Model Names

SeargeOutput4 is a fan-out node. It takes the Searge workflow's bundled "model names" and splits it back into individual named selectors - base checkpoint, refiner checkpoint, VAE, two upscalers, and a LoRA - each as its own typed output. Where other Searge nodes pack settings into one connection, this one unpacks the model choices so each loader can grab the exact filename it needs.

It's under Searge/_deprecated_/UI/Outputs, from the pack's older architecture. In that design, all the model filenames were selected together in one place and travelled as a single MODEL_NAMES bundle; this node is how the workflow pulled that bundle apart at the point where the actual checkpoint and upscaler loaders live.

How it works

Feed in the model_names bundle and the node emits it again (so you can keep chaining) plus seven broken-out values, each typed for the loader that consumes it. The base and refiner checkpoint loaders take base_model and refiner_model; the VAE loader takes vae_model; the upscale loaders take main_upscale_model and support_upscale_model; a LoRA loader takes lora_model. It's a demultiplexer - one wire in, the constituent filenames out.

This kind of node is invisible when you use the stock workflow, because it's already wired. It matters if you're tracing how Searge routes its model selections, or rebuilding part of the graph.

The inputs and outputs that matter

• model_names (input) - the bundled MODEL_NAMES from the workflow's selection stage.

Outputs, each a filename-typed value:

• model_names - the bundle passed through, for further chaining.
• base_model, refiner_model - the two SDXL checkpoints.
• vae_model - the VAE (point this at the fp16-fix VAE to avoid black images).
• main_upscale_model, support_upscale_model - the ESRGAN upscalers.
• lora_model - a LoRA filename.

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: python -m pip install opencv-python in ComfyUI's Python env (required), then cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, restart. Windows portable has an installer script that will also download the matching model files.

The names this node emits only resolve to real models if those files exist - SDXL base and refiner in models/checkpoints, the fixed VAE in models/vae, ESRGAN upscalers (like 4x-UltraSharp) in models/upscale_models.

Where people get burned

• Names that point at nothing. This node passes filenames, not the models themselves. If a loader downstream errors, the file named here isn't actually installed. Check the folder.
• Black images. If vae_model resolves to SDXL's embedded VAE rather than the fp16-fix one, you get the classic all-black result. Route the fixed VAE.
• Building new with it. It's deprecated. The current v4.x workflow handles model selection with its own nodes - you'd only meet this one inside an older graph.