Shima Multi Pass XL
The full SDXL payload — base, refiner, and everything between
- image
- mask
- latent
- model
- vae
- clip
- positive
- negative
- refiner model
- refiner clip
- refiner positive
- refiner negative
- sdxl tuple
- image
- mask
- latent
- model
- vae
- clip
- positive
- negative
- refiner model
- refiner clip
- refiner positive
- refiner negative
- sdxl tuple
Shima Multi Pass XL is the supersized member of the Passer family, built for SDXL's base-plus-refiner pipeline. It carries thirteen typed sockets through untouched: the usual nine (image, mask, latent, model, vae, clip, positive, negative, sdxl tuple) plus the refiner quartet - refiner model, refiner clip, refiner positive, refiner negative. Everything in, everything out, nothing computed.
The mechanism is identical to Multi Pass: execute walks a fixed list and returns whatever landed, in order, with None for anything unwired. Every input is optional, so you can pass a partial payload without erroring - which is the property that makes these nodes work as Use Everywhere broadcast targets at island boundaries. The pack's island model leans on passers as "stable airports," and this one is the airport with the refiner gate.
When you'd pick this over Multi Pass
The deciding factor is whether your SDXL setup runs a refiner stage. If it does, Multi Pass simply can't carry it - it has no refiner sockets, and the refiner model/clip/conditioning would have to travel by separate wires or a second passer. Multi Pass XL keeps the whole base+refiner payload in one place with typed sockets, so when the two-stage pipeline crosses an island boundary, nothing has to hitchhike. If you're on SD1.5 or a single-pass SDXL workflow, Multi Pass XL is overkill - the plain Multi Pass has everything you need and less clutter.
The interface
Thirteen inputs in, thirteen outputs out, one-to-one, all optional. Only wire what exists in your graph. The sdxl tuple slot is worth a special note: it's the SDXL conditioning tuple (width/height/crop info) that the base and refiner stages both consume, and this passer keeps it riding along with everything else so it's not the wire everyone forgets.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf
then restart, or ComfyUI Manager → search "Shima". No models, no extra dependencies.
The honest take
Same verdict as its smaller sibling, scaled up: outside the island system it's thirteen reroutes in a trench coat and you don't need it; inside, it's the correctly-typed anchor for a full two-stage pipeline. One practical complaint: thirteen sockets is a lot of surface area, and half-connected graphs are easy to misread - if you're not using most of them, the smaller Multi Pass or single-type passers will keep your canvas legible. And remember the big alternative: the MultiPipe XL In node packs the same payload into one PIPE_LINE wire if your priority is wire hygiene over individual socket access. Multi Pass XL is the middle ground - bundle, but keep every signal reachable.
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| imageopt | IMAGE | — | |
| maskopt | MASK | — | |
| latentopt | LATENT | — | |
| modelopt | MODEL | — | |
| vaeopt | VAE | — | |
| clipopt | CLIP | — | |
| positiveopt | CONDITIONING | — | |
| negativeopt | CONDITIONING | — | |
| refiner modelopt | MODEL | — | |
| refiner clipopt | CLIP | — | |
| refiner positiveopt | CONDITIONING | — | |
| refiner negativeopt | CONDITIONING | — | |
| sdxl tupleopt | SDXL_TUPLE | — |
Outputs (13)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| mask | MASK | — |
| latent | LATENT | — |
| model | MODEL | — |
| vae | VAE | — |
| clip | CLIP | — |
| positive | CONDITIONING | — |
| negative | CONDITIONING | — |
| refiner model | MODEL | — |
| refiner clip | CLIP | — |
| refiner positive | CONDITIONING | — |
| refiner negative | CONDITIONING | — |
| sdxl tuple | SDXL_TUPLE | — |