IAMCCS Motion Guide Bridge

LTX-2's motion control has a vocabulary problem: the model understands motion via guide frames and attention, but the plans you write - camera pushes, subject paths, shot timing - live in JSON. IAMCCS_MotionGuideBridge is the translator. You feed it a motion-sketch JSON and it produces the MOTION_GUIDE_DATA plus a render plan that the rest of the Shotboard V4 chain (renderer, IC-LoRA apply) actually consumes.

It's one of the Cine/Shotboard V4 nodes from IAMCCS-nodes, the production-video tooling by IAMCCS. These nodes form a chain - sketch → bridge → renderer → apply - and this is the node where the human-readable plan becomes machine-readable guidance.

How it works

The motion_sketch_json describes the intended motion: what moves, when, over how many frames. The bridge parses it and emits two things. First, motion_guide_data - the typed guide object that carries per-shot motion intent downstream. Second, motion_render_plan_json - a concrete render plan the MotionSketchRenderer can turn into control frames.

Two default strengths get applied across the board: default_video_strength (0.75) for how strongly the guide drives the video content, and default_attention_strength (0.65) for the attention-side control. When a shot in the sketch doesn't specify its own values, these are the fallbacks - so they're the dials you tune first.

The bridge also carries the cine_linx (the pack's IAMCCS_SUPERNODE_LINX data contract) through, so it can slot into a supernode chain and pass context along. control_video_path is the interesting optional: if you've already rendered a control video (say from the sketch renderer), you point the bridge at it and it activates an actual video segment; leave it empty and the bridge emits a render plan with no active segment yet - a planning-only run.

Inputs and outputs

• motion_sketch_json (STRING, default {}) - the motion sketch plan.
• default_video_strength (FLOAT, 0.75) / default_attention_strength (FLOAT, 0.65) - fallback strengths.
• resample_mode (nearest/linear) and resize_method (crop/pad/stretch to fit/maintain aspect ratio) - how guide frames get matched to the target resolution.
• cine_linx (optional) - chain context in, chain context out.
• control_video_path (STRING, optional) - pre-rendered control video to activate.

Outputs: cine_linx, motion_guide_data, motion_render_plan_json, report (STRING).

How to install

Part of IAMCCS-nodes. ComfyUI Manager search "IAMCCS", or:

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

Restart. No model downloads - the LTX-2 model, VAE, and IC-LoRAs are separate. Pack floors: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

Gotchas

This is a coordinator node, not a magic wand: feed it {} and it produces an empty plan - the report output is how you tell whether the sketch parsed into real guidance. The strength defaults are sane but generic; per-shot values in the sketch override them, so tune the sketch first. And the resize settings (crop vs stretch to fit) change what the renderer actually draws - "stretch to fit" will distort the guide motion, which is usually not what you want for a shot you bothered to sketch.