IAMCCS Cine Shotboard V4 Backend

The planners are the screenwriting; this is the production company. IAMCCS Cine Shotboard V4 Backend is the execution half of the V4 pair: it takes the cine_linx plan from a V4 (or V3) planner, plus your loaded model and clip, and returns a conditioned model, video and audio latents, guide data, and motion guide data - everything a sampler needs, with all the LTX housekeeping done for you.

This is a big node under the hood (it's the largest source file in the cine family), and most of that complexity is about one thing: making the editorial timeline the source of truth. It resolves the timeline's duration, start frame, and frame rate; it handles the LTX 8n+1 length rounding; it builds audio for the exact duration you asked for and separately for the padded latent length the sampler needs; it encodes the audio latent through your audio_vae; and it packs image guides into guide_data and camera/motion timing into motion_guide_data. If a retake is active, it also resolves the source clip's in/out window so a stale canvas duration never retimes the source footage.

The inputs. Required are just model, clip, and cine_linx. The optional controls are where you get opinions:

• audio_vae - needed for the audio latent path. No audio VAE, no audio.
• use_custom_audio (default off) and use_custom_motion (default on) - whether the timeline's audio/motion segments override the defaults.
• inpaint_audio (default on) and override_audio (default off) - audio inpainting and full override switches.
• start_frame and duration_frames - manual overrides, only used if the timeline doesn't already define them.
• custom_width/custom_height, resize_method (default crop), divisible_by (32) - canvas control when you want to force resolution.
• epsilon (0.001) - the PromptRelay timing tolerance, passed through to relay encoding.
• timeline_data and global_prompt - fallback text if the cine_linx doesn't carry them.

The outputs. model (patched/relay-enabled), positive (conditioning with local prompts and guide timings baked in), video_latent, audio_latent, guide_data, motion_guide_data, frame_rate (FLOAT, resolved), and combined_audio (the mixed timeline audio as an AUDIO tensor). Wire the latents into your sampler, the audio latent into the audio VAE decode side, and keep guide_data/motion_guide_data for any guide-aware nodes downstream.

Install. Same pack: ComfyUI Manager (search "IAMCCS"), or git clone https://github.com/IAMCCS/IAMCCS-nodes into custom_nodes, restart. ComfyUI ≥ 0.3.0, Python ≥ 3.12. No bundled models, but this node genuinely needs the full LTX-2.3 audio+video stack - model, CLIP, video VAE, and audio VAE - or the audio/motion outputs won't have anything to work against.

Gotchas. Because this node re-resolves everything from the timeline, the backend widgets are explicitly a fallback - if your cine_linx already carries duration, prompt, and audio segments, the widgets get ignored, and the report-style debug dict (inside guide_data["iamccs_backend"]) will tell you which source won. Check that when results surprise you. And the usual: keep the pack updated - the V4 contract moves as the author ships, and a stale install is the most common cause of "planner says one thing, backend does another."