IAMCCS Cine Reference Board

IAMCCS_CineReferenceBoard is the load-your-references step of the IAMCCS Cine pipeline: you paste a list of image paths into one multiline box, and out come up to 50 individually-named IMAGE outputs (image_1 through image_50) plus a batch multi_output. Every image is loaded, resized to your target width/height, snapped to a multiple of 32, and normalized - so the reference board the model sees is uniform regardless of what you dropped in.

Why a dedicated node for what sounds like "just load some images"? Because in this pack the reference board feeds the FLF engine and the ShotPlanner, and those need a predictable batch: a known count, a known resolution, and per-image slots (image_1, image_2, …) that map 1:1 to timeline reference indices. The CineInfoV2 / CineAllInOneFLFEngine side then talks about "ref 3" and it means this board's image_3. If the board feeds 5 images but your timeline references 8, you get an immediate, understandable mismatch - which is a lot easier to debug than a wall of silently-wrong frames.

How it works

image_paths takes one path per line. resize_method decides how non-matching images fit the target: crop (fill the frame, cut the overflow), pad, keep proportion, or stretch. interpolation picks the resampling filter (lanczos is the default and the right call for stills; nearest only if you're pixel-testing). multiple_of (default 32) snaps dimensions up to the video-friendly multiple the latent pipeline wants. img_compression (0–100) re-compresses on load - a debug/bandwidth knob; leave it at 0 unless you're deliberately testing quality loss.

Outputs: multi_output (the whole batch as one IMAGE tensor) plus image_1 through image_50 (each a single IMAGE). If you have fewer than 50 paths, the trailing outputs are simply empty - no error.

Installing it

Part of the IAMCCS-nodes pack. ComfyUI Manager → search IAMCCS, or:

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

Restart ComfyUI; it's under IAMCCS/Cine/01 Reference Board. No pip requirements, no model downloads. README floor: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

Where people get burned

• Path format: this reads filesystem paths, not ComfyUI's input-folder relative names. A path ComfyUI's own Load Image would accept ("myrefs/face.png") won't resolve here unless it's reachable from the working directory - use absolute paths or paths relative to where ComfyUI runs.
• The output-count confusion: multi_output is the batch you feed a board-style consumer; image_N slots are for per-reference wiring. Mixing them up is the classic "my timeline references the wrong character" bug, and it's usually because someone wired a slot with the wrong index.
• resize_method stretch on mixed-aspect inputs is a fast way to get everyone looking like a funhouse mirror. crop or keep proportion with padding keeps faces sane - which matters when these frames become identity references for a character.