WanVacePhantomSimple

This is the node the pack started with. WanVacePhantomSimple is the original single-context VACE + Phantom node: control video in, reference images in, and it produces the conditioning package that makes Wan follow both. If you're pulling this up fresh in 2026, the honest framing is that you're looking at the v1 of the thing - it still works, but the V2 version (WanVacePhantomSimpleV2) is the one the author recommends and the one new workflows are built on.

The big practical difference is model patching. WanVacePhantomSimple has no model input, so you have to run your Wan model through VaceAdvancedModelPatch yourself before the sampler. The V2 node does that internally. Miss the patch and per-frame strength lists won't work - the VACE strength behaves as a single scalar no matter what you feed it.

Inputs and outputs

Required: positive, negative, vae, and the geometry - width (832), height (480), length (81), batch_size. Optional:

• control_video - your pose/depth/layout control video
• control_masks - masks for the control regions
• vace_reference - reference image for the VACE context
• vace_strength / vace_ref_strength - control and reference strength, default 1.0
• phantom_images - the character reference images that give you likeness

Outputs are the standard five: positive, negative, neg_phant_img, latent, and trim_latent. That neg_phant_img output is Phantom-specific - merge it with your main negative using a ConditioningCombine node. And trim_latent is the count of frames the node reserved for the reference embedding, which is why the latent that comes out of an 81-frame length may not be 81 frames long.

Install and models

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

You need Phantom 14B (Phantom-Wan-14B_fp16.safetensors from Kijai/WanVideo_comfy) plus a VACE module (Wan2_1-VACE_module_14B_bf16 or fp8), or the pre-merged Wan2.1_VACE_Phantom from InnerReflections. The 2.2 A14B Low Noise LoRA from the README is worth a look for motion quality without losing Phantom/VACE behavior.

Phantom has two rules that don't change between V1 and V2: run CFG above 1.0 (1.5–3.5), and don't overcomplicate the reference image - subject in focus, background can stay. People who hit bad inpainting or lost likeness on this node are almost always violating one of those, or they forgot the patch node.

When should you reach for this over the V2? Realistically: when you're opening an older shared workflow and don't want to rewire it, or when you're deliberately chaining VACE nodes and want a patch-once setup. Otherwise start with WanVacePhantomSimpleV2. The pack is by drozbay (AbleJones), who maintains the WanTests comparison site and, since early 2026, the RES4LYF sampler pack - so "simple" here is a working name, not a promise of simplicity.