ComfyUI Custom Extensions
Browse 8,530 ComfyUI extensions and 65,510 custom nodes.
Tile the latent, not the pixels: latent-native tiled decoding for NVIDIA PiD. Seam-free masters up to 244MP; size-picker dropdowns with render-tested presets for Krea2/qwen AND FLUX.2 Klein 4B/9B; avoids the past-envelope color collapse of single-shot decoding. No VAE re-encode round-trip.
A silly POC Video Player for ComfyUI
Pymunk is a easy-to-use pythonic 2d physics library that can be used whenever you need 2d rigid body physics from Python
A mixture of effects and quality of life nodes. Nodes: ImageGlitcher (gives an image a cool glitchy effect), ColorStylizer (highlights a single color in an image), QueryLocalLLM (queries a local LLM API though oobabooga), SDXLReslution (resolution picker for the standard SDXL resolutions, the complete list), SDXLResolutionSplit (splits the SDXL resolution into width and height).
Extract the most common colors from an image, up to any number. Convert colors to plain English names using various color naming systems.
Qwen Image 2.1: Turbo Prediction and Strix Halo packed attention for text-to-image and image editing. Start edits with full evaluations; edit prediction is opt-in.
Nodes related to [Spark-TTS](https://github.com/SparkAudio/Spark-TTS).
A tool for benchmarking ComfyUI performance.
A powerful extension for ComfyUI that enables adding notes to many node in your workflow. Keep track of important settings, reminders, and workflow documentation directly within your ComfyUI canvas.
A one-click collection of alternate connection styles for ComfyUI.
Perform a Fast Fourier Transform on the image, and then users can freely select the filtering range to filter the image. The main function is to remove the grid patterns on the image, and it can also perform high-pass filtering and low-pass filtering. The detailed workflow is shown in the figure below. The PNG file contains the ComfyUI workflow.The working principle is similar to the FFT filter in Photoshop.
A powerful all-in-one ComfyUI extension with comprehensive professional nodes for AI models, image/mask/SVG processing, video synthesis, and workflow management. Features modular architecture with node alignment tool and multi-platform API support.
Comfy UI node to prompt build for https://huggingface.co/cagliostrolab/animagine-xl-4.0 model
LIST and BATCH utilities which support: create, convert, get or slice items
A ComfyUI extension that makes your node connections delightfully droopy. (Disclaimer: despite what it may look like, this extension will not make your monitor taste like spaghetti.)
It Load Images with subfolders form arbitrary folders previous on node outputs lists- convinient selection via file browser
Powerful tools for image and text processing, including cross-platform workflow fixes, optimized resolution, dynamic text/image combos, and batch processing. Unlock seamless AI art creation and boost your productivity!
This is a simple node for creating prompts using a .csv file. I created this node as an easy way to output different prompts each time a workflow is run.
Custom nodes for ComfyUI that generate images and videos using Google's Gemini API, supporting text-to-image, image-to-image, text-to-video, and image-to-video generation
Advanced multimodal prompt generation nodes for ComfyUI with local GGUF models (Qwen-VL) and cloud API support for vision-based prompt enhancement.
Give each reference image a job, and every attribute a dial. Guide cards assign per-image roles (keep the same subject, suggest the visual style, copy lighting or pose, big shapes only, avoid copying text/logos) with per-card strength, timing, and user-defined recipes - and Concept Slider cards turn any attribute you can name (brightness, age, fog density) into a training-free -6..+6 dial derived from Krea 2 text encoding.
A comprehensive ComfyUI plugin suite that brings professional Photoshop-style layer effects and advanced compositing capabilities. Features modular architecture, 24 blend modes, and enhanced positioning system.
This is set of custom nodes for your ComfyUI local installation. It offers the very basic nodes that are missing in the official 'Vanilla' package. It is a research Node based project on Artificial Intelligence using ComfyUI visual editor. This repository also includes a set of workflows to test the nodes. NOTE:Renamed from 'ComfyUI-Compositing-Nodes-Pack'
A ComfyUI custom node for auditing 3D mesh files by rendering them with a headless renderer and displaying an interactive carousel of renders with detailed mesh statistics. Features 16 renders per execution (4 camera angles x 4 shading modes), an interactive image viewer, and an asset statistics panel with geometry and quality metrics. Supports Linux x64 and Windows x64 with Vulkan-capable GPUs.
Node:Prompt Worker. A text manipulation node for postprocessing of prompt.
ComfyUI nodes for adaptive FP8/BF16 mixed precision quantization of diffusion models with automatic fragility detection and outlier protection.
