … learn many tools and apply your creativity to their use.

Much real world now so I won’t be able to do much on hobbies for a month or more.

Nice to see your progress on this.

It was quite a journey to go through this thread. I am also a hobbyist so I empathize with having to fit within free time for an inherently time-consuming hobby (I knew computer graphics would take a lot of time to learn, just underestimated how complex everything is). Sometimes I console myself with the fact that materials, rigging, animating, modelling, and sometimes even modelling just clothes and hair are things that people specialize in professionally.