SN1000 Lora Checklist

A LoRA is never just a file. It comes with a trigger word, a strength that actually works, and a prompt fragment that makes it sing - and ComfyUI stores exactly none of that. You keep it in your head, or in a text file next to the monitor. SN1000 Lora Checklist (class SN1000Checklist, from the SmartNode1000 pack) is the fix: a multi-select checklist where every item is its own prompt + LoRA pair, so ticking a box drops both in at once.

It's your per-workflow scratch space. Where its sibling SN1000 Lora Theme saves presets to shared JSON files, the checklist stores everything inside the node itself - the item list travels in the workflow file, so two checklists in one graph can hold completely different things. Make one for crying/smiling variants, one for which character appears, and flip them like switches.

How it works

At runtime the node reads items_data - a JSON string the UI manages for you. That's the source of truth; the visible prompt field is just the editing box for whichever item's settings are open. It then does three things:

• Composes the prompt: the prompts of all checked items get joined with the delimiter (default ", "), in order. A prompt_in from a previous node is prepended if connected.
• Applies the LoRAs: each checked item's LoRA stack is loaded with ComfyUI's own load_lora_for_models, same strength to model and clip. LoRAs are matched by filename against your models/loras folder; a file that isn't there is skipped and logged to the console rather than killing the run.
• Expands wildcards: __file__ and {a|b|c} syntax resolves using wildcard_seed, so the same seed gives the same picks.

One behavior worth knowing: if the same LoRA appears in two checked items, it's applied only once (the first occurrence). Applying it twice would multiply the strength and fight itself - the dedupe is a feature, not a bug.

The inputs that matter

Only two toggles really run this node:

• prompt_enabled - off drops the node's prompt and passes through prompt_in untouched. LoRAs still apply.
• loras_enabled - off skips every LoRA while each item keeps its own on/off state.

model and clip are optional. Connect them to apply LoRAs; leave them unconnected and the node becomes a pure prompt composer. That per-item on/off plus the two master toggles is what makes A/B testing a single click.

Outputs

Three sockets: MODEL, CLIP, and prompt (a STRING). Wire prompt into your CLIP text encoder, and model/clip down the normal sampler path.

Install

Same pack for all three SN1000 nodes. Via ComfyUI Manager, search SmartNode1000 and install. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/1000Pro-1997/SmartNode1000.git

Restart ComfyUI. No extra dependencies - nothing to pip install, no model downloads.

Where people get tripped up

• LoRAs silently skipped. If you checked items but didn't connect model/clip, the LoRAs just don't apply - there's no red error, just a [FLM] line in the console. Connect both and rerun.
• The checklist travels with the workflow. That's the design, but share a workflow file and the whole checklist comes along - including things you'd rather keep local.
• Strengths are clamped to -10..10 and item strengths are remembered per item, so a value that worked gets saved with the checklist.

Reach for it when you're tired of editing prompt text to try one style versus another - check the box, get the LoRA, done.