Load Image Sequence

Video work in ComfyUI usually starts with a folder full of numbered frames, and Load Image Sequence is the node that turns that folder into one batch. It reads in filename order, applies a range and a sampling strategy, resizes every frame to a common size, and hands you the whole run on a single IMAGE socket - the difference between this and Load Image Batch being that the batch version serves one frame per run, while this serves the run of frames a video pipeline wants. It defaults to 16 frames because a folder can hold thousands and a batch is one tensor in memory.

The sampling controls

• strategy - how num_frames are chosen from the range: uniform (evenly spaced), head, center, tail, random (seeded), or every_nth.
• nth - a thinning step applied before the strategy picks, so head with nth 2 takes the opening on alternate frames.
• num_frames - how many to keep. 0 takes every frame in range, up to the 4096 ceiling.
• seed - matters only for random, so a re-run grabs the same frames.

Here's the part that separates it from a dumb folder loader: only the chosen files are opened. Sample a 5,000-frame capture down to 16 uniform frames and the node reads sixteen files, not five thousand. The description makes the point deliberately - "sampling a long capture costs the frames you keep rather than all of them."

Size and the round-trip outputs

Frames of mixed sizes still have to stack, so resize_mode decides how each meets width/height: fit and pad (keeps the whole frame, pads the rest), fill and crop (fills and trims the overhang), stretch (distorts), crop or pad (never resamples). Leave width and height at 0 and the size comes from the first frame kept - with max_size capping the long edge so a folder of huge frames doesn't blow up memory.

The real gem is the paired outputs. images and count are the batch and its length; filenames lists every loaded name. But image_list and filename_list hand you the same frames one at a time - image_list runs a downstream node once per frame, and filename_list pairs with Image Save's filename_prefix (extension dropped) so every frame is written back under its own name. That's the folder round-trip the README advertises: load a sequence, process it, save each frame with the name it came in with.

Installing it

Part of WAS Node Suite:

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/was-node-suite-comfyui.git

or ComfyUI Manager → "WAS Node Suite v3", then restart. Requires ComfyUI 0.14.0+ and Python 3.10+; v3 installs no pip packages.

Where people trip

The memory surprise is the classic: wire a folder of 2,000 frames straight in at num_frames 0 and you've asked for one giant tensor - that's why the default is 16 and why max_size exists. Second, pattern matches inside the folder only, and filename order is lexical, so frame_10.png sorts before frame_2.png - pad your numbering (frame_02) or use start/end ranges rather than fighting the sort. And remember random is seeded per seed, not per run: leave the seed alone and "random" is deterministic, which is usually what you want for reproducible workflows.