ComfyUI Custom Extensions

Add a node to save images with metadata (PNGInfo) extracted from the input values of each node. Since the values are extracted dynamically, values output by various extension nodes can be added to metadata.

A ComfyUI custom node implementation of PersonaLive: Expressive Portrait Image Animation for Live Streaming, enabling portrait animation driven by reference images. (Description by CC)

A ComfyUI node that intelligently composites a generated (AI-edited) image back onto an original image by detecting only what actually changed between the two images and blending the generated content in cleanly.

Nodes: MultiText, TextBox, TitlePlus, SeamlessTexture, AspectRatioPlus, DisplayEverything, ComparerPlus, AnySwitch, Node Design Tools...

Nodes:Ood_hd_CXH, Ood_hd_CXH. [a/OOTDiffusion](https://github.com/levihsu/OOTDiffusion)

This is an extension for ComfyUI that makes it possible to use some LLM models provided by Ollama, such as Gemma, Llava (multimodal), Llama2, Llama3 or Mistral. Speaking specifically of the LLaVa - Large Language and Vision Assistant model, although trained on a relatively small dataset, it demonstrates exceptional capabilities in understanding images and answering questions about them. This model presents similar behaviors to multimodal models such as GPT-4, even when presented with invisible images and instructions.

The main node makes your conditioning go towards similar concepts so to enrich your composition or further away so to make it more precise. It gathers similar pre-cond vectors for as long as the cosine similarity score diminishes. If it climbs back it stops. This allows to set a relative direction to similar concepts. There are examples at the end but [a/you can also check this imgur album](https://imgur.com/a/WvPd81Y) which demonstrates the capability of improving variety.

Nodes for running the JoyCaption image captioner VLM.

This is a diffusers (0.27.2) wrapper node for Geowizard: [https://github.com/fuxiao0719/GeoWizard]. The model is autodownloaded from Hugginface to ComfyUI/models/diffusers/geowizard

Interactive 3D interface for selecting camera angles for Qwen-Image-Edit-2511, supporting 96 unique camera angle combinations (8 directions × 4 heights × 3 shot sizes) with multi-selection and visual feedback.

ComfyUI nodes for Viggle-Animate (MiniMax-H3 ref2va finetune): character replacement in video from a driving clip and a reference still

Lightweight ComfyUI wrapper that bring Character.AI's Ovi video+audio generator to ComfyUI with streamlined setup, selectable precision, and attention-backend control.

ComfyUI-SparkTTS is a custom ComfyUI node implementation of SparkTTS, an advanced text-to-speech system that harnesses the power of large language models (LLMs) to generate highly accurate and natural-sounding speech.

Local MCP server and browser bridge for ComfyUI. Inspect workflows, edit graphs, queue generations, manage models, and connect MCP-compatible agents to your local ComfyUI setup.

A memory-efficient implementation for upscaling videos in ComfyUI using non-diffusion upscaling models. This custom node is designed to handle large video frame sequences without memory bottlenecks.

Update support Qwen Image 2.1

Extended Save node for ComfyUI

ComfyUI-OmniGen2 is now available in ComfyUI, OmniGen2 is a powerful and efficient unified multimodal model. Its architecture is composed of two key components: a 3B Vision-Language Model (VLM) and a 4B diffusion model.

Logical Utils (compare, string, boolean operations) for ComfyUI

Janky implementation of [a/HiDiffusion](https://github.com/megvii-research/HiDiffusion) for ComfyUI. Enables generating at resolutions higher than what the model was trained for. Only supports SD 1.x (maybe 2.x) and SDXL.

PuLID adaptation for Flux.2 bringing consistent identity to generated images using face embedding and visual feature injection into Flux.2 model.

Some patches for Flux|HunYuanVideo|LTXVideo etc, support TeaCache, PuLID, First Block Cache.

