Shima ReBNDLer (Latent Maker)

Shima ReBNDLer (Latent Maker) is the smallest of Shima's three packer nodes, but it packs the most metadata for its size. It takes a latent tensor and wraps it - along with an optional seed and the width/height it was made at - into a single latentmaker.bndl wire. One input in, one tidy bundle out.

It's the same trick as the Model Citizen packer: instead of running the latent, the seed, and the dimensions across the canvas as separate wires, you bundle them so a downstream Shima node gets everything in one shot. The nice part here is that the metadata actually travels with the latent. The Latent Maker's own bundle already carries s33d (the seed) and the dimensions, so when you pack an external latent you're giving it the same shape - downstream samplers and savers can read the seed and size off the bundle instead of you wiring them separately.

You'd reach for this when you're generating a latent outside Shima - say from a vanilla Empty Latent Image or a VAE encode - and you want to hand it to a Shima Panel Sampler or a DeBNDLer chain without losing the seed and size context.

How it works

Mechanically it's a dict with bndl_type: "latentmaker" plus the latent, s33d, width, and height keys. The s33d, width, and height inputs are optional, so you can pack a bare latent if you have nothing else to attach. Shima consumers check bndl_type, pull the latent, and read the rest if present.

The inputs and outputs that matter

• latent (LATENT) - the required payload.
• s33d (INT), width (INT), height (INT) - optional metadata; all three are forced-input style (connect wires, they're not typed in). In the source they default to None when left unwired.

Output is a single latentmaker.bndl. Feed it to Shima's Panel Sampler's latentmaker.bndl input, or to a DeBNDLer (Latent Maker) to get the pieces back.

How to install it

Part of 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 auto-clones ComfyUI-Impact-Pack and cg-use-everywhere on first install; nothing extra needed for this node.

Common issues & troubleshooting

The bundle has no seed and the sampler ignores it. That's by design - if s33d isn't wired it packs as None, and the consumer falls back to its own seed logic. Connect the seed input if you want the packed seed respected.

Nothing but the latent comes out downstream. If you expected width/height to appear as wires, you're thinking of the DeBNDLer, not the Packer. Packing is one-way; unpack with the matching DeBNDLer to expose the metadata as individual sockets.