Model Names

Model Names is the node where you pick which files the Searge workflow loads - the SDXL base checkpoint, the refiner, the VAE, a couple of upscale models, and a LoRA. It's basically the workflow's central "which models am I using" panel, gathered into one node instead of scattered across a dozen loaders.

It ships with SeargeSDXL, the base-plus-refiner SDXL workflow pack Searge put out at SDXL 1.0's launch in 2023. This node is in Searge/_deprecated_/UI/Inputs - the v4.x rewrite replaced the input-chain approach with a data stream - but conceptually it's still just a bank of file dropdowns.

How it works

Each field is an enum that ComfyUI populates from the files in your models folders. So base_model lists your checkpoints, vae_model lists your VAEs, and so on. You're not typing paths - you click the dropdown and choose. The node bundles those choices into a MODEL_NAMES output that the rest of the workflow uses to actually load the files at generation time. Centralizing model selection like this is the whole point: change your base checkpoint in one spot and the entire graph follows.

One thing you'll notice in a fresh brief: the choice lists show up empty. That's just because the dropdowns fill from your installed files - until you've downloaded the models the workflow expects, there's nothing to select.

The inputs and output

• base_model - your SDXL base checkpoint. The workflow ships expecting sd_xl_base_1.0_0.9vae.
• refiner_model - the SDXL refiner checkpoint.
• vae_model - pick the fp16-fix VAE here to avoid black images.
• main_upscale_model / support_upscale_model - your 4x ESRGAN upscalers (4x-UltraSharp, NMKD-Siax, etc.).
• lora_model - one LoRA to apply.
• model_settings (optional MODEL_SETTINGS) - feeds in extra settings.
• Output model_names (MODEL_NAMES) - the bundle the workflow loads from.

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, then restart. Manual installs need python -m pip install opencv-python in ComfyUI's Python environment first, or the README says the pack won't load.

This is the node that most directly needs the downloads. At minimum grab the SDXL base checkpoint (sd_xl_base_1.0_0.9vae, ~7 GB) and, if you want the two-stage pipeline, the refiner (~6 GB). The optional-but-recommended sdxl_vae fp16-fix (335 MB) prevents black-image bugs, and the README lists the upscalers and ControlNet files too. Drop each into the matching ComfyUI/models/<folder>, creating the folder if it doesn't exist.

Common issues

The most common one is a red error because a selected filename doesn't exist - this happens constantly when you load someone else's workflow JSON that references model names you don't have. Fix it by clicking each dropdown and reselecting your actual files, exactly as the README tells you to after downloading.

Black images almost always mean the wrong VAE - point vae_model at the fp16-fix VAE. And a candid note: the refiner slot exists because this workflow predates the community consensus that the SDXL refiner mostly isn't worth it. You can leave the refiner unused and run a single fine-tuned checkpoint, which is what most people do in 2026. As ever with Searge, keep the workflow JSON and the node version in sync.