IAMCCS Cine Resolution Parity Translator

The smart way to work with expensive video models is to iterate cheap and render expensive: nail the plan at low resolution, then do the full-res pass when you're happy. The problem is that a shotboard plan isn't resolution-neutral - reference strengths, motion multipliers, and identity anchors were tuned for one resolution, and blindly re-running at 2× gives you subtly wrong results. IAMCCS Cine Resolution Parity Translator is the node that translates a proven low-res shotboard plan into a high-res run. It reads the plan, computes the scale, and adjusts the strength and motion parameters so the big render behaves like the small one did.

The docstring says it directly: "Translate a proven lower-resolution shotboard into a higher-resolution run." It's a quality-of-life node for people who iterate - and if you're rendering 200-frame LTX shots, you iterate a lot.

How it works

It reads the cine_linx plan bus and the target resolution already in it (cine_image_width/cine_image_height), then compares against the reference_width/reference_height you ran your test at (defaults 1024×576). From the scale factors it derives:

• Motion multiplier - motion_multiplier defaults to 0, meaning "auto": computed as 1 / linear_scale. A 2× render gets a motion multiplier that compensates for how resolution scaling changes apparent motion, so the high-res pass doesn't move twice as far as your test did.
• Identity multiplier - identity_multiplier (default 1.05) adjusts reference/identity strength for the higher-res run.
• Guide prefiltering - guide_prefilter (default on) with prefilter_method (bicubic, lanczos, bilinear, nearest) resamples your reference/guide images so they don't smear when scaled to the new resolution.

It rewrites the plan's visual segments with the translated strengths and image paths, then returns the same cine_linx with everything adjusted. enabled (default true) gives you a clean bypass.

The inputs that matter

• reference_width / reference_height - the resolution your test run used. Get this wrong and the "parity" is computed against the wrong baseline.
• motion_multiplier - leave at 0 for auto unless you have a reason; manual values (0–2) override the scale-derived one.
• identity_multiplier - 1.05 is a gentle bump; raise if high-res identity drift shows up, lower if the render looks over-anchored.

Outputs: cine_linx (translated) and report.

Install

In IAMCCS-nodes:

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

or Manager → search IAMCCS. README floor: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

Where people get burned

• Wrong baseline resolution. The whole node is a ratio between your test resolution and the plan's target. If reference_width/height don't match what you actually tested, the translation is nonsense and you'll be chasing differences that came from this node, not the model.
• It's a plan transformer, not a upscaler. It doesn't touch your latent or your sampler - it rewrites plan metadata. If the plan was empty to begin with, there's nothing to translate.
• Bypass mode is loud. With enabled: false it returns the plan untouched and the report says so - that's the intended way to A/B the effect.
• Auto motion math is approximate. The 1/linear_scale heuristic is a good default, not physics. For extreme resolution jumps (480p → 4K), hand-tune motion_multiplier.
• Zero search impressions, one-author tooling. Useful only if you actually iterate low-res-then-high-res with the Cine pipeline; for a one-shot workflow it's dead weight.