Trajectory ▸ Drawer

TrajectoryDrawer is the one node in the ComfyUI-WanMove-Trajectory pack, and despite the "Wan Move" in the repo name, it's not a video model at all. It's a tiny motion-graphics generator: you draw a path on top of any single image, and it renders an animation of a glowing dot tracing that path with a light trail behind it. Out come a full batch of frames ready for SaveImage or VideoHelperSuite, plus the coordinates as JSON.

Why would you want that? The author's framing is Wan Move - a reference animation showing exactly where motion happens, which you can feed into a motion-transfer workflow as a guide. But the honest use is broader: annotated path overlays for tutorials, neon-sign-style text, presentation pointers, debugging where a ControlNet or video model actually moved. There's no AI in here. No model download, no sampler, no API key, no VRAM. Just PIL, NumPy, and a canvas you draw on with your mouse.

How it works

The node reads three things: your image, a path_json trajectory, and a pile of style knobs. The trajectory is the interesting part. Instead of typing coordinates, you open the built-in editor - click Edit Trajectory (Shift Draw) on the node and a canvas overlay appears. You can load any PNG/JPG as a reference (the canvas adopts its resolution), then hold Shift and drag to draw strokes. Apply Path writes the JSON back into the path_json input, and because it lives in the workflow JSON, you can version, reuse, and share trajectories without touching any files.

At runtime the draw() function takes that JSON, interpolates between your drawn points (frames_per_segment controls how densely), resamples the stroke to fit the target duration, and composites one RGBA overlay per frame: a circular head, a fading gradient tail, and an optional pointer arrow. trail_decay is the tail's memory - the lower the value, the longer the trail persists. Pure PIL drawing, tens of frames a second, no GPU needed.

The inputs that matter

You'll set most of these once and leave them alone. The ones you actually touch:

• path_json - the required trajectory. Leave it empty and the node will tell you there's nothing to draw; open the canvas and make one.
• stroke_width (1–200) and line_color - the look. Color takes hex (#00f0ff is the default neon cyan) or r,g,b.
• trail_decay (0–1) and background_dim (0–1) - trail length, and how much to darken the base image so the glow pops. background_dim around 0.2–0.4 gets you close to the reference-video aesthetic.
• duration_seconds and fps - set duration_seconds to 0 to derive the length from the stroke; otherwise it's target duration. Keep fps the same downstream when you assemble the video.

Also available: frames_per_segment, tail_alpha, line_style (solid or neon), and show_pointer (a debug arrow that doesn't touch the glow).

Outputs

• frames - the IMAGE batch, the whole animation. Pipe it into SaveImage for a PNG sequence or VideoHelperSuite ▸ ImagesToVideo for an MP4/GIF.
• info - a STRING of JSON metadata (frame count, resolution, fps, decay settings).
• tracks_json - a STRING listing every pixel coordinate per stroke, if you want to feed the path into a script or another node.

Installation

Install via ComfyUI Manager (search "WanMove-Trajectory") or the manual way:

cd ComfyUI/custom_nodes
git clone https://github.com/starsFriday/ComfyUI-WanMove-Trajectory

Then restart ComfyUI - the node appears under the 🧭 Trajectory category. There are no dependencies beyond what ComfyUI already ships, so no pip installs and no model downloads. The README's install line mentions a ComfyUI-Trajectory folder, which is a stale copy-paste; clone the repo under its actual name and you're fine.

Where people get burned

• Only the first frame of base_image is used. Feed it a batch and the rest is silently ignored - split batches upstream if you need several trajectories.
• Desktop needs Shift. The canvas draws on touch devices directly, but on a desktop holding Shift is required; otherwise you just get a reminder.
• The JSON embeds the reference image as Base64, not a file path. Keep the original file around; don't rely on the workflow to hold it forever.
• duration_seconds = 0 derives length from stroke length, so a tiny scribble can turn into a very short clip - set a fixed duration if your pacing looks frantic.