ml deform and houdini

[anichar]

Hello

Starting to read up about the new ml deform in UE5.2, wonder if anyone have started to use it and have some insight in the procedure in houdini?  should it be possible to do without going through maya?

 

[vinyvince]

I thought the training was done using a complex houdini simulation (vellum?) , do you think it's more Ziva based?

[chovasie]

Only part that is done in Maya, is generation of training ROM.  In Maya there is a script that takes your skeleton, you define degrees of freedom for every bone in body, and generates N number of randomly generated poses. You can do that on your own in Houdini or any other software and export as FBX.

That ROM fbx is referenced in houdini, and from that point there is no any attachments with maya anymore.
UE 5.2 presentation uses Vellum, it's not Ziva.

Let me know if you have more questions.

Cheers
C
 

[vinyvince]

What does force you to use Maya for the first step ?

I suspect we will see a generalization of this approach, you could probably do the same for a car rigid body deformer for example no?

 

 

 

 

[omnilen]

Will pick it upp again when h20 realeses, learn ml onnx  in houdini 20 and ml in UE5.4 in pararell.

[chovasie]

@vinyvince There is no need for the first part to be done in Maya. Main reason is that all pipeline other than simulation is in Maya. Also some advanced nodes needed for scapula rigging, like RBF pose driver and other used, were more convenient to do in Maya at the time, than import baked FBX in Houdini and drive skeleton that way.
But if you want to rig skeleton using kineFx in Houdini you can, there is no reason not to do it that way other than p[personal preferences.