Vetted LoRA Loader

PseudoVettedLoRA Loader - yes, the display name is "Vetted LoRA Loader" and the class spells it PseudoVettedLoraLoader, with one lowercase "r", because of course it does - is the pack's curated style-LoRA slot. Where the regular LoRA Loader gives you every LoRA you've ever hoarded, this one offers seven named art styles:

• AlphonseMucha.safetensors, Brushstrokes.safetensors, ColoringBook.safetensors, DelicateWhimsy.safetensors, EldritchCharcoal.safetensors, PixelArt.safetensors, SketchIt.safetensors

That list is the whole pitch. In an architectural-rendering pipeline, the model structure and depth come from ControlNet and the geometry from Rhino; what's left to vary is the art direction. Material renderings, concept sketches, charcoal studies, graphic-novel takes - the pseudotools team curated a small style shelf so the workflow can switch moods without a trip to Civitai for every look. You pick a style, dial in how strong it should be, and the render reads differently without touching the geometry.

How it works

Standard LoRA machinery under the hood. You feed it:

• model_input (MODEL) - your checkpoint's model coming out of the Vetted Checkpoint Loader (or any other loader)
• model - the style LoRA dropdown (populated at startup from https://tools.pseudotools.com/api/models, same as the other vetted loaders)
• strength_model (FLOAT) - how hard the LoRA applies, default 1.0, range -100 to 100 in 0.01 steps
• record_id - pseudotools API bookkeeping, ignore it

Output is a single model (MODEL), the patched weights, which flows on to your sampler. Two details worth knowing from the source: a strength of exactly 0 short-circuits and returns your input model untouched - handy when you want the style slot present in a workflow but disabled - and the node caches loaded LoRA weights, so re-running the same style doesn't re-read the file from disk every time.

The strength question

A LoRA is a small low-rank adapter that patches a checkpoint's behavior without retraining it; strength is how much of that patch you apply. For style LoRAs the community default of 1.0 is a reasonable start, but style bleeding is the classic failure mode - at high strength the style overpowers the scene and every render looks like the same poster. If your materials and architecture start vanishing under a mush of brushstroke, dial the strength down before you blame the LoRA. And since each style file here is tuned by the same team against the same recommended checkpoints, 1.0 is more trustworthy than it is with random community downloads. Note the slider goes negative - that inverts the patch, which is usually an experiment, not a technique.

Gotchas

• Same family as every vetted loader: the dropdown needs the Pseudotools API reachable at startup, and the node doesn't download the files - the seven .safetensors files have to be in ComfyUI/models/loras under the exact dropdown names.
• Model must be a MODEL wire, not a widget. If you plug this straight after the Vetted Checkpoint Loader, you're golden. If you hand it a checkpoint that doesn't match the LoRA's base (an SD 1.5 style LoRA on an SDXL checkpoint), you'll get garbage or a load error - architecture lock-in is absolute, and this team ships SD 1.5 and SDXL styles, so pick the right slot.
• The spelling: search your node menu for "Vetted LoRA Loader" and it'll be there regardless of the class-name typo; the class name is just what you'll see in a raw workflow JSON.

Installation

Pack install, both routes:

# ComfyUI Manager: Custom Nodes Manager > search "Pseudocomfy" > Install

# or manually:
cd ComfyUI/custom_nodes
git clone https://github.com/Pseudotools/Pseudocomfy.git

Restart ComfyUI. Windows portable: install diffusers into the embedded Python first (README step one). Then drop the style LoRAs into models/loras.

Verdict

A style-switcher for people who do a lot of look-development and want seven curated moods on tap. If you already have favorite style LoRAs, the regular loader does the same job without the API dependency - this one earns its keep specifically inside the Pseudocomfy workflow, where a team wants everyone rendering with the same shelf of styles.