Wan Blend Visualize

The name says "Visualize," so your first instinct is a pretty picture. What WanBlendVisualize actually returns is a STRING - a text/ASCII map of your prompt transitions. That's not a bug, and once you've stared at one it's genuinely useful. This is the diagnostic node from the ComfyUI-WanSeamlessFlow pack (fblissjr's "wanvideo - seamless flow" set), and its whole job is to show you where each transition lands in your video and how hard the blend is at each frame.

Think of it as the printout you check before committing to a long render. A 257-frame Wan generation with three |-separated prompts costs real minutes of GPU time; knowing your blend zone actually covers the cut - and isn't 2 frames wide, which is a hard cut wearing a costume - is the difference between one render and three.

How it works

The node reads the blend settings straight out of the embedding dict it's handed - blend_width, blend_method, and the prompt count. That's why the wiring matters: feed it the output of WanSmartBlend (or WanAdaptiveFlow), not raw encoder output, or it has nothing to draw.

It then divides num_frames by the number of prompts to find each section boundary, marks boundaries with ┃, and shades the blend zone around each one with characters that get denser as the blend ratio climbs (▓ strongest, then ▒, ░, ·) - with the density following the interpolation curve you selected. smooth transitions fade gently; a bounce curve produces that silly stutter pattern in text form. The whole thing comes back as a runnable string with frame numbers down the left and a legend on the bottom.

The inputs that matter

• text_embeds (WANVIDEOTEXTEMBEDS, required) - the blended embedding bundle. Wire this from WanSmartBlend or WanAdaptiveFlow.
• num_frames (default 81) - the total frame count of your video. This has to match the real video length or the map is lying to you.
• show_blend_zones (default true) - whether to shade the blend zone or just show bare section boundaries.

Output is a single blend_info string. Preview it by routing it into a text-viewer node, or just read it from the console log. It's not an image node - don't wire it to a preview that expects a picture.

Install

Same as the rest of the pack - ComfyUI Manager, search wanvideo - seamless flow (or "WanSeamlessFlow"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/fblissjr/ComfyUI-WanSeamlessFlow

Restart ComfyUI. No extra dependencies, no models to download; it needs the WANVIDEOTEXTEMBEDS type from Kijai's ComfyUI-WanVideoWrapper, which you should have installed first.

Where people get burned

• It returns a one-liner no-op message - "No blend zones configured or only one prompt available." - whenever blend_width is 0 or there's a single prompt. That's usually a wiring problem: you fed it unblended embeddings, or your text encoder only has one prompt.
• Frame count mismatch. If you say 81 frames but the actual generation is 257, the boundaries are in the wrong place and the map is worse than useless. Keep num_frames in sync with the sampler.
• Expecting an image. It's a text diagnostic. Wire it to a string/text preview, not an IMAGE input.

This is a small, young pack with basically no community footprint - treat it as a helpful debug tool, not a pillar of your setup. But for sanity-checking your multi-prompt Wan transitions, it's the quickest look you'll get at whether your blend is a gradual fade or a hard cut.