ReBNDLer (Shima Sampler)

The ReBNDLer (Shima Sampler) is the bundle packer for the output side of a generation. While the other ReBNDLers package what a sampler needs to start (model, prompt, latent), this one packages what a sampler produced - the decoded image, the sampled latent, and the s33d_used seed - into a single shimasampler.bndl wire.

Think of it as the "result envelope." A Shima Sampler already emits a shimasampler.bndl on its own, so you usually don't need to build one by hand. But when you sample with a non-Shima node and still want to speak the pack's bundle language downstream - feeding a panel, a post-processing island, or a save pipeline that expects the envelope - this ReBNDLer lets you construct it from any sampler's outputs.

How it works

Three optional inputs, one dict out:

{"bndl_type": "shimasampler",
 "image": image, "latent": latent,
 "s33d_used": s33d_used}

image is the decoded IMAGE, latent the raw LATENT, and s33d_used the INT seed that was actually used for the run (useful because with randomization the used seed differs from the one you typed). All optional - pack an image-only or latent-only envelope and the DeBNDLer returns None for the rest.

The inputs that matter

• image - if you're handing the result to an image pipeline.
• latent - if downstream wants to re-sample or img2img from the result.
• s33d_used - if you're logging or reproducing, wire the actual used seed so the bundle tells the truth.

Where it fits

Feed shimasampler.bndl into a DeBNDLer (Shima Sampler) to unpack back to image / latent / seed used sockets for standard nodes, or into pack nodes that consume the result envelope. It completes the BNDL story: loaders pack inputs, samplers pack outputs, and the pack/unpack nodes are the two ends of every bundle wire. The Shima docs explain why the pack ships explicit nodes like this rather than one universal bundle node - ComfyUI serialization would drop connections on reload with the dynamic version.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf.git Shima
pip install -r requirements.txt

Restart, then Shima/Routing → ReBNDLer (Shima Sampler). ComfyUI Manager: search "Shima". First boot auto-clones ComfyUI-Impact-Pack and cg-use-everywhere if missing (Use Everywhere is the island auto-linking backbone) - expected.

Common issues

The most common confusion is thinking this node samples. It doesn't - it only packs whatever is already on the wires. If the image inside the envelope is a pre-sampled source rather than your final render, that's on the wiring, not the packer. Also, s33d_used has no default: unwired, it ships as None, and a DeBNDLer consumer that assumes a seed will quietly produce seed-less logs. Wire it whenever reproducibility matters.