AI Trending

Project directory

Projects

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

Python 127 stars

ant-research/edicho

[ICCV 2025] Edicho: Consistent Image Editing in the Wild

### Overview: Consistent image editing in the wild has become a pressing need due to the variability in object poses, lighting conditions, and photography environments. Edicho, an innovative solution, addresses this challenge by leveraging explicit image correspondence to generate precise and consistent editing results in zero-shot scenarios. ### Key Features: - **Training-Free Solution**: Edicho is a training-free method that achieves consistent editing across images without requiring any pre-training. - **Attention Manipulation Module**: This module enhances the editing process by manipulating attention features. - **Classifier-Free Guidance (CFG)**: The CFG denoising strategy guides the denoising process in two levels, ensuring precise consistency across different image editing tasks. ### Tech Stack: - **Python**: The primary programming language used for implementing the solution. - **Diffusion Models**: The underlying technology that enables the generation of consistent edits. ### Why It Matters: Consistent image editing across in-the-wild images is crucial for applications such as post-processing in the field of computer vision, ensuring that edits are applied uniformly and effectively across diverse scenes. ### Repository: The project is hosted on GitHub under the repository `ant-research/edicho`, with 127 GitHub stars, showcasing its growing interest and adoption in the community. ### How to Run: 1. **Environment Setup**: Ensure that all necessary dependencies are installed. 2. **Pretrained Model Weights**: Prepare the editing model and base diffusion model. 3. **Run Inference**: Organize your input images and run the inference script with appropriate parameters. ### Outputs: Inference results are saved under
Python 127 stars

cyberark/agentwatch

A powerful AI observability framework that provides comprehensive insights into agent interactions across platforms, enabling developers to monitor, analyze, and optimize AI-driven applications with minimal integration effort.

### Overview agentwatch is a powerful AI observability framework designed to provide comprehensive insights into AI agent interactions across various platforms and frameworks. By seamlessly intercepting, logging, and analyzing interactions, it enables developers to monitor, analyze, and optimize AI-driven applications with minimal integration effort. ### Key Features - **A One-liner Observability Framework**: Instant insights with just a single import. - **Comprehensive Interaction Tracking**: Monitor LLM and tool calls in real-time. - **Advanced Visualization**: Generate intuitive graphs for in-depth analysis. - **Detailed Metadata Capture**: Log tool inputs, arguments, and performance metrics. - **Multi-Framework Support**: Compatible with various AI development frameworks. - **MCP Support**: Tracks tool calls over remote MCP servers (HTTP/SSE only atm). ### Tech Stack Python ### Why It Matters agentwatch empowers developers to understand and optimize their AI-driven applications with unprecedented visibility, making it essential for modern software development.
C++ 127 stars

intuitive-robots/MoDE_Diffusion_Policy

[ICLR 25] Code for "Efficient Diffusion Transformer Policies with Mixture of Expert Denoisers for Multitask Learning"

### Overview: This repository contains the code for the "Efficient Diffusion Transformer Policies with Mixture of Expert Denoisers for Multitask Learning" paper, presented at ICLR 2025. The project focuses on developing a policy-based approach for multitask learning using diffusion models and expert denoisers. ### Key features: - **Efficient Diffusion Models**: Utilizes diffusion models to learn policies that can perform multiple tasks efficiently. - **Mixture of Expert Denoisers**: Combines multiple denoisers to improve the robustness and generalization of the diffusion models. - **Multitask Learning**: The policy can be trained on multiple tasks simultaneously, improving performance and reducing the need for separate models. ### Tech stack: - **Programming Languages**: The project is written in C++. - **Dependencies**: Requires specific versions of libraries and tools like pyhash, pybind11, and others. ### Why it matters: This project offers a novel approach to multitask learning by leveraging diffusion models and expert denoisers, potentially improving the efficiency and generalization of policy-based multitask learning systems.
AutoHotkey 127 stars

kdalanon/LLM-AutoHotkey-Assistant

An AutoHotkey v2 application that utilizes OpenRouter.ai to seamlessly integrate Large Language Models into your daily workflow. Process texts with customizable prompts by pressing a hotkey and interact with multiple AI models simultaneously.

