Nodes/Shima/Shima Panel BNDLer
ComfyUI Node

Shima Panel BNDLer

One node that configures every panel in the workflow

By KDB-USJP·Created 6 months ago·Updated 6 months ago· 2
Shima Panel BNDLer
  • ====== MODEL CITIZEN ======
  • ====== MASTER PROMPT ======
  • ====== LATENT MAKER ======
  • ====== SHIMA SAMPLER ======
  • ====== CTRL ======
  • panelinputs.bndl
ckpt_name
vae_name
lora_1_name
lora_1_strength
lora_2_name
lora_2_strength
lora_3_name
lora_3_strength
filter_by_model_type
positive
negative
clip_l_weight
clip_g_weight
t5_weight
positive_l
positive_g
positive_t5
negative_l
negative_g
negative_t5
flux_guidance
lumina_sysprompt
width
height
batch_size
scale
aspect_ratio
orientation
s33d
steps
cfg
sampler_name
scheduler
denoise
start_at_step
end_at_step
randomize
add_noise
return_with_leftover_noise
vae_decode
upscale_enabled
upscale_method
upscale_factor
upscale_denoise
upscale_steps
upscale_cfg
control_type
strength
fit_method
bypass_preprocessing

Shima Panel BNDLer is the control tower for Shima's Panel system. Where each panel (Model Citizen, Master Prompt, Latent Maker, Sampler) configures itself through its own double-click modal, this node collects every panel setting into one place and broadcasts them as a single panelinputs.bndl. Connect that bundle to any panel's panelinputs.bndl input and the panel's settings get overridden by what's in the bundle.

If that sounds like overkill for a three-node workflow, it is. This node earns its keep when you're building a template or an "island" - a workflow you'll share or reuse - and you want one obvious place for whoever opens it to change everything. Instead of opening four modals and hunting for the CFG slider, they find one node that drives model, prompts, size, sampler, and control settings at once. It's Shima's answer to "how do I make this graph configurable without burying the settings."

The clever part is how the settings get in. Every input is an optional forceInput pin - you wire values in from primitives or other nodes rather than typing them. So you can drive the whole panel stack from a handful of Primitive nodes, or even programmatically. The node also merges a shima.commonparams bundle if you wire one, so shared values like seed, width, height, and model type automatically flow into the corresponding panel settings.

How it works

Mechanically it builds a panelinputs.bndl dict from every connected input, skipping the visual separator pins and anything left unwired. When a panel receives this bundle, its own panelinputs.bndl handling overrides its local widget values - checkpoint name, LoRA names and strengths, positive/negative prompts, encoder weights, latent width/height, sampler steps/CFG/sampler, upscale settings, and the control agent's control_type/strength/fit_method/bypass_preprocessing. The separators (the ====== MODEL CITIZEN ====== pins) are dummy wildcard inputs whose only job is to make the giant node readable.

The inputs and outputs that matter

There are no required inputs. The optional inputs are grouped: Model Citizen (ckpt, vae, three LoRAs), Master Prompt (prompts, encoder weights, flux guidance, lumina sysprompt), Latent Maker (width, height, batch, aspect ratio, orientation), Sampler (seed, steps, cfg, sampler, scheduler, denoise, upscale block), and Control (control_type, strength, fit_method, bypass_preprocessing). Output: panelinputs.bndl.

Practical tip: you don't have to use all of it. Wire in just the seed, steps, and CFG and leave the rest unwired - panels keep their local values for anything the bundle doesn't carry.

How to install it

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. No extra dependencies beyond the pack's core install.

Common issues & troubleshooting

Panel ignores the bundle. Only settings the bundle actually carries override. If a pin is unwired, the panel keeps its modal value. Wire the pin or accept the fallback - that's the designed behavior, not a bug.

The node is huge and intimidating. The long separator pins make it look worse than it is. Most people use maybe five inputs. Right-click the unused pins to disconnect any strays, and don't feel obligated to populate the whole thing.

CategoryShima/Panels

Inputs (55)

NameTypeDefaultDescription
====== MODEL CITIZEN ======opt*
ckpt_nameoptSTRING
vae_nameoptSTRING
lora_1_nameoptSTRING
lora_1_strengthoptFLOAT
lora_2_nameoptSTRING
lora_2_strengthoptFLOAT
lora_3_nameoptSTRING
lora_3_strengthoptFLOAT
filter_by_model_typeoptSTRING
====== MASTER PROMPT ======opt*
positiveoptSTRING
negativeoptSTRING
clip_l_weightoptFLOAT
clip_g_weightoptFLOAT
t5_weightoptFLOAT
positive_loptSTRING
positive_goptSTRING
positive_t5optSTRING
negative_loptSTRING
negative_goptSTRING
negative_t5optSTRING
flux_guidanceoptFLOAT
lumina_syspromptoptSTRING
====== LATENT MAKER ======opt*
widthoptINT
heightoptINT
batch_sizeoptINT
scaleoptFLOAT
aspect_ratiooptSTRING
orientationoptSTRING
====== SHIMA SAMPLER ======opt*
s33doptINT
stepsoptINT
cfgoptFLOAT
sampler_nameoptSTRING
scheduleroptSTRING
denoiseoptFLOAT
start_at_stepoptINT
end_at_stepoptINT
randomizeoptBOOLEAN
add_noiseoptBOOLEAN
return_with_leftover_noiseoptBOOLEAN
vae_decodeoptBOOLEAN
upscale_enabledoptBOOLEAN
upscale_methodoptSTRING
upscale_factoroptFLOAT
upscale_denoiseoptFLOAT
upscale_stepsoptINT
upscale_cfgoptFLOAT
====== CTRL ======opt*
control_typeoptCOMBO6 options: canny, depth, pose, lineart, scribble, color
strengthoptFLOAT
fit_methodoptCOMBO3 options: crop to fit, pad to fit, stretch
bypass_preprocessingoptBOOLEAN

Outputs (1)

NameTypeDescription
panelinputs.bndlBNDL