When composing a scene with outputs from FX, bgeo sequences are still the norm. Is it possible to use value clips to load a bgeo sequence? I noticed reference/sublayer set default values for attributes coming from bgeos, which would interfere with value clips working as expected. But I couldn't get the sequence recognized at all.

Is exporting as USD or alembic the only plausible way to load animation?

Here's a simple example file that uses BGEO files directly as value clips. The trick is that the BGEO files need to have a detail attribute added to them instructing the BGEO USD import plugin to generate attribute values at time samples at a particular time.

mtucker
Here's a simple example file that uses BGEO files directly as value clips. The trick is that the BGEO files need to have a detail attribute added to them instructing the BGEO USD import plugin to generate attribute values at time samples at a particular time.

Ah, okay. A little hard to do retroactively, but maybe I can configure the caching HDA to pre-emptively put these attributes on future geos in the expectation that they might be used in value clips.

On a related note, I've been trying to follow some recent discussion on the USD interest group about changes to value clips in a new version of USD. Something about manifest now being required. I haven't been able to sort out how to generate a manifest from Houdini yet. I think Sebastian mentioned using usdstitchclips, but I thought that was for actually joining files.

A manifest is now required, but you don't need to provide it… If one isn't provided, it will be generated at composition time automatically, if I'm not mistaken. Which is cool, but probably a feature you only want to use in very limited circumstances. Probably makes more sense when the clip files are animation sequences that you're typing together. Not so great for file-per-frame FX data.

usdstitchclips creates two files. One is the manifest file. The other is a file containing the reference to the manifest file and the value clip metadata.

mtucker
A manifest is now required, but you don't need to provide it… If one isn't provided, it will be generated at composition time automatically, if I'm not mistaken. Which is cool, but probably a feature you only want to use in very limited circumstances. Probably makes more sense when the clip files are animation sequences that you're typing together. Not so great for file-per-frame FX data.

usdstitchclips creates two files. One is the manifest file. The other is a file containing the reference to the manifest file and the value clip metadata.

Thanks for the file and explanation.

Are there memory constraints to worry about with USD stitch clips? Do all of value clips have to be loaded in memory at the same time?

USD is generally very good about only loading stuff into memory when it is needed, whether the source is a single USD file or many (be they references, value clips, or sublayers). So I feel pretty confident saying that you shouldn't have to worry about the total size of your value clip files. We do have a bug report (and I think I recall a similar report on usd-interest) that there may be an issue with running out of memory when running usdstitchclips on a large sequence of value clips. But that mey be a Windows-only issue, and I haven't actually had time to replicate it myself.

is it possible to sop import Packed Disk prims into LOPs as a reference to the files instead of being converted to geometry?
also
maybe it would be handy to be able to import Packed Disk Sequence directly as a Clip to not have to deal with this lower level juggling

In my quick test, a Packed Disk prim did get impoerted as a payload of the BGEO file on disk. Please post a hip file if this isn't working for you.

Adding support for packed disk sequence prims as value clips already exists as an RFE. So yes, good idea

mtucker
Here's a simple example file that uses BGEO files directly as value clips. The trick is that the BGEO files need to have a detail attribute added to them instructing the BGEO USD import plugin to generate attribute values at time samples at a particular time.

is this advisable for huge liquid sim caches? I couldnt make it work for deforming caches.