What does VACE conditioning being "stackable" mean?

[Boingboingsplat]

I've been experimenting with Wan 2.1 VACE workflows, and digging into its implementations to better understand how I can construct and manipulate the control data that is encoded by the WanVaceToVideo node for more complex workflows. I came across this PR #8240 which seems to refactor the vace_strength property in the conditioning to be a list instead of a single value, though the node seems to wrap the given strength as a list with a length of one anyways. I don't think that passing in a list of FLOATs would work, though maybe I'm wrong?

My first guess is that this has something to do with supporting the conditioning concat/average/combine etc. nodes. From my understanding the CONDITIONING type is a tuple with latents and a dict with additional parameters, and I'm not sure if/how those nodes interact with the extra params. I guess I need a better mental model for what those nodes do in general.

Or maybe this PR is implemented for another use case I'm entirely unaware of?

[Om7035]

Good question — you’re on the right track. The “stackable” aspect of VACE conditioning refers to the ability to combine or layer multiple conditioning inputs (each with its own vace_strength) rather than being limited to a single one.

Before that PR (#8240), the VACE conditioning schema only accepted a single strength value, which made it hard to flexibly merge or mix multiple conditioning sources (for example, when blending outputs from several Wan VACE encoders, or when using the Conditioning Combine / Average / Concat nodes).

After the change, the vace_strength property was updated to a list, enabling multiple conditioning entries to be stored and processed together — essentially making VACE “stackable.” Even though most nodes still wrap a single float in a list for backward compatibility, this structure now allows future workflows (and custom nodes) to pass, merge, or weigh multiple conditioning strengths in a unified way.

So, to summarize:

• vace_strength → list now, so multiple conditionings can be combined.
• This supports stacking/merging behaviors (combine, average, concat).
• Current nodes still use single-element lists, but the new schema makes multi-conditioning workflows possible going forward.

You don’t need to manually pass a list right now — existing nodes handle it automatically — but this change opens the door for advanced control workflows where several VACE conditionings interact.

[Boingboingsplat]

Ok, after much scratching my head I figured it out, and it's much simpler than I had anticipated.

The mentioned PR made it so that you can chain WanVaceToVideo nodes. I failed to notice when I was first looking that it wasn't just strength wrapped in a list, but all of the extra parameters for VACE. The addition of append=True should have tipped me off sooner, but hey, at least I had a reason to dig in and understand data flow in ComfyUI.

Each of these parameters are applied in sequence during the inference. In theory this means you could have a different weight for the reference image and the control video data, or apply two different control videos simultaneously. I'll have to experiment and see what kind of results this can achieve, now.