Pack In - LD

Every pack that grew its own "context bus" eventually invents a wire that carries a whole workflow's state in one object instead of a dozen cables. PromptMasterLD's version is the pack - an LDP_PACK - and Pack In - LD is the node that fills it. It's the assembled "here's everything the sampler needs" moment in the pack's spine of H3 Studio → Unpack / Pack In → your H3 or LTX graph, and its job description is one line: seconds plus your model rails, packed, so the render side can pull what it needs.

What it does

The H3 Studio panel writes a pack that carries the script, the canvas size, and the references - the creative side. What it doesn't carry is your actual models. Pack In is where you attach the hardware: it takes the pieces of a real render and merges them into the pack as a dictionary. What comes out is a single pack output that already knows everything - pass it onward and downstream nodes never have to guess which checkpoint, VAE or sampler you meant.

The mechanism is deliberately non-destructive. It reads any pack_in you feed it, copies the fields it has, and only overwrites what this call actually supplies - so a pack already carrying a model keeps it if you only add a sampler. seconds is the one value you'll reach for: a FLOAT override (1–60) that beats the seconds dial set on the Studio node, handy when you want a quick 3-second proof that reuses a saved 8-second script.

The inputs that matter

Most of the optional sockets are just "wire your loaders here":

• model, clip, video_vae, audio_vae - the model rail. Video VAEs and audio VAEs both get slots because H3 and LTX-2-generation models keep them separate.
• sampler and sigmas - optional; leave empty and let the downstream sampler decide.
• mod_name - a RefMod's exact name as a string (wire RefMod Picker - LD's output straight in). Loaded internally at strength 1.0 via the ComfyUI-MiniMaxH3Mod pack's own code, so a single mod needs no loader node on the canvas. Handy, but note the subtlety: mods and mod_name accumulate rather than replace, so you can stack a pre-loaded bundle with a picked one.

One output, pack, which then feeds Unpack - LD or an H3 sampler directly. There's a quirk worth knowing: the pack assumes 24 fps unless told otherwise - fps is implied, not carried.

Install and reality check

Same install as every node in the pack:

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

The honest notes: the mods socket is typed H3_REF_MODS, but the pack declares it as a string type so it loads fine even without the ComfyUI-MiniMaxH3Mod pack installed - the socket just won't connect to anything, which can silently do nothing if you expected ref mods to flow. And one licensing flag for the whole pack: it vendors ComfyUI-H3-Motion-Context (GPL) to make multishot work in one install, which drags the distribution under the GPL. If that matters to you, the README tells you exactly which folder to delete and how to install that dependency separately. For everyone else, Pack In is the boring, reliable junction node that makes the pack bus actually work - you'll notice it only by noticing the graph is far less spaghetti than it should be.