Audio-only refinement pass for MiniMax H3: freeze the video stream of a sampled latent and run extra denoising steps on the audio only, with an optional frozen-video KV cache that makes those steps several times cheaper.

the custom code for [a/UVR5](https://github.com/Anjok07/ultimatevocalremovergui) to separate vocals and background music

Nodes for use of Chroma model and other prototype models

CRT-Nodes is a collection of custom nodes for ComfyUI.

BMAB for ComfyUI. BMAB is an custom nodes of ComfyUI and has the function of post-processing the generated image according to settings. If necessary, you can find and redraw people, faces, and hands, or perform functions such as resize, resample, and add noise. You can composite two images or perform the Upscale function.

Nodes:Text_Image_Zho, Text_Image_Multiline_Zho, RGB_Image_Zho, AlphaChanelAddByMask, ImageComposite_Zho, ...

A simple local text generator for ComfyUI utilizing [a/ExLlamaV2](https://github.com/turboderp/exllamav2). [w/NOTE:Manual package installation is required.]

A drawing and sketching node for ComfyUI with layers, multiple brush types, and a focused, professional interface.

Supported by anima_lora trainer daemon, this introduces simplified comfyui ui for training in background.

This custom node provides advanced settings for FreeU.

NODES: Comfly_Mj, Comfly_mjstyle, Comfly_upload, Comfly_Mju, Comfly_Mjv, Comfly_kling_text2video, Comfly_kling_image2video, Comfly_video_extend, Comfly_lip_sync, Comfly_kling_videoPreview, Comfly Gemini API, Comfly Doubao SeedEdit, Comfly ChatGPT Api,Comfly Jimeng API, Comfly_gpt_image_1_edit, Comfly_gpt_image_1

A custom LoRA-loading node designed to prevent issues such as blurriness and other artifacts when loading multiple LoRAs in HunYuan Video. Usage Instructions: The connection method remains unchanged from the original. The only difference is the additional blocks_type option. Please select double_blocks.

Nodes:Load Ultralytics Model, Ultralytics Inference, Ultralytics Visualization, Convert to Dictionary, BBox to XYWH

Nodes: Int, Float, String, Operation, Checkpoint

Civicomfy seamlessly integrates Civitai's vast model repository directly into ComfyUI, allowing you to search, download, and organize AI models without leaving your workflow.

ComfyUI nodes for automatically generating XML-style prompts compatible with NewBie models using LLM APIs, with style customization and preset management capabilities. (Description by CC)

This extension automates the creation of structured XML prompts for the NewBie model by using LLM APIs to transform natural language or images into formatted descriptions. It provides specialized nodes for high-robustness prompt generation and seamless artistic style injection to enhance image generation workflows.该插件利用大语言模型 API 将自然语言或图片自动转化为适用于 NewBie 模型的结构化 XML 提示词。通过提供高度健壮的提示词生成与画面风格注入节点，它显著提升了图像生成流程的效率与效果。

Production sparse attention, token ordering, and memory optimization nodes for MiniMax H3 in ComfyUI

Agent sidebar panel for ComfyUI — chat with an autonomous agent (Claude, Codex/GPT, Gemini, or local Ollama) that builds, runs, and debugs workflows on your live canvas. UI-only pack (also on the Comfy Registry as comfyui-agent-panel); pairs with the comfyui-mcp orchestrator: npx -y comfyui-mcp@latest --panel-orchestrator

Shows Lora information from CivitAI and outputs trigger words and example prompt

This custom node for ComfyUI integrates the Flux-Prompt-Enhance model, allowing you to enhance your prompts directly within your ComfyUI workflows.

For unloading a model or all models, using the memory management that is already present in ComfyUI. Copied from [a/https://github.com/willblaschko/ComfyUI-Unload-Models](https://github.com/willblaschko/ComfyUI-Unload-Models) but without the unnecessary extra stuff.

Nodes to use Florence2 VLM for image vision tasks: object detection, captioning, segmentation and ocr

RadialAttention in ComfyUI native workflow

ExLlamaV2 nodes for ComfyUI.

