Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Python 69 stars

lansespirit/image-gen-mcp

An MCP server that integrates with gpt-image-1 & Gemini imagen4 model for text-to-image generation services

Overview: Image Gen MCP Server is a server that integrates with multiple LLM-powered chatbot clients, enabling them to generate professional-quality images through the Model Context Protocol (MCP). Key features: - Supports multiple AI image generation models including gpt-image-1, gpt-image-2, dall-e-3, dall-e-2, and Google's Imagen series. - Works with any MCP-enabled LLM client, providing universal compatibility. - Seamless integration with no context switching, ensuring workflow continuity. - Standardized protocol for one server, multiple client support. - Access to OpenAI and Google's latest image generation models. Tech stack: Python. Why it matters: This server democratizes access to multiple AI image generation models, transforming text-only conversations into rich, visual experiences. It solves the limitations of traditional AI chatbot interfaces by providing universal compatibility, seamless integration, and access to multiple AI providers.

Scala 69 stars

lightbend/model-serving-tutorial

Code and presentation for Strata Model Serving tutorial

### Overview: This tutorial provides a hands-on introduction to serving machine learning models in the context of streaming data services. It discusses implementation options like TensorFlow Serving, embedded ML libraries, and Apache Kafka as the "data backplane." The core "use case" implemented is a stream processing application that ingests updated parameters for a machine learning model and scores the data. ### Key Features: - **Model Serving with TensorFlow Serving** - **Kafka as Data Backplane** - **Akka Streams, Spark, and Flink Implementations** - **Comparison of In-Memory State Management** ### Tech Stack: Scala ### Why It Matters: This tutorial offers a practical guide for developers interested in implementing machine learning models in real-time streaming applications. It provides insights into different deployment options and showcases how to manage in-memory state effectively. ### Repository: Repository: lightbend/model-serving-tutorial Primary Language: Scala GitHub Stars: 69 GitHub Description: Code and presentation for Strata Model Serving tutorial ### GitHub Description: The repository contains code and presentation materials for a Strata Model Serving tutorial. It provides a hands-on introduction to serving machine learning models in the context of streaming data services. The tutorial discusses implementation options like TensorFlow Serving, embedded ML libraries, and Apache Kafka as the "data backplane." The core "use case" implemented is a stream processing application that also ingests updated parameters for a machine learning model and then uses the model to score the data. Several implementations of this use case are provided, where we compare and contrast the use of Akka Streams, Spark,

Python 69 stars

linden-li/collage-diffusion-ui

An open source, layer-based web interface for Collage Diffusion - use a familiar Photoshop-like interface and let the AI harmonize the details.

### Overview: Collage Diffusion UI is an open-source project that provides a user-friendly interface for interacting with image generation models. It allows users to specify the composition of an image in a familiar Photoshop-like interface, ensuring that the details are harmonized and plausible. ### Key Features: - **Layered Interface**: Precisely outlines how a scene should be composed, from perspectives to lighting. - **Harmonized Details**: Ensures that everything from perspectives to lighting is plausible, unlike traditional text prompting supported by diffusion interfaces. - **Stable Diffusion Model**: Modified version of Stable Diffusion v1.5, using HuggingFace Diffusers library. - **React App**: Frontend built with Chakra UI, TypeScript, and Vite. - **Ray Serve**: Custom scheduler runtime for inference. - **Google Cloud Storage**: Configured for uploading models and assets to a Google Cloud Bucket. ### Tech Stack: - **Frontend**: React App with Chakra UI UI components. - **Server**: Modified Stable Diffusion v1.5 model using HuggingFace Diffusers. ### Why It Matters: Collage Diffusion UI provides a more intuitive and visually appealing interface for interacting with image generation models, making it easier and more intuitive for users to create and edit images. The ability to precisely outline scene composition ensures that the resulting image is harmonized and plausible, enhancing the user experience.

