LASED Animate

An edge map tells you where the contour ended up. LASED Animate shows you how it got there - the swarm spawning, wandering, latching onto the boundary, and the contour emerging over the run, captured as an IMAGE frame batch you can feed to a video-combine or preview-animation node.

Under the hood it's the same swarm simulation as Swarm Edge, but with a hook that captures an overlay frame at regular intervals while the run plays out. Every steps / frames ticks it grabs a frame, and it always appends the final state, so you get roughly frames evenly-spaced snapshots plus the finished contour. You can see the ants converge on the boundary in a few seconds instead of reading a gray field and hoping.

The inputs that matter

• image, seed_x, seed_y - same as the rest of the family: the source image and the click point in input-image pixels. Note the node uses only the first image of a batch, so feed it a single image and keep batches for elsewhere.
• frames - how many overlay frames to capture across the run, default 60, range 2–600. More frames = smoother video, not a longer simulation - the simulation cost is set by steps, not frames. There's no point cranking frames if your steps count is low; you'll just get near-duplicate frames.
• Everything else is the standard LASED swarm set: tolerance, color_space, ants, steps, hand, beta, epsilon, seed, mode, max_side.

The output is a single frames IMAGE batch. The natural sink is anything that turns a batch into a video - a video-combine node, or SaveAnimatedWEBP/GIF, which is exactly what workflows/lased_animate.json does.

The honest use case

Yes, it's a demo toy - a glowing contour assembling itself is a satisfying thing to watch and a nice workflow to share. But the genuinely useful reading is diagnostic. One animated run tells you in a glance what a dozen static edge maps won't: whether the swarm is converging on your region's boundary or wandering around it. If the ants spend most of the run circling a neighboring blob and only settle on yours at the end, you'll see it immediately - and then you know tolerance is too high (or your seed point is too close to a border). Static output would just show you a messy field and make you guess.

A couple of things worth knowing: the contour's completeness tracks steps - short runs freeze mid-convergence, so don't animate a 200-tick run and expect a finished boundary. And since everything is seeded, the same parameters animate identically every time, which makes it a genuinely reproducible way to compare tolerance settings side by side.

Install

Nothing special - the pack installs as one unit. ComfyUI Manager search for ComfyUI-LASED-mask, or git clone https://github.com/MushroomFleet/ComfyUI-LASED-mask into custom_nodes and restart. No model downloads; the core is numpy-only with optional scipy. If the animation is your first stop, open workflows/lased_animate.json, point the loader at any image, set the seed coordinates inside a color region, and queue.