Sick Ollie Node Pack

✨Sup✨
Unpack the zip into your Comfyui 'custom_nodes' folder or install from the git URL:
https://github.com/sickollie/SickOllie
Or search in the Nodes Manager - sick-ollie (v2 still validating there, so need to manually switch to that version if doing it this way)

After that, drop my main display images into ComfyUI for the basic setup, Studio for the new nodes, Classic got some updates as well for anyone the prefers the old nodes and also so old workflows dont break on import.

HUGE UPDATE, pasting the full README after this, but Shuffle randomization was huge for me, randomize would often repeat LoRAs or prompts where some LoRAs wouldn't get used at all. shuffle fixes that by randomly using every value once before repeating.

Built a brand new Image Metadata Core node that reads all available metadata in an image, with extra metadata compatibility from the rest of the Studio nodes. Immediately populates metadata on upload with easy copy buttons. Can also be linked into the workflow or even replace the preview node entirely.

Things like prompt index values are now dropdown menus that preview every log entry.

Folder and subfolder selection got massive overhauls. As well as the entirety of the full node pack is MUCH more intuitive now. Need wider usage testing on these to find any more bugs, so if you catch any please let me know!

Sick Ollie Node Pack v2.0.0

A ComfyUI node pack for LoRA testing, prompt-log automation, generation, metadata-aware output, image inspection, and repeatable batch workflows.

v2 ships two parallel node families:

• Classic — the familiar v1-style workflow with the original node IDs and explicit wiring, plus the final Classic feature upgrades.

• Studio — the redesigned CMYKG interface with streamlined sockets, Auto Context, integrated VAE decode, hierarchical browsers, and structured metadata.

Requirements

• ComfyUI

• rgthree-comfy — required by Loader Core's secondary-LoRA interface

• Python 3.10+

Install

1. Copy the ComfyUI-SickOllie folder into ComfyUI/custom_nodes/.

2. Make sure rgthree-comfy is installed.

3. Restart ComfyUI completely.

4. Hard-refresh the browser if the frontend was already open.

Node Menu

Sick Ollie
├── Classic
│   ├── Loader Core
│   ├── Prompt Core
│   ├── Generation Core
│   ├── Output Core
│   ├── Preview Core
│   ├── Persist + Display Resolved Prompt
│   └── Image Metadata Core
└── Studio
    ├── Loader Core
    ├── Prompt Core
    ├── Generation Core
    ├── Output Core
    ├── Preview Core
    ├── Persist + Display Resolved Prompt
    └── Image Metadata Core

What's New in v2

Classic

Classic keeps the original node IDs and v1-style wiring so older Sick Ollie workflows and workflow-bearing PNGs can continue to resolve to the node family they were built with.

• Shuffle — randomize without repeats. Every eligible value is used once before the pool refills in a new random order.

• Shuffle is available for the main LoRA, Prompt, Outfit A/B/C, and Scene streams.

• Prompt/outfit/scene logs detect the number of usable nonblank lines.

• Prompt/outfit/scene indexes use searchable index · line preview selectors instead of blind integer values.

• Outfit and Scene shuffle bags only advance when their tokens are actually used in the source prompt.

• Clean-name selection handles a wider range of delimiter-based LoRA filename structures.

• Loader Core exposes main_folder, the immediate parent folder of the active main LoRA.

• Generation Core accepts optional external positive/negative conditioning while retaining the normal prompt-text and empty-negative fallbacks.

• Resolved metadata records only prompt/outfit/scene/substitution values that actually participated in the final prompt.

Studio

Studio keeps the same production ideas but removes most of the bookkeeping from the visible graph.

• Custom black CMYKG dashboard UI.

• Hierarchical folder and log browsers.

• Separate folder scope and filtered master LoRA selection.

• Live eligible LoRA counts.

• Searchable prompt/outfit/scene line previews and no-repeat shuffle.

• Streamlined graph sockets.

• Auto Context carries resolved Loader, Prompt, and Generation data into Output Core without requiring dedicated metadata wires.

• Output Core can decode samples + vae directly, so a separate VAE Decode node is optional.

• Structured Sick Ollie metadata is written into saved images.

• Image Metadata Core can inspect those images immediately and also works as a standalone metadata utility.

