Custom nodes for ComfyUI by Winnougan
Nodes (48)
Point it at a directory and press Run once
Caption a folder of clips without hitting Queue 200 times
Two Krea 2 characters in one image without the blend β this is how
A before/after comparison grid that lives inside your graph
Bounding Boxes without Nodes 2.0 β draw the layout, hand the model a map
SillyTavern cards inside ComfyUI β chatting with the LLM you already loaded
Load a raw Diffusers folder and get MODEL + CLIP + VAE out
Why your 5-second clip is 107 frames β duration to frames, without the math node
Stop retyping 'preserve the composition' β template your edit prompts
Ref_strength, explained
Bake your reference images into one file instead of re-encoding them every run
The receiver half of wireless data β grab a Set node's value from anywhere
Grab any in-workflow image and get IMAGE + MASK, no disk detour
GIFs that don't look like mud β two-pass palette, and where Discord's 10MB bites
MiniMax H3 image editing in one node β conditioning, sampling, decode, done
The one-toggle uncensor for Krea 2, with LoRAs shipped in the pack
One uncensor stack, five Krea 2 models β the multi-model version
Reusable Krea 2 references β if you've done the two things Krea 2 demands first
The text-encode node that swaps in for Krea 2 Ostris Edit and reads baked references
Chat with a local LLM from inside ComfyUI β vision, audio, video included
A live, editable reader for whatever your local LLM just said
Point at any LoRA and it tells you who it is β live from Civitai
A LoRA stack editor that doesn't make you squint at text menus
Make one image match another's dimensions β without the resize math
MOSS-Image references, scaled the way Ostris's example actually scales them
Swap the Ostris MOSS encode node for one that reads your baked references
Six Load Image nodes? One node with a plus button instead
Bake Qwen Image 2.1 references β and why this one stores the picture too
Qwen Image 2.1 encode with baked references β and the extra latent output
H3 refmods applied in place
Bake a character β portraits and a voice β into one MiniMax H3 reference file
Baking chat characters into safetensors
Four characters, one H3 graph β loading baked refmods with separate face and voice strength
The node that frees your VRAM mid-graph β and how to wire it so it doesn't backfire
Two characters, no identity soup
One node for aspect ratios that survive snapping
Save video with the workflow inside it β codecs, CRF and the NVENC trap
Beam any value across your graph without the wire spaghetti
A no-frills text box that emits a clean STRING
Skip the CLIPTextEncode β ConditioningZeroOut two-node dance
Styled title banners that stay readable at any zoom
The Nougan V2 PNG Baker
The Nougan Workflow PNG Baker
Tuning the segment boundaries kijai's node draws
Give a video different prompts for different moments
Drag your scene changes instead of typing frame numbers
The Nougan Regional Character LoRA
The node that gets .txt files into your dataset folder
π Winnougan (Nougan's Nodes)
<img width="1536" height="1024" alt="Nougan Nodes" src="https://github.com/user-attachments/assets/e292034a-5854-43a4-9141-705c170dfd3f" />Custom nodes for ComfyUI
Twenty-five nodes across nine families β loaders, LoRA tooling, masking, layout boxes, smart image saving, dataset captioning, remote data, temporal video prompts, and a local LLM bridge β every one with its own themed, live UI.
</div>β¨ Overview
- π Loaders β a folder-aware Diffusers Loader and the one-toggle Krea 2 Β· Uncensored stage with three bundled LoRAs.
- π LoRA tooling β a themed LoRA stack editor with favourites, folder filters and a randomizer, plus the Lora Inspector that identifies any LoRA live from Civitai.
- π¨ Image & masks β Get Image for in-graph grabs, a full Mask Editor and Mask Composite.
- π² Layout β Bounding Boxes to draw a layout on a canvas (with layers and a right-click menu), plus a Layout LLM Prompt that lets a local LLM draw the boxes for you.
- π Smart image saving β the Image Converter / Saver saves or converts to PNG, WebP, JPEG and more, and Smart Squeeze finds the smallest file that still looks right.
- π Dataset captioning β folder batch loaders and a caption saver that names each
.txtafter its image, with whole-folder auto-run. - π‘ Remote data β Set / Get nodes pass data across your graph without wires.
- π Characters & video β Regional Character LoRA for two identities without blend, and Prompt Relay for per-segment prompts in LTX video.
- π§ Local LLMs β the LM Studio Bridge streams chat, vision, audio and video through any model in LM Studio, with a live on-node console.
Every node ships a custom frontend UI. The pack also includes a global Tab-cycler quality-of-life extension that works on every node in your graph.
