IAMCCS Cine Shotboard Planner V5 V2V

The Shotboard Planner is the node that turns "I want a 20-second cinematic timeline" into an actual per-clip production plan, and V5 is the version that learned to do video-to-video. Its docstring says it plainly: it keeps the V4 timeline/editor contract, then adds a V2V mode layer - emitting the V4 timeline data plus V2V modelLayers, controlTracks, takes, and a router-compatible V2V payload. In plain terms: V4 planned a video you'd generate from scratch; V5 plans a video you're going to make from existing footage, clip by clip, and it writes the directions that the backend routers will later read.

The required inputs are the editorial spine. global_prompt is your one-sentence creative brief (the default is a surprisingly well-written "cinematic video-to-video production timeline..." that's a fine starting point). timeline_data is where the plan materializes - the tooltip calls it out as edited by the planner, and it emits V4 timeline data plus the V2V layers. duration_seconds, frame_rate, and the guide-policy trio (guide_policy, min_guide_gap_seconds, max_guides) control how aggressively the planner seeds control points along the timeline. ltx_round_mode (up_8n_plus_1 default) is the LTX frame-grid guarantee you'll recognize if you've ever had a clip come back 2 frames short.

Then the big one: v2v_mode. Six choices - ltx_v2v, ltx_loop, scail_single, scail_multi, wananimate_bg_locked, pose_transfer - and this is really the whole personality of V5. It's a single planner that can emit plans for four completely different generation backends. You pick the backend up front, the planner shapes the takes/control tracks for it, and downstream routers dispatch to the matching BusToLTX / BusToScail / BusToWanAnimate / BusToPoseTransfer converters. The other controls are mostly that choice, fleshed out:

• Source handling: source_video_path / source_image_path with trim_start_s / trim_end_s and frame_load_cap (241 default) tell the planner what footage you're working from. continuation_mode (source_video default) decides how each segment continues - off the source, off the last frame, off a guide frame, or a manual cut.
• Geometry: generation_width/height (1280x720 default) vs the reference image_width/height (768x432) - the first is what you generate, the second is how references get normalized.
• Segmentation: segment_seconds with segment_preset (5/10/15/20 sec, videoclip, monologue) and overlap_frames (9 default) control chunking, which is where long-form continuity is won or lost.
• Backend-specific toggles: pose_mode + dwpose_enabled/dwpose_strength for pose-controlled routes, scail_identity_mode and scail_output_stage for SCAIL, wan_background_lock / wan_character_mask_mode (SAM31 identity mask) for the Wan background-locked route, audio_vae_name (default ltx-2.3-22b-dev_audio_vae.safetensors) with device/dtype for the audio side.

Output is a single cine_linx carrying the whole plan. From there you hand it to IAMCCS_CineShotboardV5V2VRouter, which loads the media and splits everything into media/control/preview buses for the backend.

Installation: search IAMCCS in ComfyUI Manager, or clone https://github.com/IAMCCS/IAMCCS-nodes.git into custom_nodes and restart. This node is pure planning logic - no models, no pip deps. But it plans for a graph that needs real weight: an LTX 2.3 AV setup, or SCAIL, or WanAnimate, per your v2v_mode. The pack's SuperNodes requirements doc lists the exact LTX node set it expects (LTXVPreprocess, LTXVImgToVideoInplace, LTXVAudioVAEEncode, etc.), and the one external-pack requirement that bites is VideoHelperSuite (VHS) for the combine step.

The trap to plan around: it's easy to set v2v_mode, get a great-looking plan, and then wire the router into the wrong backend converter. The planner names the backend it planned for - respect it. Also remember render_scope (selected_clip / selected_range / all_clips / dirty_clips) is your iterative-render friend: draft one clip, check it, then widen the scope instead of re-rendering everything every time.