Shima Panel Model Citizen

Shima Panel Model Citizen is the loader at the heart of Shima's "panel" way of building workflows. It's the same unified model loader as the regular Model Citizen - checkpoint, CLIP, and VAE in one node, with three LoRA slots on top - but the presentation is completely different. The panel variant renders as a compact PCB-style chassis: no widget soup on the canvas, just a small board with one output wire. You pick your checkpoint and LoRAs by double-clicking the node and using its settings modal.

The point of the Panel system is legibility. A normal loader with checkpoints, VAE, and three LoRA stacks exposes maybe a dozen widgets and three or four outputs. When you're building a template workflow or an "island" you'll share, that's clutter. This node collapses all of it to a single modelcitizen.bndl output - one wire carries MODEL, CLIP, and VAE together, which is the BNDL "context bus" idea done right: the values that always travel together stay on one wire.

You'd reach for this as your standard loader if you're adopting Shima's Panel workflow at all. It's the first node in the pack's own getting-started pipeline: Panel Model Citizen → Panel Master Prompt → Panel Latent Maker → Panel Sampler.

How it works

Under the hood it's a subclass of the regular Model Citizen with the same load_stack logic - it calls the parent loader and just trims the return down to what the panel exposes. The filter_by_model_type widget scopes the checkpoint dropdown to a model family (SDXL, SD 1.5, SD3, Flux, and friends), so you don't scroll past 300 checkpoints that can't run with the rest of your graph. The frontend JS hides the native widgets and draws the chassis; the double-click modal is where the real controls live.

Two inputs worth knowing about beyond the obvious:

• panelinputs.bndl - an override bundle from a Shima Panel BNDLer node. Connect one and it can drive this loader's settings (checkpoint, LoRA names and strengths) externally, so a single Panel BNDLer can configure several panels at once.
• lora_stack - lets you feed in a LoRA stack from a Shima Lora Stack node instead of the built-in three slots.

The inputs and outputs that matter

Required: filter_by_model_type, ckpt_name, vae_name, and three lora_N_name + lora_N_strength pairs. Optional: lora_stack, shima.commonparams, use_commonparams, allow_external_linking, and panelinputs.bndl. Outputs are MODEL, CLIP, VAE as individual sockets and the combined modelcitizen.bndl - so you're not locked into the bundle; the individual outputs are right there if the next node doesn't speak BNDL.

How to install it

Ships in the Shima pack. ComfyUI Manager → search "Shima", or:

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

restart, and let install.py clone its two core deps (ComfyUI-Impact-Pack and cg-use-everywhere). No models to download - it loads whatever's already in your checkpoints and loras folders.

Common issues & troubleshooting

The modal won't open / the chassis looks broken. The Panel nodes are frontend-heavy, which makes them sensitive to ComfyUI frontend changes. The Nodes 2.0 (Vue) rewrite broke a lot of JS-rendered nodes across the ecosystem; if you're on that frontend and the panel renders wrong, switch back to the legacy LiteGraph canvas - this pack is built against it.

Checkpoint dropdown is empty. filter_by_model_type is set to a family you don't have checkpoints for. Set it to match what's in your models/checkpoints folder, or to the family your workflow actually needs.