WanVideo Block Activation Viewer

Somewhere between wiring up WanVideoActivationEditor, forgetting which preset you picked in the builder, and queueing a 14B video that takes minutes per clip, you'll wonder whether the injection is even configured. That's what this node is for. It's the "check my work" node of the pack: connect your model, optionally connect the activation string, and it tells you what state the model is actually in - no render required.

It's deliberately boring, and that's the appeal. Video generation is expensive; a 30-second sanity check beats a 5-minute generation that silently did nothing because the activation pattern was all zeros.

What it shows

Feed it the model from the activation editor chain and it reads transformer_options for the config the editor stored:

• which blocks are active, the injection strength, whether runtime patching is on
• a checkmark-style patching status: "✓ Transformer blocks are patched" vs "✗ Patching disabled - injection configured but not active"

Feed it block_activations (the STRING from WanVideoBlockActivationBuilder) and show_visual on, and it renders the pattern as a grouped 4×10 block visualization so you can eyeball "yes, I meant the late blocks" without decoding the string yourself. It'll even try to detect common patterns for you.

Inputs: model (required), show_visual (default true), plus optional block_activations and text_embeds. Output is a single visualization STRING - a report you read, not something you wire onward. Treat it as an output/display node; it doesn't modify the model.

Install

It ships with fblissjr/ComfyUI-WanActivationEditor, so the install is the same as the rest of the pack. ComfyUI Manager: search "ComfyUI-WanActivationEditor". Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/kijai/ComfyUI-WanVideoWrapper   # required first
git clone https://github.com/fblissjr/ComfyUI-WanActivationEditor
cd ComfyUI-WanActivationEditor && pip install -r requirements.txt

Restart ComfyUI afterward. The dependencies are light (torch, numpy, duckdb, zstandard - the last two auto-install).

Gotchas

The viewer reads what the editor wrote into transformer_options. So the order matters: it must sit after the editor in the graph (or receive the edited model), or it'll report "No activation editor modifications found" and you'll think you've gone mad. That message is the giveaway - it means the node ran before the editor did.

Also worth knowing: the viewer reports configuration, not ground truth about whether injection visibly worked. If you want to verify the effect actually lands, that's what WanVideoInjectionTester is for - it compares the main and injection embeddings and tells you whether the difference is big enough to see. Viewer answers "is it configured?", tester answers "will it work?".