Now in H21 one can use blendshapes in Shot Sculpt. But I think the other way around is more important for game development: to "bake" a Shot Sculpt pose as a blendshape, and bake the corresponding Shot Sculpt track curve into the animation, so we can export an .fbx with the Shot Sculpt.

Is this supported?

That's not an off the shelf workflow right now so it would be a good RFE to submit. It might be possible make a custom script that reaches inside the HDA internals to do create a Character BlendShapes set up but it would be a bit of work.

edward
That's not an off the shelf workflow right now so it would be a good RFE to submit. It might be possible make a custom script that reaches inside the HDA internals to do create a Character BlendShapes set up but it would be a bit of work.

Yeah, I tried to reach into the guts of Shot Sculpt but it's a really complicated HDA. I got lost. I think it might be a little easier to just calculate the Blendshape by diffing the output mesh and input mesh. That part is easy. However, I don't know how to dump the curves from Shot Sculpt and somehow wire them back to APEX Scene Animate or export them as a part of animated Blendshape channel of an .fbx...

This shouldn't involve APEX at all. The diffing process would be handled by the Character Blendshapes Add SOP. There's no explicit curves for the Shot Sculpt. Other than the in/out regions of a track, the shapes are always linearly interpolated right now. If we're to convert to an unpacked KineFX format for export to FBX, then the weights will be effectively baked out to per frame values anyways. I think the trickier bit, and possibly not even doable is if you have multiple ShotSculpt tracks. Because then those need to be composited together and I don't how you could even represent that in FBX. At that point, you might as well just export VATs.

Mutltiple tracks should not be an issue (unless using track types or transform modes caveated below), as I assume those are effectively just multiple pose keys active at the same time

Essentially, pose keys within track or clip usually have their weights crossfade
While pose keys on different tracks can be 1 at the same time therefore they add up

Id assume this should be pretty doable for sculpts on static base geo or potentially also for joint space sculpts

Since then you just extract each pose from the internal cache, rebuild its shape from clustermesh and compute its active weights over time based on multiparms, considering its clip in/out, neighbor pose time, track weight, ...
ideally as channel prims

Where it gets tricky is if you use animated blendshape keys or tangent space keys or world space keys on deforming base mesh as those would not be easily represented with a single blendshape with weights