Nodes/PromptMasterLD/H3 Pack Bridge - LD
ComfyUI Node

H3 Pack Bridge - LD

H3 Pack Bridge Feeds Your Own H3 Graph

By Lora-Daddy·Created 6 days ago·Updated 2 days ago· 5
H3 Pack Bridge - LD
  • pack
  • clip
  • vae
  • audio_vae
  • positive
  • latent

When you want Studio's writing but your own sampler

H3 Master renders the whole pack itself, and it's great at it. But there are times you don't want the pack's render rail - you have a favourite H3 sampler setup, a turbo LoRA curve you trust, or a graph already wired around ComfyUI's stock MiniMax H3 encode nodes. H3 Pack Bridge (from PromptMasterLD, under LD / PromptMaster / Video rail) is the adapter that lets you keep H3 Studio's writing while sampling on your own terms.

It takes the Studio pack and turns it into the two things a normal H3 sampler graph wants: a positive CONDITIONING and a latent. From there you sample with whatever you like - the MiniMax H3 Turbo sampler, SamplerCustomAdvanced, your own schedule - and decode with your own VAEs. Studio writes; your graph renders.

How it decides what to encode

The pack carries Studio's mode, and the bridge reads it. No settings, no dropdowns - the tooltip's phrasing is "no settings: the pack decides," and it means it.

  • T2V / I2V → the bridge encodes using ImageToVideo conditioning.
  • REF (reference mode) → it encodes via ReferenceToVideo instead.

Only references that are actually attached get sent, and they're sent at the size Studio already gave them - there's no re-fitting on the bridge's side, which is exactly what you want, because a reference re-encoded at a different size than the panel prepared is a reference with different framing.

The three required inputs are pack, clip, and vae (the video VAE). The one that catches people: audio_vae. It's required for REF mode - and for video soundtracks - because the reference path carries audio too. On a pure T2V/I2V encode it's simply unused, which is the subtle way this node says "your mode doesn't need sound." Leave it unwired on a text-to-video job and nothing complains; forget it on a REF job and the encode can't carry the reference's audio.

Outputs are positive (CONDITIONING) and latent (LATENT) - wire them into your sampler's conditioning and latent inputs.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Lora-Daddy/PromptMasterLD
cd PromptMasterLD
pip install -r requirements.txt

Restart ComfyUI, or install via ComfyUI Manager → search "PromptMasterLD".

Where it fits, and where it doesn't

Reach for the bridge when you've outgrown the pack's opinionated rail and want Studio's output feeding your own carefully tuned H3 graph - that's the whole reason it exists, and it's the cleanest way to mix this author's writing with the wider H3 ecosystem's samplers. Don't reach for it if you want Studio's pack to just render: H3 Master does that with fewer moving parts. And if you're wondering why it exists alongside Master at all - the pack keeps both paths on purpose. Master is the one-queue movie maker; the bridge is the escape hatch for people whose real work happens downstream of a conditioning + latent pair.

CategoryLD/PromptMaster/Video rail

Inputs (4)

NameTypeDefaultDescription
packLDP_PACK
clipCLIP
vaeVAE
audio_vaeoptVAERequired for REF (and for video soundtracks). Unused on a pure T2V/I2V encode.

Outputs (2)

NameTypeDescription
positiveCONDITIONING
latentLATENT