Turn your workflow into an easy-to-use web application without enabling Node2.0
Add unified control panels to your workflows and convert them to WebApp without using Node 2.0
Nodes:Inversed Euler Sampler, Mix Noise with Latent, Combine Latent Noise
Professional-grade frame manipulation tools for ComfyUI, providing advanced video editing capabilities with native IMAGE tensor support.
Official companion node for Image MetaHub. Saves A1111/Civitai-compatible parameters plus extended Image MetaHub metadata (workflow JSON, SHA256 model hashes, VRAM peak/total, GPU device, generation time, and steps/sec) for reproducibility and benchmarking.
The Flux Prompt Saver is set of simple nodes for saving images generated with Flux with A1111-style metadata.
Aspect ratio selector for ComfyUI based on [a/sd-webui-ar](https://github.com/alemelis/sd-webui-ar?tab=readme-ov-file).
Cached alternatives to ComfyUI's native MiniMax H3 FL2VA and Ref2VA conditioning nodes that avoid repeatedly loading the large Qwen3-VL text/vision encoder when the same H3 conditioning was already computed.
This project integrates [a/StyleShot](https://github.com/open-mmlab/StyleShot) functionality into ComfyUI, thanks to the foundational work by continue-revolution.
ComfyUI custom nodes to apply various latent travel techniques.
NODES:(Down)Load SAM2-Realtime Model, Sam2RealtimeSegmentation
The ultimate text processing suite for ComfyUI. Now features Batch Cleaning, LLM Parsing, Dynamic Wildcards, Image and Logic Tools for sophisticated automation.
this node contains a lot of small little helpers like switches, passers and selectors that i use a lot to build my workflows.
[WIP]Custom nodes for handling, inspecting, modifying and creating various model files.
Ultra fast frame interpolation using RIFE with TensorRT with Automatic downloading
NTSC/VHS analog tape effect for ComfyUI, with a live preview inside the node.
using InstantX's CSGO in comfyUI for style
Accelerate ComfyUI Nodes for Faster Image Generation, Ensuring Consistency Pre and Post-Acceleration, Ideal for Bulk Image Production.
This is a set of custom nodes for ComfyUI that replaces nodes like WanImageToVideo but using a Tiled VAE approach to reduce VRAM requirements.
A custom node for ComfyUI that generates images using Google's Gemini API, supporting both text-to-image and image-to-image generation.
Auto-MBW for ComfyUI loosely based on sdweb-auto-MBW. Nodes: auto merge block weighted
A ComfyUI node for setting image sizes based on the model being used.
Resolution calculator nodes for ComfyUI with model-specific constraints and optimal bucket resolutions
A very simple and easy to use node to automaticaaly create (square) image crops and masks using YoloV8. This can be very useful when using controlnet and ip adapters
These are just some nodes I wanted and couldn't find where anyone else had made them yet.
This node captures images one at a time from your webcam when you click generate. This is particular useful for img2img or controlnet workflows. NOTE:This node will take over your webcam, so if you have another program using it, you may need to close that program first. Likewise, you may need to close Comfyui or close the workflow to release the webcam.
H3-native flow-aligned and progressive-resolution regeneration for ComfyUI
A collection of nodes for batch processing texts and images in ComfyUI
ComfyUI Node to create Object-Oriented Prompts
ComfyUI nodes for QwenVL (Qwen2.5-VL / Qwen3-VL): image caption, OCR, boxes, video summary.
Steerable, resumable long-form MiniMax H3 video for ComfyUI
Various custom nodes for ComfyUI
Advanced real-time control for Z-Image models and Qwen3-4B Text Encoders. These nodes allow you to change your model's weights to boost or lower specific generative phases.
A local, single-user workflow library inside ComfyUI: save workflows as structured entries with versions, example media, and Markdown docs, then browse, tag, search, and reopen them.
Used to process video redrawing, frame skipping, frame ending early, etc.
Super fast multilingual speech recognition model based on Whisper Large-v3 Turbo. A node for ComfyUI.
This extension for ComfyUI adds a autocompletion feature that helps you write prompts faster and more efficiently. It provides suggestions from premade and custom data files as you type, allowing you to quickly insert and expand predefined keywords.
Comprehensive ComfyUI nodes for 3D mesh processing, PBR texture baking, caching, character consistency, and prompt iteration
ComfyUI-Benripack is an extension for ComfyUI that provides a CharacterPipe node. This node allows for managing various elements such as images, prompts, and models in a single structure, simplifying the workflow for character-based image generation.
A simple set of nodes to generate a point cloud from an image and its depth map, perform transformations and some basic operations.
A set of ComfyUI nodes to quickly test generated QR codes for scannability. A companion project to ComfyQR.
