mocap Motion capture retargeting in Blender

[Arcturus]

Here's a .blend file with a Darkmod NPC rig: https://drive.google.com/file/d/1-odjc8A54ED5qgO7pA01Nr59U2_2jFmL/view?usp=sharing

set up to be used with this motion capture animation pack https://www.rokoko.com/mocap/free-mocap-assets

It's a stripped down version of this Blender rig. Exporting works the same way.

For retargeting you'll need to install this addon: https://github.com/Mwni/blender-animation-retargeting It works in Blender 3.6 as well as 4.5. Animation playback of longer files is waaay faster in 4.5 but the drawback is that md5 file exporter won't work with anything newer than 3.6.

This pack contains different skeleton standards. This setup was made for Mixamo which seems to be majority of files.

• Import .fbx file with scale 48, manual orientation: x forward to match the Darkmod skeleton. When selected, keyframes of the imported animation will be displayed.
• Select armature_control. In the retargeting addon tab (in right panel), in 'Source' box choose the imported skeleton name.
• Load config file: Darkmod_male_NPC_mocap_Mixamo.blend-retarget This should pair up both skeletons.
• First frame in animations is a t-pose. Proper animation starts at frame 2.
• Typically animations start offset from the center. You can move the character to the center at the start of the animation simply by dragging armature_control object.
• Similarly, character model may be slightly below ground level. Move armature_control up to adjust.
• Origin bone in tdm_ai_proguard has constraints to follow the pelvis. Otherwise it would stay in center.

Here's what happens when origin bone stays in the same place. This is idle animation playing:

And here's what happens when origin bone follows the character:

I found that in order to match imported skeletons I had to bend darkmod skeleton forward a little bit. That pose is saved as "tpose_mixamo" single frame animation.

I don't guarantee that this is the best way you can do it. Perhaps skeletons could be matched in a more ideal way.

Remember that there are bones for clothing in Darkmod, that aren't in the motion capture files and need to be animated by hand :)

[Arcturus]

Small update: added copy rotation constraint in head bone in head_old_man_MD5_Armature.

[Arcturus]

Update 2: fixed position of the head.

Update 3: Added 'Copy Rotation' constraint to origin bone in tdm_ai_proguard, in addition to 'Copy Location'. Those can be turned off if needed. 

[Arcturus]

Updated with latest changes to old man's model.

[Arcturus]

Somebody on Blenderartists linked to this AI tool that extracts motion capture data from video:

https://github.com/PozzettiAndrea/ComfyUI-MotionCapture/tree/main

There's an older open source mocap project:

https://freemocap.org/

[Arcturus]

AI animation generator from Nvidia. Requires 17GB of VRAM to run locally:

https://research.nvidia.com/labs/sil/projects/kimodo/

[thebigh]

I for one would not want TDM to contain too much AI slop.

[JackFarmer]

8 hours ago, Arcturus said:

AI animation generator from Nvidia. Requires 17GB of VRAM to run locally:

https://research.nvidia.com/labs/sil/projects/kimodo/

Could this be used to create new animations for TDM?

 

6 hours ago, thebigh said:

I for one would not want TDM to contain too much AI slop.

I know what you mean, and I’m torn on this issue. But by now I’m starting to wonder why we shouldn’t make things easier for ourselves here, after all, we’re doing all this for free in our spare time. If we were getting profits for it, that would be a completely differentd story.

After several voice actor casting fiascos (HHBAM was the worst I’ve ever experienced—it pushed the release back by at least three months not to mention the time - and energy! - consumption needed for new voice actor briefings/instructions) that I’ve encountered in the past, I can understand why people might be considering other options here.

Animations are just one of the hardest things to get right. Just think about how many of our friends here have already tried their hand at the animations for the Werebeast spending countless hours, and now, after x years, we still don't have a fully functional character.

[Arcturus]

7 hours ago, thebigh said:

I for one would not want TDM to contain too much AI slop.

In Nvidia's tech report linked on the website you can find that they trained it on 700 hours of motion capture data from: https://bones.studio/ai-datasets

How is using Nvidia's tool different from using mocap animations directly?

[Arcturus]

3 hours ago, JackFarmer said:

Could this be used to create new animations for TDM?

Possibly. I assume there is a way to convert those animations to a format usable in Blender or other editor. The GPU requirement are a bit high though. That Comfy UI mocap model looks interesting and probably doesn't need 17GB of VRAM.

[Arcturus]

On 3/27/2026 at 11:13 AM, Arcturus said:

That Comfy UI mocap model looks interesting and probably doesn't need 17GB of VRAM.

It's a wrapper on this tech: https://zju3dv.github.io/gvhmr/

It's for non-commercial uses only. That ComfyUI wrapper requires a lot of dependencies. There's this version for ComfyUI that seems to streamline things, but I couldn't get it to work. Not enough people with right technical knowledge are interested in it so far. From the published examples it would seem that it gives better results with a single camera than Freemocap with multiple.

[datiswous]

So the idea is that it converts model animation from a video file into blender and you have to export it to md5 files and then somehow activate those animations in tdm?

[Arcturus]

@datiswous On Github pages for those ComfyUI plugins you can see that they they were able to convert those videos to BVH format, which is an old standard for skeletal animation. And then convert to a different format. The second one even mentions Blender installation linked directly to ComfyUI.

But when I try to install it, there is always a problem with this or that module missing. There's always possibility that I wouldn't be able to run it anyway, if it requires top of the line GPU.