Shell 69 stars

opensource-spraakherkenning-nl/Kaldi_NL

Code related to the Dutch instance and user groups of the KALDI speech recognition toolkit

Overview: This repository contains scripts and models for Dutch speech recognition using the Kaldi open-source toolkit. It includes scripts for converting speech to text and a pre-built container image with models and a web interface for automatic speech recognition. Key features: - Convert speech to text using Kaldi's speech recognition system - Container image with models and web interface for automatic speech recognition - Support for various Dutch models and decoder graphs Tech stack: Shell, Java, Perl, Python 3, SoX Why it matters: This tool is useful for developers and researchers working on Dutch speech recognition projects, providing a convenient and efficient way to integrate Kaldi models into their workflow. GitHub stars: 69

Python 69 stars

pinecone-io/pinecone-claude-code-plugin

The official Pinecone marketplace for Claude Code Plugins

Overview: Pinecone is a vector database that integrates directly into Claude Code, offering semantic search, index management, and Retrieval Augmented Generation (RAG) capabilities. This plugin streamlines the integration process, making it easy to use Pinecone’s powerful indexing and search functionalities within Claude Code. Key features: - Pinecone Assistant: A fully managed RAG service for document Q&A with citations, natural language support, and incremental file syncing. - Pinecone MCP Server: Full integration with the Pinecone Model Context Protocol server for index creation, listing, searching, and more. - Slash Commands: Quick access to common Pinecone operations directly from Claude Code. - Semantic Search: Query your vector indexes using natural language. - Natural Language Recognition: Assistant commands work without explicit slash commands. Tech stack: Python Why it matters: This plugin simplifies the integration of Pinecone’s vector database into Claude Code, enabling powerful semantic search and RAG workflows, making it easier and more efficient for developers to manage and utilize vector database capabilities within their projects.

Python 69 stars

satojkovic/DeepLogo2

A brand logo detection system by DETR

Overview: DeepLogo2 is a brand logo detection system utilizing DETR, a Transformer-based object detection model. It provides a training and inference environment for creating logo detection models. Key features: - Training and inference environment for DETR logo detection models. - Uses the Flickr logos 27 dataset for fine-tuning and evaluation. - Supports conversion of the dataset to COCO format for fine-tuning. Tech stack: Python. Why it matters: This system is useful for developers looking to implement logo detection capabilities in their applications, leveraging the power of DETR for accurate and efficient logo recognition.

Python 69 stars

speediedan/finetuning-scheduler

A PyTorch Lightning extension that accelerates and enhances foundation model experimentation with flexible fine-tuning schedules.

Overview: FinetuningScheduler is a PyTorch Lightning extension designed to accelerate and enhance foundation model experimentation with flexible fine-tuning schedules. Key features: - Flexible specification of fine-tuning schedules with explicit or regex-based parameter selection. - Automatic restoration of best per-phase checkpoints driven by early-stopping criteria. - Composition of early-stopping and manually-set epoch-driven fine-tuning phase transitions. - An experimental PyTorch API that brings the same schedules to a plain PyTorch training loop. Tech stack: Python Why it matters: FinetuningScheduler simplifies model research and exploration by providing a powerful and flexible tool for fine-tuning experiments, making it easier to experiment with different parameters and fine-tuning schedules without the need for complex setup or extensive configuration.

Python 69 stars

tanyuqian/redco

NAACL '24 (Best Demo Paper RunnerUp) / MlSys @ NeurIPS '23 - RedCoast: A Lightweight Tool to Automate Distributed Training and Inference

### Overview: RedCoast (Red Coast) is a lightweight tool designed to automate distributed training and inference for large models. It simplifies the ML pipeline development process without requiring MLSys expertise, making it suitable for users of any background. ### Key Features: - Supports large models beyond Transformers, such as Stable Diffusion. - Supports complex algorithms beyond cross entropy, including Meta Learning, Differential-Private (DP) Training, and Federated Learning. - Simplifies the ML pipeline development process without requiring MLSys expertise. - Automatically adjusts Jax to GPU/TPU version based on the user's device. - Automatically installs the required Jax version for GPU/TPU compatibility. ### Tech Stack: Python ### Why it Matters: RedCoast is a game-changer for researchers and developers who need to train large models but lack the expertise in distributed training. It streamlines the process, making it accessible to a wider audience, and ensures that the models can be trained efficiently and effectively on any available hardware, including GPUs and TPUs. ### Repository: tanyuqian/redco ### GitHub Stars: 69 ### GitHub Description: NAACL '24 (Best Demo Paper RunnerUp) / MlSys @ NeurIPS '23 - RedCoast: A Lightweight Tool to Automate Distributed Training and Inference ### Installation: ```bash pip install redco ``` ### Examples: - **Classification/regression (GLUE & MNIST)** - **Differential-private_training** - **Federated learning (FedAvg)**

Swift 69 stars

tuff-app/TUFF

The local AI app built for the Mac. Run LLMs like Qwen, Gemma and GPT-OSS on Apple Silicon, including models far bigger than your Mac's memory, by streaming MoE experts from the SSD.