Studio Quick Start

Minimum generation path:

Loader Core model
    → Generation Core model

Prompt Core final_prompt
    → Generation Core positive_text

Generation Core samples
    → Output Core samples

Generation Core vae
    → Output Core vae

Output Core images
    → Preview Core images

Optional identity substitution:

Loader Core clean_name
    → Prompt Core NAME value

Output Core's Auto Context automatically collects the resolved naming, LoRA, prompt-log, seed, model, and generation information needed for metadata and filename/subfolder recipes.

Output Core can also accept a decoded IMAGE. When both IMAGE and samples + vae are connected, IMAGE takes priority.

Shuffle

shuffle is randomize with complete coverage:

A, B, C, D, E
→ random order, each used once
→ refill
→ new random order

It is intended for LoRA and prompt-log testing where ordinary random selection can repeat the same value several times before the rest of the pool has been seen.

Studio Loader Core

• Choose a Folder scope without flattening the entire LoRA directory into one giant path list.

• Choose the main LoRA separately from the filtered master list.

• Main LoRA · N available shows the size of the current testing pool.

• Epoch filtering remains independent of folder navigation.

• Include subfolders defaults ON.

• New secondary LoRAs default OFF.

• Clean Name and Trigger are shown directly in the dashboard.

• main_folder returns the active LoRA's immediate parent folder.

Visible Studio outputs:

model
clean_name
main_trigger
main_folder

Studio Prompt Core

• Switch Prompt Source between Manual and Prompt Log.

• Prompt Log mode shows file, selected line, mode, preview index, and usable line count together.

• Outfit A, Outfit B, Outfit C, and Scene each have independent logs, modes, indexes, counts, and shuffle state.

• NAME and ITEM live under Substitutions.

• Prefix and Suffix live under Prompt Additions.

• Only the useful external text sockets remain visible: NAME value, ITEM value, Prefix text, and Suffix text.

• final_prompt is the only Studio graph output; the rest of the resolved state travels through structured context.

Studio Generation Core

Generation Core groups Encoding, Conditioning, Canvas, Sampling, and Seed controls into one dashboard.

Default Studio setup:

• preferred text encoder: qwen3vl_4b_fp8_scaled.safetensors when installed

• preferred VAE: qwen_image_vae.safetensors when installed

• resolution mode: preset

• aspect: 3:4 Portrait Standard

• megapixels: 1.0

• batch: 1

• sampler: Euler

• scheduler: beta

• steps: 9

• CFG: 1.0

• denoise: 1.0

• shift: 1.25

• seed: random every run

If the preferred encoder or VAE is not installed, the dropdown falls back to an available local file.

Optional positive and negative CONDITIONING sockets can override the internal conditioning path for image-edit/reference workflows.

Studio outputs:

samples
vae

Studio Output Core

Output Core handles naming, saving, metadata, and optional VAE decoding.

• Output root defaults to Project Folder Name.

• Build subfolders and filenames from literals plus resolved variables.

• Resolved Inputs shows the values Auto Context actually received; click values to copy them.

• Last Save shows the final relative output\... path.

• Prompt metadata, workflow metadata, and Civitai metadata can be embedded independently.

• Connect decoded images, or connect samples + vae directly.

Studio output:

images

Image Metadata Core

Image Metadata Core is part of the Studio-era metadata system but also works completely standalone.

Load an image by:

• clicking choose file to upload

• dragging a PNG/JPG/WebP onto the node

• wiring an IMAGE into the node for live workflow inspection

The node parses available metadata immediately. A workflow run is not required for a manually loaded image.

Focused copy actions:

• Seed

• Final Prompt

• Source Prompt

• Full Metadata Report

When Sick Ollie structured metadata is present, the inspector can recover generation settings, model/LoRA information, trigger data, prompt/outfit/scene files and indexes, substitutions, and other resolved inputs. It can also fall back to common Comfy workflow/prompt and parameters metadata when available.

Outputs include:

images
final_prompt
source_prompt
seed
generation_settings
models
resolved_inputs
full_report
metadata_json
has_metadata

Classic vs Studio

Use Classic when:

• opening an older Sick Ollie workflow or workflow-bearing PNG

• you prefer explicit graph wiring

• you want the original native-style node interaction model

Use Studio when:

• building a new workflow

• you want the dashboard interfaces and hierarchical browsers

• you want fewer bookkeeping wires

• you want Auto Context and integrated VAE decode

• you want the richest structured metadata workflow

Classic and Studio can coexist in the same ComfyUI installation.

