Magic Box Pipeline Start

If you opened the Searge SDXL EVOLVED workflow, went looking for the KSampler and the CLIP encoders, and found a wall of little grey boxes wired into one fat cable instead, this node is where that cable begins. SeargePipelineStart (it shows up as "Magic Box Pipeline Start") is the first stage of Searge's "Magic Box" - the pipeline system that runs the whole v4.x workflow through a single data stream rather than the usual spaghetti of explicit wires.

Some context, because it explains the whole design. Searge SDXL was one of the first complete all-in-one SDXL workflows for ComfyUI, back when SDXL 1.0 had just dropped in mid-2023 and everyone was figuring out how to chain the base and refiner models. It packed txt2img, img2img, inpainting, revision, ControlNet and up to five LoRAs into one file. To keep that from becoming an unreadable graph, the author built the Magic Box: instead of wiring twenty nodes together by hand, you drop in stage nodes that each read from and write to one shared bundle of state. Start the bundle, pass it through the stages, terminate it at the end.

How it works

Think of the Magic Box as a conveyor belt. SeargePipelineStart puts an empty crate on the belt. Every downstream Searge node - prompt adapters, samplers, loaders - reaches into that crate, does its job, and puts the crate back. Because everything travels on one wire, you don't rewire the graph to change modes; you change settings inside the boxes and the same belt carries them.

The one thing you have to get right here is the version. The required input wf_version is an enum offering 4.0, 4.1, 4.2, 4.3 - it tags the data stream with which workflow generation you're running so the rest of the Magic Box behaves consistently. The README is blunt about this: always use the latest workflow JSON with the latest version of the custom nodes. A mismatch between the version you declare and the nodes you have installed is exactly the kind of thing that produces a cryptic mid-graph error.

The inputs and outputs that matter

There's really only one knob:

• wf_version (required enum) - pick the workflow version, 4.0 through 4.3. Match it to the JSON you loaded.
• data and additional_data (optional, SRG_DATA_STREAM) - two inlets for feeding in an existing stream, used when the workflow chains Magic Box segments together. In a stock single-pipeline setup you leave these empty; Start creates the stream from scratch.

The single output is data (SRG_DATA_STREAM) - the initialized crate. This is what you wire into the next Searge stage. It only speaks to other Searge Magic Box nodes; it's not a normal ComfyUI type you can hand to a KSampler.

How to install it

Easiest path is ComfyUI Manager: open Manager, search for SeargeSDXL (or "Searge SDXL"), install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/SeargeDP/SeargeSDXL.git

then restart ComfyUI. One dependency bites manual installers: the README says to run python -m pip install opencv-python in ComfyUI's Python environment at least once, or the pack fails to load. On Windows portable there's also an official installer script that grabs the nodes and the model files (SDXL base, refiner, the upscalers, ControlNet annotators) in one go.

Common issues

The recurring failure mode with the entire pack, this node included, is version drift. ComfyUI's core changes, and an older Searge checkout stops loading - the v4.3.1 and v4.3.2 releases exist purely to "make the workflow and custom node extension compatible with the latest changes in ComfyUI." If you get a Magic Box traceback after a ComfyUI update, git pull the pack (or update it in Manager) before you debug anything else. And keep the JSON and the nodes on the same version; the Magic Box assumes they agree.