This extension helps generate images through NAI.

Maintained by Eden.art, this is a growing suite of custom nodes for building advanced pipelines.

This is a request node tool designed for making HTTP requests (GET/POST) to APIs and viewing the responses. It is useful for API testing and development.

ComfyUI LayerDivider is custom nodes that generating layered psd files inside ComfyUI[w/Please follow readme and run install_windows_portable_win_py311_cu121 for ComfyUI embedded python.]

A reference-aware MiniMax H3 still-image studio for ComfyUI

Nodes:Word Cloud, Load Text File

An enhanced OpenPose Editor for ComfyUI with pose auto-completion, pause/edit workflow support, and UI fixes.

Vision-aware text conditioning for the Krea2 / K2 model. Feeds reference images through the Qwen3-VL-4B vision path with the Krea2 descriptor template; optional per-image mask crops to the masked region. No VAE (Krea2 has no reference-latent pathway). Auto-growing image+mask slots.

Multi-segment MiniMax H3 video director for ComfyUI with continuity, selective reruns, asset management, post-processing, and live preview.

This is a node to converts models into Fp8, bf16, fp16.

🍭 大炮-Qwen3VL ComfyUI 自定义节点集合，集成了阿里云 Qwen 团队开发的 Qwen3-VL 多模态大语言模型系列。

This project provides a TensorRT implementation of [a/RIFE](https://github.com/hzwer/ECCV2022-RIFE) for ultra fast frame interpolation inside ComfyUI

LowVRAM Animation : txt2video - img2video - video2video , Frame by Frame, compatible with LowVRAM GPUs Included : Prompt Switch, Checkpoint Switch, Cache, Number Count by Frame, Ksampler txt2img & img2img ...

An interactive image cropping node for ComfyUI, allowing precise visual selection of crop areas directly within your workflow. This node is designed to streamline the process of preparing images for various tasks, ensuring immediate visual feedback and control over your image dimensions.

Custom AI prompt generator node for ComfyUI.

[Original] In the field of portrait video generation, the use of single images to generate portrait videos has become increasingly prevalent. A common approach involves leveraging generative models to enhance adapters for controlled generation. However, control signals can vary in strength, including text, audio, image reference, pose, depth map, etc. Among these, weaker conditions often struggle to be effective due to interference from stronger conditions, posing a challenge in balancing these conditions. In our work on portrait video generation, we identified audio signals as particularly weak, often overshadowed by stronger signals such as pose and original image. However, direct training with weak signals often leads to difficulties in convergence. To address this, we propose V-Express, a simple method that balances different control signals through a series of progressive drop operations. Our method gradually enables effective control by weak conditions, thereby achieving generation capabilities that simultaneously take into account pose, input image, and audio. NOTE: You need to downdload [a/model_ckpts](https://huggingface.co/tk93/V-Express/tree/main) manually.

Long-form MiniMax H3 video generation for ComfyUI with AV latent continuation, reference conditioning, chunked sampling, and resume support.

Using Qwen-2.5 in ComfyUI

a comfyui custom node for [a/MimicBrush](https://github.com/ali-vilab/MimicBrush),then inpainting with reference image.

This extension offers various nodes that are useful for Deforum-like animations in ComfyUI.

A ComfyUI custom node to read LoRA tag(s) from text and load it into checkpoint model.

Houdini connection plugins

Enhanced QwenVL node with MiniMax H3 loop mode and presets, multilingual prompt generation, camera & style tag dropdowns, Pony prompt converters, Qwen 3.8 uncensored models, GGUF backend, LTX 2.3 FL2VA presets, and local model discovery for professional multimodal video workflows.

Custom nodes by IAMCCS, including a critical fix for LoRA loading in native WANAnimate workflows, especially useful for advanced models like FLUX and WAN 2.1. This bypasses the need for the WanVideoWrapper for LoRA functionality.