### Overview: An AutoHotkey v2 application that integrates Large Language Models (LLMs) into your daily workflow. It allows you to process texts with customizable prompts by pressing a hotkey and interact with multiple AI models simultaneously. ### Key Features: - **Text Processing with Hotkeys**: Highlight any text and press a hotkey to access AI-powered text processing. - Summarize: https://github.com/user-attachments/assets/c9bab953-ee7f-478a-b381-bbc39aeaa735 - Translate: https://github.com/user-attachments/assets/b8f01752-0791-4332-a3ac-fac4b82bb74d - Define: https://github.com/user-attachments/assets/a9b43770-f7a3-4c24-9cfd-c9391be0abf6 - Type custom prompts manually: https://github.com/user-attachments/assets/34da227d-e3ec-40e1-9084-432f75e3f99c - Chat: https://github.com/user-attachments/assets/2ffaf8e8-d07b-4ea1-a0e3-5c5fb850da4b - Copy: https://github.com/user-attachments/assets/1682f45e-f3
HTML 127 stars

kehanlu/DeSTA2

Code and model for ICASSP 2025 Paper "Developing Instruction-Following Speech Language Model Without Speech Instruction-Tuning Data"

### Overview: This project introduces DeSTA2, a novel approach for developing instruction-following speech language models (SLMs) that do not rely on speech instruction-tuning data. The model leverages text-based large language models to generate speech-text pairs from rich metadata, achieving impressive performance on speech understanding benchmarks while preserving advanced reasoning capabilities. ### Key Features: - **DeSTA2**: A speech language model designed to follow instructions without requiring speech instruction-tuning data. - **Rich Metadata**: Utilizes metadata rich in context and information to generate speech-text pairs. - **Performance**: Achieves high accuracy in speech understanding tasks. - **Model**: Available on Hugging Face as `DeSTA-ntu/DeSTA2-8B-beta`. - **Dataset**: Available on Hugging Face as `DeSTA-ntu/DeSTA2-Llama3-8B-Instruct`. ### Tech Stack: - **Model**: Developed using Hugging Face Transformers and other relevant libraries. - **Dataset**: Created and managed using Hugging Face datasets. ### Why It Matters: - **Innovation**: Breaks the traditional paradigm of speech language model development by eliminating the need for speech instruction-tuning data. - **Performance**: Maintains high performance in speech understanding benchmarks while retaining advanced reasoning capabilities. - **Versatility**: Can be applied to various speech understanding tasks without the need for additional speech instruction-tuning data. ### Repository: - **Repository**: `https://github.com/kehanlu/DeSTA2` - **Stars**: 127 - **Description**: Code and
Python 127 stars

lxa9867/ControlVAR

This is the official implementation for ControlVAR.

Overview: ControlVAR is an official implementation of Controllable Visual Autoregressive Modeling, focusing on exploring the capabilities of visual autoregressive models. Key features: - Pretrained VQVAE model for faster speed, shorter length, and better FID metrics. - Pseudo-labeled ImageNet dataset for training and evaluation. - Support for Intel HPU training, making it compatible with GPU versions. - Pretrained checkpoints available for download. Tech stack: Python. Why it matters: ControlVAR provides a comprehensive solution for researchers and developers interested in exploring the potential of visual autoregressive models, offering improved performance and compatibility with modern hardware. Repository: lxa9867/ControlVAR Primary language: Python GitHub stars: 127 GitHub description: This is the official implementation for ControlVAR. README excerpt: [ControlVAR: Exploring Controllable Visual Autoregressive Modeling](https://arxiv.org/pdf/2406.09750) - Xiang Li, Kai Qiu, Hao Chen, Jason Kuen, Zhe Lin, Rita Singh, Bhiksha Raj - arXiv paper - 2024 [![arXiv](https://img.shields.io/badge/arXiv%20paper-2404.02905-b31b1b.svg)](https://arxiv.org/pdf/2406.09750)&nbsp; <p align="center"><img src="Illustration.png" width="700"/></p
Unknown 127 stars

sangyun884/Diffusion-Models-Seminar

### Overview: This repository contains a collection of presentations and discussions related to diffusion models, a class of generative models that can generate images, videos, and other types of data by learning from a series of random perturbations. The presentations cover recent advancements in denoising diffusion probabilistic models, improved models, and applications in image synthesis and text-to-image synthesis. ### Key features: - Presentations on various diffusion models and their improvements. - Discussion on the use of diffusion models in image synthesis and text-to-image synthesis. - Talks on conditioning methods and deep unsupervised learning using non-equilibrium thermodynamics. ### Tech stack: Python, Jupyter Notebook, LaTeX ### Why it matters: These presentations highlight the importance of diffusion models in generating high-quality images and videos, making them valuable tools in fields such as computer vision and artificial intelligence. The discussions also underscore the potential of these models in addressing complex problems in image synthesis and time series imputation.
Python 127 stars

teddylee777/langchain-teddynote

LangChain 을 더 쉽게 구현하기 위한 유틸 함수, 클래스를 만들어서 패키지로 배포하였습니다.

