IAMCCS Cine Videoclip Maker Planner

Making a music video with local AI is a two-model job: an image model for the stills, then a video model to animate them. IAMCCS Cine Music Video Planner is the node that plans that pipeline for you, shot by shot. For each shot in the sequence it produces the image prompt (with your style modifier), the video prompt for animating that still, the beat segmentation, and the exact frame counts - so a Z-Image Turbo still can become an LTX-2.3 audio-i2v shot that lands on the beat.

The display name says "Videoclip Maker Planner," which is accurate: it's the planning layer, not the generation. You still wire up the image generator and the video backend; this node just makes sure every shot is sized, prompted, and timed consistently. It's the most "music-sequencer-inspired" node in the pack, and it shows.

How it works

You have three ways to size a shot:

• manual_shot_seconds (default 4) with manual_total_shots (default 8) - flat, even shots.
• audio_divided_by_total_shots - split the connected audio evenly across shots.
• max_manual_audio_slice - the longer of manual or audio-divided.

Or skip all that and feed sequence_json from an external music-video sequencer (it reads shots/sequence/segments/timeline arrays, with per-shot duration, prompt, action, frame counts). The planner then assembles each shot from two banks: image_prompt_bank (stills: "lead singer in a moody neon rehearsal room…") and video_action_bank (motion: "the performer breathes with the beat…"), cycling through them per shot. image_style_modifier gets appended to every image prompt; beats_per_shot (default 4) splits each shot's frames into sub-beats for the video segment math. The image_model_hint (z_image_turbo, flux_9b, …) and video_backend_hint (ltx23_i2v_audio, ltx23_i2v, wan22_i2v, custom) are metadata that keep your downstream choices consistent.

The outputs that matter

• image_prompt - the still prompt for this shot, ready for your image model.
• video_global_prompt and video_local_prompts - for the video backend (and PromptRelay-style segment conditioning).
• segment_lengths - the beat/sub-beat frame counts.
• frame_count_true vs frame_count_adjusted - the raw frame count and the 8n+1-rounded one LTX actually wants. If these differ, feed frame_count_adjusted to the latent.
• start_seconds, duration_seconds, total_shots - timeline positioning.
• music_linx - the plan bus for the rest of the pipeline.
• negative_prompt - passthrough.

Install

In IAMCCS-nodes:

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

or Manager → search IAMCCS. README floor: ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8. For the full intended stack you'll also want an image model (Z-Image Turbo is the hinted default), an LTX 2.3 audio-i2v setup, and - if you're making something that actually syncs to music - MelBand vocal conditioning per the pack's SuperNodes requirements doc.

Where people get burned

• It plans; it doesn't sync. The node sizes shots to the beat, but whether motion actually lands on the beat is up to the video model and your audio conditioning. Manage expectations: this gives you planned rhythm, not guaranteed sync.
• The prompt banks are the product. The default banks are good starting points, but the quality ceiling is your image_prompt_bank/video_action_bank. Thin banks → repetitive shots; the node cycles them, so keep at least as many entries as you have shots.
• shot_index is 0-based here (starts at 0), unlike the Multi-Gen Director (1-based). Inconsistent across the pack - double-check when looping.
• Frame rounding. Watch frame_count_adjusted vs true. The adjusted value is what LTX accepts; feeding the true count into an LTX latent is a classic "why is my latent the wrong length" bug.
• Zero search impressions, one-author tooling. It's a solid planner but it's opinionated about the whole stack - adopt the whole Cine/Videoclip pipeline or just steal its prompt-bank structure for your own workflow.