Shima MultiPipe XL Out

Shima MultiPipe XL Out is the SDXL big brother of the MultiPipe 1.5 Out node, and the difference is basically the refiner. Both are legacy pipe nodes from before Shima moved to its BNDL bundle system - you route a whole bunch of values down one PIPE_LINE wire on the In side and fan them back out on the Out side. The XL flavor is what you want when the old workflow you're salvaging was built for SDXL's two-stage pipeline: base model plus a refiner.

Let's be direct about when to use this. Fresh Shima builds use the Panel system and BNDL wires, and the pack itself calls these pipes legacy. This node exists for one job: unpacking an XL pipe from an old workflow (or from Shima.MultiPipeInXL) into individual sockets so you can rewire it into a modern graph. If you're not rescuing an old SDXL workflow, you probably don't need it.

How it works

The source is, again, a straight tuple unpack. MultiPipeInXL collects the SDXL payload into a pipe - image, mask, an sdxl_tuple, latent, model, vae, clip, positive and negative conditioning, then a second set for the refiner (refiner_model, refiner_vae, refiner_clip, refiner positive/negative), plus the four dimensions. The Out node unwraps all of it.

Two differences from the 1.5 version are worth knowing:

• It carries an sdxl_tuple - the SDXL-only encoding pair used by style/prompt tuples in the Efficiency ecosystem.
• It has no pipe passthrough output. The 1.5 Out keeps the pipe flowing for chaining; the XL Out consumes it entirely and hands you 18 individual outputs. Plan your chain accordingly.

The inputs and outputs that matter

Single required input: pipe. The outputs in order: image, mask, sdxl_tuple, latent, model, vae, clip, positive, negative, then the refiner block (refiner_model, refiner_vae, refiner_clip, refiner_positive, refiner_negative), then image_width, image_height, latent_width, latent_height.

For a typical SDXL workflow you'll wire model and vae into the sampler, positive/negative into the conditioner, and the refiner outputs into a second-pass sampler or ignore them entirely. The four INT dimension outputs are the shared source of truth for size - one place sets them, everything reads the same numbers.

How to install it

It ships in the Shima pack - install via ComfyUI Manager by searching "Shima", or manually:

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

then restart. The pack's install.py auto-clones ComfyUI-Impact-Pack and cg-use-everywhere as core dependencies, and its own requirements.txt is just aiohttp and requests. No models to download for this node specifically.

Common issues & troubleshooting

XL pipe into the 1.5 Out (or vice versa). The two pipes have different tuple lengths - XL carries 18 fields, 1.5 carries 12. Mismatch means a Python unpack error at execution. Match the In flavor to the Out flavor.

sdxl_tuple comes out as None. The XL In node defaults most optional inputs to 0, so if nothing was connected upstream the tuple slot is empty. That's normal for a partial workflow - you only get a real tuple if the source provided one.