IAMCCS Cine Shotboard Planner V3

Shotboard Planner V3 is where the IAMCCS shotboard stops being "a list of image guide rows" and becomes a real timeline. The docstring calls it an editor-grade timeline: the timeline_data JSON now carries segments, rows, and audioSegments - so your plan can describe not just where images land in time, but where audio lands too. That's the leap from the Pro generation of nodes. It's built for the LTX-2.3 world where video and audio are generated together, and the whole plan still rides through the pack's single cine_linx contract.

It also inherits something new: an optional cine_linx input. V3 can chain - you can feed it an upstream planner's cine_linx and merge that plan's helper layers into your output instead of starting from zero. That's how you compose a complex timeline from smaller pieces.

The inputs that matter. Most are carried over from Pro (global_prompt, timeline_data, frame_rate, default_force, promptrelay_epsilon, ltx_round_mode, image controls), but V3 changes the defaults for its era:

• duration_seconds - default jumped to 20s. LTX-2.3 can genuinely hold a 20-second window, and V3 is aimed at that.
• guide_policy - default flipped to every_checked_row. V3 trusts you to check only the rows you want; safe_core_guides and prompt_only are still there.
• min_guide_gap_seconds - default 0, and max_guides default 50. In other words, V3 is willing to fire a guide on essentially every checked row across a long timeline. That's a lot of injected frames - it's powerful and it's also exactly where you can overshoot. If the generation starts feeling over-constrained or the model locks onto guide images, back off to safe_core_guides.

timeline_data here is specifically JSON with segments, rows and audioSegments - this is not the flat per-line format of the earlier planners, so an old timeline_data string may not parse the same way.

Output. The single cine_linx socket, carrying the V3 timeline (visual segments + multi-track audio metadata) to the backend. The pack describes it as "inspired by LTX-style guide data while remaining fully IAMCCS-owned" - meaning downstream backends that speak cine_linx (like the V4 backend) can consume it.

Install. The whole pack installs together: ComfyUI Manager (search "IAMCCS"), or:

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

then restart. ComfyUI ≥ 0.3.0, Python ≥ 3.12. No bundled models - you supply the LTX-2.3 model, video VAE, and audio VAE. The audio path in particular wants a working Audio VAE; without it, the audioSegments in your plan have nothing to be encoded against.

Gotchas. V3 is newer, so treat its 20s/default-everything defaults as starting points, not gospel - a 20-second multi-guide generation is a long, expensive run to discover your settings are wrong. Validate a short segment first. And the house rules: keep the pack updated (the author ships fast and stale installs cause weird validation), don't duplicate the IAMCCS-nodes folder in custom_nodes (double-registered routes and startup failures), and remember the LTX 8n+1 tail gets trimmed by the backend, not by you.