### Overview: TUFF is a local AI app designed for macOS, enabling users to run large language models on Apple Silicon without exceeding their Mac's memory capacity. It supports models like Qwen, Gemma, and GPT-OSS, allowing users to interact with these models seamlessly on their Macs. ### Key features: - **Chat with Reasoning**: Engage in conversations with models like Qwen, Gemma, and GPT-OSS, with the ability to include reasoning, Markdown, math, and file attachments. - **Web and File Search**: Access web and file search capabilities, which can be turned on or off. - **Benchmarks**: Measure and share the performance of installed models on real Macs. - **Background API**: A local OpenAI-compatible server for agents and other apps, loading the specified model for requests. ### Tech stack: Swift and Metal ### Why it matters: TUFF provides a practical solution for running large language models on Apple Silicon Macs, enabling users to interact with models like Qwen, Gemma, and GPT-OSS without the need for additional memory. This makes it ideal for developers and users looking to leverage advanced AI capabilities on their Macs.

Unknown 69 stars

watsonyanghx/Image-Text-Papers

Image Caption and Text to Image papers.

Overview: This repository contains papers related to image captioning and text-to-image generation. It covers various techniques such as visual-semantic embeddings, encoder-decoder models, reinforcement learning, and more. Key features: - Survey of models, datasets, and evaluation measures - Visual-semantic embedding-based approaches - Encoder-decoder architectures - Reinforcement learning techniques - RNN-based methods Tech stack: Python, TensorFlow, PyTorch, and other related frameworks. Why it matters: This repository provides a comprehensive overview of recent advancements in image captioning and text-to-image generation, helping developers and researchers stay updated with the latest techniques and methodologies in the field. Repository: watsonyanghx/Image-Text-Papers GitHub stars: 69 Primary language: Unknown GitHub description: Image Caption and Text to Image papers.

Python 69 stars

worldbench/VideoLucy

[NeurIPS 2025] Deep Memory Backtracking for Long Video Understanding

### Overview: VideoLucy is a deep memory backtracking framework designed for long video understanding. Inspired by human memory processes, it employs a hierarchical memory structure with progressively granular details. VideoLucy mines video-wide, question-relevant deep memories through an agent-based iterative backtracking mechanism, providing confident answers by gathering sufficient information. ### Key Features: - **Hierarchical Memory Structure**: A progressive granularity hierarchical memory structure for detailed video understanding. - **Agent-Based Iterative Backtracking**: Systematically mines video-wide, question-relevant deep memories. - **EgoMem Benchmark**: A new benchmark designed to evaluate a model's ability to understand complex events and capture fine-grained details in long videos. ### Tech Stack: - Python ### Why It Matters: VideoLucy offers a novel approach to long video understanding by leveraging human memory processes, providing a structured and granular way to mine video data. This method not only enhances the accuracy of the model but also makes it more efficient and effective in handling complex events and long videos. ### Repository: - Repository: worldbench/VideoLucy - GitHub Stars: 69 - GitHub Description: [NeurIPS 2025] Deep Memory Backtracking for Long Video Understanding - Primary Language: Python ### News: - **Demo Release**: The VideoLucy demo and EgoMem benchmark are now available for use. - **Paper Publication**: The paper "VideoLucy: Deep Memory Backtracking for Long Video Understanding" has been accepted at NeurIPS2025. - **Demo Environment**: The

JavaScript 69 stars

xialeistudio/echoic

### Overview: Echoic is an AI-powered speaking practice tool that allows users to import any audio and practice sentence by sentence. It offers three modes: Read Aloud, Situational, and Monologue, with phoneme-level pronunciation scoring and word-level breakdowns. ### Key Features: - **Oral Practice**: Three modes—Read Aloud, Situational, and Monologue—with phoneme-level accuracy, fluency, and completeness scoring. - **Content Gallery**: Browse and import episodes from VOA Learning English and BBC Learning English. - **Audio Import**: Upload local files or import from any direct audio URL. - **Collections**: Organize audio into named collections. - **Sentence Practice**: Practice each sentence with adjustable playback speed. - **Pronunciation Scoring**: Accuracy, fluency, and completeness scores with word-level breakdown. - **Phoneme Display**: IPA transcription per word; phonemes color-coded by score after assessment. - **Word Error Review**: Aggregate word accuracy across all sessions to identify weak spots. - **A/B Compare**: Play original then your recording back-to-back in one click. - **AI Sentence Analysis**: Translation and grammar breakdown via OpenAI or local Ollama (optional). - **Practice History**: Every attempt saved with full score details; click to replay any recording. - **Sentence States**: Bookmark sentences for review; mark sentences as mastered to hide them. - **Sentence Search**: Filter sentences by text within any audio file. - **Practice Heatmap**: 365-day activity calendar on the overview page. - **Keyboard

Python 68 stars

AI4Science-WestlakeU/wdno

[ICLR 2025] Wavelet Diffusion Neural Operator (WDNO) uses diffusion models on wavelet space for generative PDE simulation and control.