Professional color grading and film emulation suite — 29 nodes: 161 film stocks with real Capture One curve data, Camera Raw tools, Resolve-level grading (LGG, Log Wheels, Hue vs Hue, Color Warper), lens optics with 102 real lens profiles. Zero API costs.

Real-time Gallery for ComfyUI with image metadata inspection. Support for images and video.

Generate Audio from any video and or text

Qwen3-TTS voice cloning and multi-character dubbing. Features script-driven unlimited multi-character dubbing, emotion-aware speech-to-speech (ASR) technology, and efficient speech asset management capabilities.

Generate MiniMax H3 prompts locally with Qwen models in ComfyUI.

ComfyUI nodes to use [a/FLATTEN: optical FLow-guided ATTENtion for consistent text-to-video editing](https://github.com/yrcong/flatten).

An extension that's adds advanced audio processing capabilities to ComfyUI with professional-grade audio effects and AI-powered audio enhancement.

NODES:Canvas View

Load mp3 files and use the audio nodes to power animations and prompt scheduling. Use with FizzNodes.

Enhanced Wan2.2 image-to-video node with dual sampler workflow, dynamic enhancement, and intelligent color drift correction for video generation. (Description by CC)

ComfyUI Custom Nodes including FrameRangedFaceLoader for consistent face editing and framing

Interactive Gaussian Splatting PLY preview and render nodes for ComfyUI.

DynamiCrafter that works natively with ComfyUI's nodes, optimizations, ControlNet, and more.

ComfyUI wrapper nodes to use the Diffusers implementation of ELLA

Advanced camera control prompt generator for ComfyUI, optimized for dx8152's MultiAngle LoRA. Simplifies multi-directional movements, rotations, and special views with bilingual output.

Using Qwen-2 in ComfyUI

This extension currently has two sets of nodes - one set for editing the contrast/color of images and another set for saving images as 16 bit PNG files.

A custom node collection for integrating various LLM (Large Language Model) providers with ComfyUI.

A custom ComfyUI node for interactive 360° panorama image previews. Panoramic 360 images are also sometimes known as VR photography (virtual reality), HDRI environments (ex: skyboxes), image spheres, spherical images, 360 pano, and 360 degree photos.

Tara is a powerful node for ComfyUI that integrates Large Language Models (LLMs) to enhance and automate workflow processes. With Tara, you can create complex, intelligent workflows that refine and generate content, manage API keys, and seamlessly integrate various LLMs into your projects.

ComfyUI custom nodes for Foundation-1 | structured text-to-sample generation for music production

SegFormer model fine-tuned on ATR dataset for clothes segmentation but can also be used for human segmentation! Download the weight and put it under checkpoints: [a/https://huggingface.co/mattmdjaga/segformer_b2_clothes](https://huggingface.co/mattmdjaga/segformer_b2_clothes)

Without fine-tuning, FLUX.1 Dev model cannot understand exact color codes. However, it is known that FLUX.1 Dev can repeatedly produce certain colors with certain prompt(color name). Fortunately, on CIVITAI, [a/“novuschroma” shared 155 pre-tested color names](https://civitai.com/models/879997/color-wildcards-for-flux-and-sdxl) that FLUX.1 Dev can handle. Thanks to his resource, color palette consists exclusively of 155 colors can be configured. ‘ColorPalette’ node from ComfyUI APQNodes converts input hex color code to the most similar color name(from pre-tested 155 color names) of which FLUX.1 Dev is aware.

A lightweight FirstBlockCache model patch for native ComfyUI MiniMax H3.

Run EbSynth, Fast Example-based Image Synthesis and Style Transfer, in ComfyUI.

twinflow:Realizing One-step Generation on Large Models with Self-adversarial Flows,you can use it in comfyUI

Non-native [a/AttentionDistillation](https://github.com/xugao97/AttentionDistillation) for ComfyUI. Official ComfyUI demo for the paper AttentionDistillation, implemented as an extension of ComfyUI. Note that this extension incorporates AttentionDistillation using diffusers.

Moondream image to text query node with batch support