DeBNDLer (Shima Sampler)

Shima's sampler story ends with a bundle, not a picture - the Panel Sampler wraps up the whole "DeBNDLer → Shima.Sampler → ReBNDLer" chain and emits a single shimasampler.bndl wire carrying the final image, the final latent, and the seed that was actually used. If you're following the bundle pattern through to the end, this DeBNDLer is the last stop: it unpacks that bundle into the plain IMAGE, LATENT, and INT outputs you need to save, preview, or feed a post-processing chain.

The mechanism is identical to its siblings, so don't overthink it. The bundle dict holds image, latent, and s33d_used; the unpacker reads the fields and hands each to its own typed socket. It's the same "one wire in, standard sockets out" pattern as every other DeBNDLer, and the same pattern the KB's plumbing doc describes across the ecosystem's context-bus nodes. Its only distinguishing feature is what it carries.

Inputs: bndl (the Shima Sampler bundle) and allow_external_linking (the island-scope toggle - off unless you want it broadcasting across islands via Use Everywhere).

Outputs, and this is where it gets nice:

• Image (IMAGE) - the decoded output, ready for a Save Image, a preview, an upscaler, or anything else that wants pixels.
• Latent (LATENT) - the final latent, so you can chain a second pass (img2img, latent upscale, refinement) without re-encoding.
• Seed Used (INT) - the seed that actually ran, not the one that was supposed to run. That's a quietly valuable output: the classic ComfyUI footgun is control_after_generate overwriting the seed you thought you locked in, and having the real value exposed as a wire means you can log it, name your file with it, or feed it to a "replicate this" node with confidence.

No gotchas unique to this node, just the family ones: it's an output node so it always executes, and if Image comes back None the bundle feeding it was assembled empty upstream. Also, same history as the others - Shima's old single dynamic DeBNDLer (type dropdown) dropped connections on workflow reload, so they replaced it with these discrete nodes. Old workflows may need a quick re-wire.

Install the pack via ComfyUI Manager (search "Shima") or

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

and restart. Nothing to download for this node; the first-run installer may pull in Impact Pack and cg-use-everywhere for the Island system, but unpacking a sampler bundle needs neither.