### Overview: WDNO: Wavelet Diffusion Neural Operator (ICLR 2025) is a novel method for generative PDE simulation and control, addressing the challenges of diffusion models in modeling abrupt changes and generalizing to higher resolutions. ### Key Features: - WDNO shares a unified input–output interface with standard neural operators. - It uses wavelet space for generative PDE simulation and control. - The method is designed to handle system states with abrupt changes and improve generalization to higher resolutions. ### Tech Stack: Python ### Why It Matters: WDNO provides a solution for generative PDE simulation and control, addressing the limitations of diffusion models in handling abrupt changes and generalizing to higher resolutions. This makes it particularly useful in applications requiring accurate and robust modeling of physical systems.

HTML 68 stars

ArrozDotachi/InvokeAI-Windows-Setup-Guide

InvokeAI 2026 Windows Setup Guide: Run Stable Diffusion Locally on Win 11 & 10

### Overview: InvokeCanvas Studio 2026 is a next-generation, browser-first creative workstation designed for Windows 11 and Windows 10. It offers a canvas-first environment where local AI image generation feels more like sitting at an easel, making it feel less like running a script and more like creating art. ### Key Features: - **Responsive Interface**: The interface is designed to be intuitive and responsive, making it easy to navigate and use. - **Multilingual Support**: The documentation and community knowledge base are available in 12 languages, catering to a global audience. - **24/7 Customer Support**: The project is actively maintained and supported, ensuring continuous improvement and assistance. - **System Requirements**: The software is designed to run smoothly on modern Windows 11 and 10 systems, requiring minimal hardware resources. - **Getting Started on Windows 11 & 10**: Detailed instructions are provided to set up and use the software on both operating systems. - **The Canvas Workflow**: The workflow is centered around the canvas, where all generation, editing, and batch rendering operations take place. - **Model Management Philosophy**: The project emphasizes local-first generation, with a layered editing model and a persistent session ledger. - **Performance & Hardware Notes**: The software is optimized for performance, ensuring stable and efficient operation even on less powerful hardware. - **Use Cases & Creative Scenarios**: The software is designed to be versatile, suitable for a wide range of creative projects and scenarios. - **SEO & Discoverability Notes**: The project is optimized for search engines

Python 68 stars

Azure-Samples/azure-ai-search-multimodal-sample

A sample app for the Multimodal Retrieval-Augmented Generation pattern running in Azure, using Azure AI Search for retrieval and Azure OpenAI large language models to power Q&A experiences.

## Overview: This repository contains the code for an application designed to showcase multimodal Retrieval-Augmented Generation (RAG) techniques using Azure AI Search, Azure OpenAI large language models, and Azure OpenAI large language models to power Q&A experiences. ## Key features: - Azure AI Search integration for text retrieval - Azure OpenAI large language models for Q&A - Multimodal capabilities to extract and reason over text and image data ## Tech stack: Python ## Why it matters: This demo provides a hands-on experience with multimodal AI techniques, showcasing how to integrate AI capabilities to create custom copilots/RAG applications that can query, retrieve, and reason over both text and image data.

Python 68 stars

CharlesPikachu/FreeGPTHub

FreeGPTHub: A truly free, unified GPT API gateway—unlike “fake-free” alternatives that hide paywalls behind quotas, trials, or mandatory top-ups. (真正免费的GPT统一接口，支持讯飞星火知识大模型，百度文心大模型，OpenAI ChatGPT，智谱大模型，幻方量化深度求索，字节豆包，阿里通义千问，稀宇科技MiniMax，魔搭社区开源模型等)

Overview: FreeGPTHub is a free, unified GPT API gateway that supports multiple models including AlibabaQwenEndpoints, BaiduQianfanEndpoints, and DeepSeekEndpoints. Key features: - Supports multiple GPT models - Provides a unified interface for accessing different models - Free and open-source Tech stack: Python Why it matters: FreeGPTHub simplifies the process of accessing various GPT models, making it an attractive option for developers looking to integrate AI capabilities without the need for paywalls or quotas. Repository: CharlesPikachu/FreeGPTHub GitHub stars: 67 GitHub description: FreeGPTHub: A truly free, unified GPT API gateway—unlike “fake-free” alternatives that hide paywalls behind quotas, trials, or mandatory top-ups.

