DeBNDLer (Master Prompt)

A prompt in Shima-land isn't just text. By the time it's been through a Master Prompt / Panel Master Prompt node you've got positive and negative conditioning and - on the multi-encoder models like Flux, SD3, and Hunyuan - separate conditioning for CLIP-L, CLIP-G, and T5. That's a lot of sockets. The bundle system collapses all of it onto one masterprompt.bndl wire, and this node is the unpacker that turns the bundle back into the individual CONDITIONING outputs a standard KSampler chain expects.

Think of it as the "explode" for prompt bundles. The bundle is a dict with pos, neg, clip_l, clip_g, t5, and the raw strings, and the DeBNDLer just pops each field onto its own typed output. It's the same context-object pattern the wider ecosystem uses to keep a graph readable - one fat wire in, clean typed sockets out - covered in the plumbing doc, so the mental model is: any time you see a BNDL input and want real wires, this (and its siblings) is the escape hatch.

Two inputs, same as the other DeBNDLers: bndl (the Master Prompt bundle) and allow_external_linking, which lets the node broadcast and receive outside its island group through Use Everywhere. Leave it false unless you're deliberately crossing island boundaries.

Outputs, in the order you'll actually use them:

• positive / negative (CONDITIONING) - wire these straight into a KSampler's positive and negative.
• CLIP_L_ONLY / CLIP_G_ONLY / T5_ONLY (CONDITIONING) - the individual encoder passes. Handy if you want to weight or mix encoders yourself (e.g., Flux workflows that put more weight on the T5 side), or feed a custom alignment node.
• pos_string / neg_string (STRING) - the plain text prompts, useful for metadata, logging, or a text-to-text node downstream.

If you're only using the base positive/negative pair, that's fine - the other outputs just sit unused.

The practical trap is the same one that hits the whole DeBNDL family: if positive comes out empty, the bundle was assembled upstream with nothing in it, so check the Master Prompt/ReBNDLer before you suspect this node. And don't go hunting for a single dynamic "DeBNDLer" - Shima moved from one dropdown-style unpacker (which dropped connections on workflow reload) to these fixed per-bundle nodes; old dynamic workflows may need a quick re-wire.

Install: ComfyUI Manager, search "Shima", or

cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf

then restart. No models or extra downloads for this node - the pack's first-run installer may pull in Impact Pack and cg-use-everywhere for the wider system, but this unpacker is standalone routing.