Shima ReBNDLer (Master Prompt)

Shima ReBNDLer (Master Prompt) is the conditioning packer in Shima's BNDL family. It takes your positive and negative conditioning and bundles them - along with optional per-encoder conditionings and the original prompt strings - into one masterprompt.bndl wire. Instead of dragging two conditioning noodles from your CLIP Text Encode across the whole graph, one wire carries the pair.

Why this is worth a node: in a plain ComfyUI workflow, positive and negative conditioning are two separate wires that must both reach every conditioning consumer. The moment you add a ControlNet or a second sampler, you're tracing four wires. Bundle them once and the pair stays together. Shima's Master Prompt node outputs this bundle natively; this Packer is how you get the same shape when your conditioning was made by vanilla CLIP Text Encode nodes or any other pack.

There's a real hidden benefit here: the optional pos_string and neg_string inputs let you attach the actual prompt text to the bundle. Downstream savers and loggers can then recover what was prompted, not just the encoded conditioning - handy when you want metadata without a text-tracking node.

How it works

Mechanically it's a dict tagged bndl_type: "masterprompt" holding pos (positive), neg (negative), optional clip_l, clip_g, t5 conditionings (for SDXL-style dual-encoder and Flux/T5 triple-encoder setups), and the two strings. Consumers read the type tag and pull what they need; the DeBNDLer (Master Prompt) unpacks all of it back into individual sockets, including CLIP-L/G/T5-only conditionings and the prompt strings.

The inputs and outputs that matter

• positive, negative (CONDITIONING) - required.
• clip_l, clip_g, t5 (CONDITIONING) - optional, only relevant if your conditioning pipeline produces separate encoder outputs.
• pos_string, neg_string (STRING) - optional; attach the raw prompt text for downstream metadata.

Output: one masterprompt.bndl. Wire it into Shima's Panel Sampler, Photo Remix, or any node accepting a masterprompt.bndl input.

How to install it

Ships in the Shima pack - ComfyUI Manager, search "Shima", or:

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

then restart. The pack's installer also pulls in ComfyUI-Impact-Pack and cg-use-everywhere as core deps; this node adds nothing on top.

Common issues & troubleshooting

Downstream sampler says "no conditioning." Both required inputs must be connected. The bundle is only built when positive and negative are present - leave one out and the pack is incomplete.

Multi-encoder conditioning ends up on the main slot. If you pack clip_l/clip_g/t5, they're stored separately from pos/neg. A consumer that only reads the main pair won't apply them. Use the DeBNDLer to confirm what actually rode along in the bundle.