Unknown 68 stars

Cordux/ComfyUI-Wan2.2-workflow

A ComfyUI Workflow for low vram users

### Overview: Generate high-quality text-to-video and image-to-video content using Wan 2.2 14B on budget hardware! ### Key features: - **Low VRAM Optimization** - Works on 6GB VRAM cards (RTX 3050, 3060, AMD equivalents) - **User-Friendly Interface** - Simple FPS & Seconds input (automatic frame calculation) - **Dual Mode Support** - Text-to-Video (t2v) and Image-to-Video (i2v) with easy switching - **Last Frame Extraction** - Optional last frame saving for video chaining/extension - **GGUF Quantization** - Leverages quantized models for memory efficiency - **Smart Memory Management** - Aggressive CPU offloading for low-VRAM systems - **Fast Generation** - 4-step sampling for rapid results ### Tech stack: ComfyUI ### Why it matters: This workflow is designed for users with low VRAM, providing efficient and fast generation capabilities on budget hardware.

Python 68 stars

Enigma-Vault/NoPII

Ready-to-run examples showing NoPII PII protection with OpenAI, Anthropic, LangChain, LlamaIndex, and more

Overview: NoPII is a PII-tokenizing reverse proxy designed to protect sensitive information in LLM API requests and responses, ensuring compliance with PCI DSS Level 1 and SOC 2 Type II standards. Key features: - PII detection and tokenization - Deterministic tokenization for LLM providers - Secure handling of API keys - Fail-safe mechanism to block requests if tokenization fails Tech stack: Python Why it matters: NoPII simplifies the process of protecting sensitive data in LLM API traffic, making it easier for developers to comply with security standards without compromising performance. Repository: Enigma-Vault/NoPII Primary language: Python GitHub stars: 67 GitHub description: Ready-to-run examples showing NoPII PII protection with OpenAI, Anthropic, LangChain, LlamaIndex, and more Security details: PCI DSS Level 1 certified, SOC 2 Type II audited, AWS Partner.

Python 68 stars

IlyaGusev/saiga_bot

Telegram bot for different language models. Supports system prompts and images

Overview: Saiga is a Telegram bot designed to interact with various language models, supporting system prompts and image processing. Key features: - Integration with the Telegram platform - Support for any number of language models (compatible with OpenAI API format) - Ability to handle system prompts, generation parameters, and characters Tech stack: Python Why it matters: Saiga is a versatile tool for developers looking to integrate language model capabilities into their projects, offering a simple yet powerful interface for interacting with different language models. Repository: IlyaGusev/saiga_bot Primary language: Python GitHub stars: 68 GitHub description: Telegram bot for different language models. Supports system prompts and images

Python 68 stars

MKFMIKU/vidm

[AAAI23 Oral] Official implementations of Video Implicit Diffusion Models

### Overview: VIDM is an official implementation of Video Implicit Diffusion Models, showcasing state-of-the-art performance in generating high-quality and diverse videos. This project leverages diffusion models to synthesize motion implicitly, ensuring plausible video sequences. The method is evaluated on various datasets and outperforms state-of-the-art methods in terms of Frechet Video Distance (FVD) and Video Inception Score (VIS). ### Key Features: - **Implicit Motion Modeling:** The model generates videos by sampling latent features of frames, enabling the modeling of motion implicitly. - **Multiple Strategies:** The project includes improvements such as sampling space truncation, robustness penalty, and positional group normalization to enhance video quality. - **Versatile Performance:** The model performs exceptionally well across different video resolutions and frame counts, demonstrating its effectiveness in various scenarios. ### Tech Stack: - **Python:** The project is built and tested using Python 3.8 with PyTorch 1.8.0 and CUDA 11.6. ### Why It Matters: VIDM represents a significant advancement in video generation, offering a powerful tool for creating realistic and diverse video sequences. Its ability to generate motion implicitly and its superior performance in FVD and VIS metrics make it a valuable contribution to the field of video synthesis.