Starter Content

The GitHub release includes optional starter content:

Starter Content/
├── input/SickOllieLogs/
│   ├── prompts/
│   ├── outfits/
│   └── scenes/
└── workflows/
    ├── Sick Nodes v2_Classic.json
    └── Sick Nodes v2_Studio.json

The example workflows reference model/LoRA selections from the test system. Replace any missing local files with your own installed equivalents.

See Starter Content/INSTALL STARTER CONTENT.txt for the copy locations.

What's New v1.1:
Major Loader and Prompt Core updates and overall improvements/bug fixes.

Automatic SafeTensor trigger detection. Built using LoRA file metadata from multiple trainer sources. Provides a Live copy button for the detected trigger, and also has a node output that can be linked elsewhere.

Prefix and Suffix boxes in Prompt Core. My workflow has the Autotrigger linked into the prefix text field, so the detected trigger automatically attaches to the front of any prompt, from the manual prompt box or prompt logs.

LoRA Epoch filtering (ex - automatically detect and test every epoch_8 from all your working character/style Loras, etc),

smarter clean names - instead of entering a bunch of 'remove this' values, you now select from a dropdown menu how much you want it to trim, and then you can output the resulting name to Prompt NAME replacement, output subfolder structure, etc.

three independent outfit log channels with editable token values that update in the copy buttons (put that token in your prompt and it will replace with a line from the attached log),
resolved-value copy tools
___________________________________________________

I got really sick of some annoying quirks in my workflow with like 70 nodes and reduced it all down to six brand new ones. FULL OF DOPE DROPDOWN MENUS

I present to you, Sick Ollie Node Pack.
A six-node complete Krea2 ComfyUI production system built for a much more cohesive workflow with a ton of quality of life improvements.
Does not include Face Detailer or Upscale nodes, but they can easily be linked in as normal if needed. I never really need to use them with Krea2.

For some dope starter prompt and outfit logs;
move the input folder found here:
custom_nodes\ComfyUI-SickOllie\Starter Content\input
to here:
ComfyUI\input

There's a lot in these nodes and this pack.
Included basic setup workflow here (also embedded in the showcase image here):
ComfyUI-SickOllie\Starter Content\workflows

I'm also including a PDF in the zip with a clickable index that explains each node here:
ComfyUI-SickOllie\docs\Sick-Ollie-Node-Pack-Field-Guide-v1.0.0.pdf
(also a new 1.1 pdf going over changes)

Filterable LoRAs that can easily loop one subfolder makes bulk testing LoRA Epochs SO much easier for me.

The Loader Core needs rgthree installed as well to work correctly for the secondary LoRA loader, but I feel like most people already have that. If not here's the github link for that.
https://github.com/rgthree/rgthree-comfy

But yeah, I am gonna paste some elaborate AI-generated description here full of bullshit about how cutting-edge this model is, but I actually put some work into these, and as simplified as they are, it's way too much to not explain somehow.

Features

• Loader Core

  • load your diffusion model

  • filter LoRAs to a specific folder or subfolder or epoch

  • loop through only that testing set

  • Choose from a dropdown menu clean LoRA filenames into usable names like becky

  • pass those names downstream for prompts, folders, and filenames

• Prompt Core

  • build prompts from reusable prompt logs

  • use outfit logs and scene logs

  • replace tokens like NAME, OUTFIT, SCENE, and ITEM

  • automate prompt variation without rebuilding prompts by hand

• Generation Core

  • compact Krea2-oriented generation workflow

  • built-in seed tools for randomizing, restoring, and reusing seeds

  • cleaner, more streamlined sampling setup

• Output Core

  • save clean, organized folders and filenames with dropdown menu variables

  • preserve prompt, seed, model, LoRA, and workflow metadata

  • improve Civitai resource detection and output traceability

• Preview Core

  • better in-workflow image previews.

  • contain, cover, fit, stretch, and blurred-background display modes

• Persist + Display Resolved Prompt

  • keep the full final prompt visible even after reloading an image later

  • preserve it when reloading the workflow from saved images

Included in the download

• the full 6-node pack

• a ready-to-use workflow

• 5 starter prompt logs

• 6 starter outfit logs

• an annotated workflow map

• a 76-page visual field guide

Requirements

• ComfyUI

• rgthree-comfy