Some simple&practical ComfyUI image processing nodes.
Some simple&practical ComfyUI image processing nodes.
LLM-powered prompt engineering node for Z-Image Turbo workflows. By Cyberdelia AI Lab.
A collection of custom nodes for ComfyUI designed to apply various image processing effects, stylizations, and analyses.
Custom ComfyUI nodes for HYPIR image restoration using Stable Diffusion 2.1. Restores and upscales scanned or generated images with optional LM Studio prompt enhancement and tiled VAE processing for large resolutions.
Self-contained ComfyUI custom nodes for Krea 2 (Raw/Turbo): text-to-image, multi-stage high-res, 2x upscale-decode, LoRA stack, conditioning rebalance, and recipe presets.
Experience the CogVideoX model on ComfyUI
You'll get a new node called SD3 Latent Select Resolution, you can pick the x and y sizes from a list.
Pragmatic utilities for ComfyUI: Lazy Prompt (history + randomizer), Unified Loader (+ mini), Just Save Image (+ passthrough).
A collection of custom nodes for ComfyUI that enable powerful batch processing capabilities for both prompts and images.
Haiper API official ComfyUI custom node.
ComfyUI custom nodes for Ideogram 4.0 image generation.
Nodes for searching videos by motion
This is an implementation of [a/Qwen2-Audio-7B-Instruct-Int4](https://github.com/QwenLM/Qwen2-Audio) by [a/ComfyUI](https://github.com/comfyanonymous/ComfyUI), including support for text-based queries and audio queries to generate captions or responses.
Latent Residue Cleaner — a ComfyUI node that removes diffusion noise residue from AI-generated images using GPU-accelerated guided, bilateral, flat snap, and hue snap filters
This extension node is intended for the use of LCM conversion for SSD-1B-anime. It does not guarantee operation with the original LCM (as it cannot load weights in the current version). To take advantage of fast generation with LCM, a node for using TAESD as a decoder is also provided. This is inspired by ComfyUI-OtherVAEs.
QoL nodes for semi-automatic calculation of the best (most optimal) sampling resolution • compatible with ANY model (from now or the future), • accounting for upscale... • ...and pixel-step.
Attach a debug node to an output to obtain more detailed information. Uncover the details of your models in ComfyUI with ease.
ComfyUI nodes to use [a/ViewCrafter](https://github.com/Drexubery/ViewCrafter/tree/main) for novel view synthesis.
Direct model downloads to your Runod pod with blazing-fast multi-connection support. No more downloading models to your local machine and re-uploading!
ThoughtBubble is a custom node for ComfyUI that provides an interactive canvas to build and manage your prompts in a more visual and organized way. Think of it as a whiteboard for your ideas, allowing you to link different concepts, create conditional logic, and dynamically generate prompts using a powerful set of commands.
Experimental ComfyUI custom nodes for LTXV2 model with support for Gemma models. Tutorial available at https://www.youtube.com/@raretutor
A custom extension for ComfyUI that integrates Photopea directly into the sidebar.
A professional post-production, color grading, and sharpness suite for ComfyUI, featuring a custom built-in web application for real-time adjustments.
Adaptation of Fooocus Prompt Expansion for ComfyUI Forked from [a/ComfyUI-Prompt-Expansion](https://github.com/meap158/ComfyUI-Prompt-Expansion) with some updates and changes based on original Fooocus, to be more specific [a/expansion.py](https://github.com/lllyasviel/Fooocus/blob/main/extras/expansion.py) and [a/LykosAI - GPT-Prompt-Expansion-Fooocus-v2](https://huggingface.co/LykosAI/GPT-Prompt-Expansion-Fooocus-v2)
This node pack is designed to adjust audio playback speed within ComfyUI, particularly to sync audio with models like FantasyTalking (WAN) that require specific frame rates. It can also be used for general-purpose audio speed control.
Fully customizable Classifier Free Guidance for ComfyUI.
Compare two images using ComfyUI. Connect images to both image_a and image_b inputs. Press RUN. Then use the slider to compare. Best used when two images are very similar and you want to see the differences very closely. Search for the manager for image compare, or imagecompare.
llama-cpp-python wrapper, with support for vision models. It allows the user to generate text responses from prompts using llama.cpp.
Reason-then-encode for Krea 2. The encoder's own Qwen3-VL rewrites your prompt into a richer one — or reads a reference image (describe → filter → combine) — before conditioning: a built-in prompt enhancer and language-level image reference, running on the exact model that conditions.
Prompt Expansion for Stable Diffusion, using Deepseek API.
Face Anonymization make simple and easy.