π¦ Node index
| Node | What it does | Docs |
|---|---|---|
| Nougan Diffusers Loader π | Loads a Diffusers folder/file β MODEL Β· CLIP Β· VAE, auto-precision, optional Sage/Flash attention | core |
| Nougan Krea 2 Β· Uncensored π | Bakes in 1β3 bundled uncensor LoRAs with per-LoRA toggle + uncapped strength | core |
| Nougan Get Image πΌοΈ | Grabs an in-workflow image β IMAGE + MASK, live previews | core |
| Nougan Text Encode + Zero Neg π― | Encodes positive and builds the negative (zero-out / empty-string) in one node | core |
| Nougan Text Box π | Reusable prompt card emitting a STRING, with counts + copy/clear | core |
| Nougan Title Font π | Zoom-proof styled title banner with glow, animations, gradient presets, web link | core |
| Nougan Lora Loader π | Themed LoRA stack editor: chooser, β favourites, folder filter, π² randomizer | core |
| Nougan Lora Loader (Multi) π | Same editor, sharing one stack across up to 5 models | core |
| Nougan Lora Inspector π | Point at a LoRA β trigger words, base model, samples, stats, live from Civitai | docs |
| Nougan Mask Editor π¨ | Paint and edit masks on-node | core |
| Nougan Mask Composite π | Combine masks for edit-then-composite workflows | core |
| Nougan Bounding Boxes π² | Draw layout boxes on a canvas β preview, BOUNDING_BOX and prompt elements. Layers, snap, right-click menu, paste/import JSON | docs |
| Nougan Layout LLM Prompt π§ | Writes the instructions that make an LLM answer with box JSON, so a local LLM can draw your layout | docs |
| Nougan Image Converter / Saver π | Save or convert to PNG Β· WebP Β· JPEG Β· AVIF Β· TIFF Β· BMP Β· GIF. Smart Squeeze + a 5β10 quality scale find the smallest file that still looks right | docs |
| Nougan Load Image Batch From Folder π | One image per run from a folder, with live preview and whole-folder auto-run | docs |
| Nougan Load Video Batch From Folder ποΈ | The video counterpart of the folder loader | docs |
| Nougan Save Caption (Match Image/Video Name) πΎ | Saves a caption as a .txt named after the file just loaded | docs |
| Nougan Set (Remote) π‘ | Sends data under a key, no wire needed | docs |
| Nougan Get (Remote) π₯ | Receives data by key from anywhere in the graph | docs |
| Nougan Regional Character LoRA π | Two character LoRAs, one image, each masked to its own region β no blend | docs |
| Nougan Prompt Relay Encode π¬ | Per-segment temporal prompts for LTX Β· LTX2 Β· LTXAV | docs |
| Nougan Prompt Relay Timeline ποΈ | Same pipeline + a draggable WYSIWYG timeline editor | docs |
| Nougan Prompt Relay Options βοΈ | Per-stream epsilon / strength / window tuning | docs |
| Nougan LM Studio π§ | Streaming LLM chat + vision + audio + video via LM Studio, live DOM console | docs |
| Nougan LM Studio Prompt Box π¬ | Live on-node reader/editor for LLM output | docs |
π§ Which node do I need?
| I want toβ¦ | Use | |---|---| | Load a Diffusers model folder | Diffusers Loader π | | Uncensor Krea 2 in one toggle | Krea 2 Β· Uncensored π | | Stack style/character LoRAs with favourites + a randomizer | Lora Loader π | | Find out what an unknown LoRA actually is | Lora Inspector π | | Encode positive + negative without the two-node dance | Text Encode + Zero Neg π― | | Place subjects and text exactly where I want them | Bounding Boxes π² | | Let a local LLM plan the layout for me | Layout LLM Prompt π§ | | Save images as small as possible without visible loss | Image Converter / Saver π | | Convert a whole folder of images to WebP or PNG | Image Converter / Saver π | | Caption a whole dataset folder in one run | Dataset Captioning π | | Pass data across the graph without wires | Set / Get π‘ | | Put two characters in one image without identity blend | Regional Character LoRA π | | Give different video segments different prompts | Prompt Relay π¬ | | Chat with / analyze images via a local LLM | LM Studio Bridge π§ |
β‘ Install
cd ComfyUI/custom_nodes
git clone <your-repo-url>
# restart ComfyUI
Each family announces itself on boot β and each loads in its own try/except, so a problem in one can never take down the others:
[Nougan] β
Lora Inspector loaded.
[Nougan] β
Bounding Boxes loaded.
[Nougan] β
Image Converter loaded.
[Nougan] β
Prompt Relay loaded (3 nodes).
[Nougan] β
LM Studio Bridge loaded (2 nodes).
Or install via ComfyUI Manager β search Nougan.
Optional: pip install pyoxipng adds an extra lossless pass to PNG saving in the Image Converter.
π Documentation
Every node family has its own page in docs/:
| Page | Covers | |---|---| | Core nodes | Diffusers Loader Β· Krea 2 Β· Get Image Β· Text Encode + Zero Neg Β· Text Box Β· Title Font Β· Lora Loaders Β· Mask Editor Β· Mask Composite | | Remote Data Nodes | Wireless data passing, auto-color sync, dynamic key management | | Lora Inspector | Civitai lookup, caching, HTTP endpoint | | Regional Character LoRA | Token-space masking, regions, recon tool | | Prompt Relay | Timeline editor, epsilon tuning, technical deep dives | | LM Studio Bridge | Dev-server setup, vision/audio/video, Krea 2 prompt director | | Dataset Captioning | Folder Batch Loader Β· Save Caption (Match Image Name) Β· whole-folder auto-run Β· LoRA dataset prep | | Bounding Boxes | Box editor Β· layers and right-click menu Β· LLM layout prompt Β· MOSS-Image layouts | | Image Converter / Saver | Save and convert Β· Smart Squeeze Β· quality scale Β· lossless PNG/WebP Β· batch convert a folder |
π License
MIT β take it, ship it, remix it.
<div align="center">
Nougan's Nodes Β· if something here saves you a render, β the repo.
</div>