WVAOptions

Every tool eventually grows a settings node, and this is WanVaceAdvanced's. WVAOptions doesn't encode anything and doesn't touch your conditioning - it assembles a bundle of global behavior flags into a WVA_OPTIONS pipe and hands it to the one node that consumes it, WanVacePhantomExperimentalV2. If you've never wired an options pipe before, it's the ComfyUI pattern where a config node outputs a custom type and you run that wire alongside your data into a node that respects it. Nothing happens until you connect the wva_options output to the ExperimentalV2's wva_options input.

The inputs (all optional, all have sensible defaults)

• use_tiled_vae (default off) - use tiled VAE encoding for large images. This is the memory-saver: if your reference/control images are big and you're OOMing during encoding, flip this on.
• enable_debug_prints (default on) - detailed logging to the console. Leave it on while you're learning; turn it off when the console spam gets old.
• debug_save_phantom (default off) - dump the phantom images to a temp folder so you can inspect what the node actually encoded. The tool you reach for when Phantom results surprise you.
• phantom_resize_mode - center (default, crop to fit) vs pad_edge (preserve aspect ratio with edge padding). If your phantom reference images get badly cropped, this is the switch.
• phantom_combined_negative (default off) - use a Conditioning (Combine) for the phantom negative conditioning, which is the manual step the README normally has you do yourself. Turn it on and the node handles the negative merging for you.
• phantom_pad_to_4 (default off) - always use 4 phantom reference latents, filling missing slots with zero tensors. Relevant if you're building for a specific model path that expects a fixed phantom count.

Output: a single wva_options of type WVA_OPTIONS. That's it. No data transformation, no VRAM, no model.

When to bother

If you're on WanVacePhantomSimpleV2 or WanVacePhantomDualV2, you can't use this node at all - neither takes a WVA_OPTIONS input. It only exists to configure WanVacePhantomExperimentalV2. So the honest framing: it's for the ExperimentalV2 crowd, and within that, most people will touch exactly one toggle in their lives - use_tiled_vae when they OOM on big references, or phantom_resize_mode when cropping eats their subject. It's the pack's plumbing, and that's fine: better a clean options pipe than a node with thirty boolean inputs.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/drozbay/ComfyUI-WanVaceAdvanced
# restart ComfyUI

or ComfyUI Manager → "ComfyUI-WanVaceAdvanced". No extra dependencies. And a small workflow tip: if you're experimenting with several flag combinations, you can drop multiple WVAOptions nodes into the graph and swap which one feeds the ExperimentalV2 - same pattern as testing VACE strengths, zero re-wiring of the data path.