Unpack - LD

If Pack In - LD is where a workflow's pieces get stuffed into one cable, Unpack - LD is the far end where they come back out. Feed it a single pack - the LDP_PACK the H3 Studio panel and Pack In produce - and it fans the contents into a wall of typed sockets you can wire into any H3 or LTX graph. It's the "make it real" step in the pack's spine: the Studio panel writes a script, Pack In adds the models, and Unpack turns all of that into the individual wires a sampler rail actually consumes.

The outputs, grouped

There are a lot of them, but they read cleanly once you group them in your head:

• The script: prompt, report, mode, plus music_caption and music_lyrics when the shot is a music piece.
• The canvas: width, height, seconds, fps - wire these straight into your empty-latent / resolution logic. fps defaults to 24 when the pack doesn't say otherwise.
• The stills: start_frame and end_frame for image-to-video and reference shots, then image_1 through image_9 for reference images.
• The media: video_frames_1..3 each paired with video_audio_1..3 (same clip, matching the H3 Refs convention), then standalone audio_1..3.
• The pass-through: pack comes back out unchanged, so you can chain Unpack nodes or keep the bundle alive for something downstream.
• The still-image subgraph: image_prompt, image_width, image_height - appended at the end for FLUX-style still renders, defaulting to 2:3 at ~1MP (816×1224) when the pack doesn't specify.

The one required input, pack, is the whole surface. Everything else is output.

The design decision you'll feel

Empty slots don't error - they come out as a small black frame for images and a short silence for audio. That's deliberate: a graph can't hold an empty IMAGE tensor, so a missing start frame would otherwise break your wiring the moment you load the workflow. The price is that black frames and silent audio are easy to mistake for "something loaded" when nothing did. If a shot renders from a black first frame or a reference comes out blank, check report (the text output) before you suspect the sampler - it's the pack's way of telling you which slots are placeholders and why.

How it fits and how to install

You'll almost always see Unpack wired right after the Studio output or Pack In, and its outputs feeding the text encoder, the resolution logic, and the sampler rail. It does no math itself - it's a fan-out, which makes it the most boring node in the pack and therefore the easiest to trust.

cd ComfyUI/custom_nodes
git clone https://github.com/Lora-Daddy/PromptMasterLD
# restart ComfyUI

The whole pack shares one caveat worth repeating: it vendors ComfyUI-H3-Motion-Context (GPL) so the multishot rail installs in one click, which puts the distribution under the GPL - the README tells you exactly how to strip it if that matters to you. For the node itself there's not much to get wrong: give it a pack, read the report when something looks black or silent, and wire what you need.