Load + Resize Segment From Dir ??

If Load Images From Dir (Lite) is the "read frames from disk" node, Load + Resize Segment From Dir is the "read frames from disk and already have them at the right size" node. It's the two-step pipeline compressed into one: pull a segment of frames out of a directory, resize them all to your target resolution, and hand them to the sampler ready to go. For a long LTX-2 extension workflow where source frames arrive at arbitrary resolutions but the model wants a specific one, this removes a whole resize stage from the graph.

The default directory is iamccs_source_frames/source_video - the naming tells you this is the front door for source video frames you've pre-extracted, before they become latents.

How it works

It's a merger of the two patterns you've already seen in this pack. The selection half uses the same mode (all/from_start/from_end/range) with count, start_index, end_index, plus the linked overrides count_in/start_index_in/end_index_in. The resize half is exactly the KJ-style API from IAMCCS Image Resize Batch Safe:

• width / height - target size.
• upscale_method - nearest-exact, bilinear, area, bicubic, lanczos (defaults to lanczos, the sharpest of the five).
• keep_proportion - stretch/resize/pad/pad_edge/pad_edge_pixel/crop/pillarbox_blur/total_pixels (default crop).
• pad_color, crop_position, divisible_by - the usual finishing knobs.

And device (cpu/gpu, default cpu) lets you keep the resize off the GPU when you're tight on VRAM.

Outputs: images, plus width/height (the actual output size, post-divisible_by rounding) and batch_size - so downstream nodes can see exactly what they're getting without inspecting the tensor. There's also a report string summarizing the load and resize.

Why the combined node wins

Two reasons. First, fewer nodes and fewer wires - this pack is visibly designed for big graphs where every hop is noise. Second, and more subtly: doing the resize right after load means you resize once, not resize-then-reload. For a 500-frame source video, that's the difference between one pass and two.

Install

Same pack, same drill:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

or ComfyUI Manager → search IAMCCS, restart. No extra deps.

Gotchas

As with the loader, an empty directory errors out loudly - confirm the frames exist before blaming the node. And watch the resize modes: if your source is a mix of aspect ratios, stretch will distort some frames; crop (the default) is the safer pick for video, even if it does chop edges.