IAMCCS Shotboard Planner V2V

IAMCCS_ShotboardPlannerV2V is the front door of the pack's multi-timeline video production path: an all-in-one planning panel that takes a source video, a source image, and a pile of generation settings, and packs them into a single cine_linx bundle that everything downstream reads. It's built around the pack's LTX-2.3 video-to-video backends, but it's fronted as "pick a backend profile, fill in the panel, get one clean output" - which is exactly the shape you want when a raw LTX graph has turned into node spaghetti.

The author ships detailed instructions for the LTX-2.3 v2v and audio+image-to-video flows on their Patreon (the README points there), so think of this node as the entry point to that whole production line rather than a standalone effect.

How it works

The panel is planning + orchestration, not rendering: it validates and normalizes your source, duration, trim, resolution, segmentation, and backend choices, then serializes everything into a cine_linx object (IAMCCS_SUPERNODE_LINX) that carries the full plan - including timeline_data, prompts, pose/audio VAE settings, and the chosen backend variant - to the shotboard editor and multigen backend nodes. You pick a backend_profile that says which engine handles the actual generation: LTX-2.3 infinite lipsync, LTX-2.3 loop anti-degrade, SCAIL-2 single/multi-person, WANAnimate extension, or Flux Klein pose transfer. SCAIL-2, if you're new to it, is Z.ai's motion-transfer fine-tune of WAN 2.1 that famously dropped the DWPose stick-figure stage; this node treats it as one of several selectable backends.

The inputs that matter

• source_video_path / source_image_path - the footage and the anchor frame the v2v works from.
• duration_seconds, fps, trim_start_s / trim_end_s, frame_load_cap - what portion of the source is used.
• generation_width / generation_height - output resolution (defaults 1280×720).
• segment_seconds, planning_mode, segment_preset, overlap_frames, ltx_round_mode - the same LTX-2 segmentation math as SegmentPlanner, folded in.
• backend_profile - the big one: which engine generates the frames.
• pose_mode (none/dwpose_openpose/source_pose_only/image_pose_transfer) and dwpose_strength - how strongly the source's pose drives the output.
• audio_vae_name - LTX-2.3's audio VAE (default ltx-2.3-22b-dev_audio_vae.safetensors), which you'll need on disk for audio-synced runs.
• global_prompt / negative_prompt - the shared prompt that applies across the timeline.
• vram_profile - normal_vram or low_vram for memory-constrained cards.

Output: a single cine_linx that wires into ShotboardVideoEditorV1 or the multigen adapter. That's it - one wire, the whole plan.

Installing

ComfyUI Manager → search "IAMCCS", or

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

Restart. This is where the pack stops being free of dependencies: the backend profiles expect a working LTX-2.3 (or SCAIL-2 / WANAnimate / Flux Klein) setup with native LTXV nodes, and LTX-2.3 audio runs want the audio VAE file in your models folder. The pack's SUPERNODES_REQUIREMENTS.md lists VideoHelperSuite (VHS) and MTB as needed for some combine paths - install them proactively.

Gotchas

The backend profiles are not interchangeable drop-ins; they have different external model requirements, and picking one you haven't installed will fail at the backend stage, not at this panel. frame_load_cap (default 241) silently bounds how much source footage is loaded - if your trim extends beyond it, later segments have nothing to work from. And keep vram_profile = low_vram for cards under ~16GB on the SCAIL-2 and LTX-2.3 profiles; the 22B LTX models do not fit casually.