Overview: This repository contains utility functions and classes for simplifying the implementation of LangChain, a popular AI library for natural language processing. Key features: - StreamResponse: Provides a function for streaming responses. - LangSmith tracking: Configures LangSmith for logging and tracking. - MultiModal: Enables multi-modal models for generating responses with images. - DeepL translation: Integrates the DeepL translation API for Korean to English translation. - Kiwi Tokenizer: A tokenizer for Korean text. - Synapsoft DocuAnalyzer: Converts PDFs to markdown, XML, and JSON formats. - OpenAI Assistant V2: Implements a RAG system with a specific set of instructions. Tech stack: Python Why it matters: This package simplifies the implementation of LangChain by providing various utility functions and classes, making it easier for developers to integrate LangChain into their projects. GitHub stars: 127 Repository: teddylee777/langchain-teddynote
Python 127 stars

uzaymacar/comparatively-finetuning-bert

Comparatively fine-tuning pretrained BERT models on downstream, text classification tasks with different architectural configurations in PyTorch.

Overview: This repository is designed to fine-tune pre-trained BERT models for text classification tasks, focusing on reducing the network depth while maintaining high accuracy. It aims to help developers optimize BERT for specific tasks without the full 12-layer architecture. Key features: - Fine-tuning BERT models for text classification tasks - Minimizing network depth without sacrificing performance - Using PyTorch and PyTorch-Transformers for implementation Tech stack: Python Why it matters: This approach can help developers reduce training time, fit models on modest GPU and storage resources, and explore BERT for specific tasks more efficiently.
Python 126 stars

AMD-AGI/Nitro-1

This repo provides a working re-implementation of Latent Adversarial Diffusion Distillation by AMD

Overview: This repository provides a re-implementation of the Latent Adversarial Diffusion Distillation method, specifically for building efficient text-to-image generation models. It includes models for both PixArt and Stable Diffusion, optimized for AMD Instinct GPUs. Key features: - Nitro-1-PixArt (1024px DiT-based model) - Nitro-1-SD (512px UNet-based model) Tech stack: PyTorch Why it matters: This implementation allows researchers and developers to advance research in the field of text-to-image generation by providing a more efficient alternative to popular models like Stable Diffusion 2.1, while maintaining performance close to the original models.
HTML 126 stars

AgustiPuigserver/opus-prompt-architect

Best Prompt Engineering Tools for 2026 AI Workflow Optimization

Overview: MythosCraft is an experimental prompt architecting environment designed to transform how you sculpt language for artificial intelligence, focusing on resonance and alchemical composition. Key features: - Dynamic Context Anchoring - Tone Matrix Engine - Trigger-Response Mapping - Version-History Weaving - Export in 12+ Formats Tech stack: HTML Why it matters: MythosCraft offers a full-spectrum prompt development suite that optimizes language structure for high-performance AI interactions. Repository: AgustiPuigserver/opus-prompt-architect GitHub stars: 125
Python 126 stars

CORE-Labet/CORE

CORE is a plug-and-play conversational agent for any recommender system.

Overview: CORE is a plug-and-play conversational agent designed for any recommender system, offering a comprehensive, flexible, and efficient solution. Key features: - Comprehensive: Provides a data manager, offline trainer, and online checker. - Flexible: Integrates seamlessly into any recommender system. - Efficient: Utilizes an online decision tree algorithm for uncertainty minimization. Tech stack: Python. Why it matters: CORE simplifies the integration of conversational agents with recommender systems, making it a versatile tool for enhancing recommendation platforms.
Python 126 stars

ChunjingXiao/DiffAD

Imputation-based Time-Series Anomaly Detection with Conditional Weight-Incremental Diffusion Models, KDD 2023

Overview: This repository contains the code for a paper presenting an imputation-based time-series anomaly detection method using conditional weight-incremental diffusion models. The paper was presented at KDD 2023. Key features: - Conditional Weight-Incremental Diffusion Models for Anomaly Detection - PooledServer Metrics (PSM) dataset for training and testing - Mars Science Laboratory Rover (MSL) dataset for testing - Soil Moisture Active Passive (SMAP) dataset for testing - Server Machine Dataset (SMD) for testing - Secure Water Treatment (SWaT) dataset for testing Tech stack: Python Why it matters: This method offers robust anomaly detection for time-series data, particularly in scenarios involving imputation and conditional weight-incremental diffusion models, making it suitable for various applications requiring real-time anomaly detection in time-series data.
JavaScript 126 stars

Drix10/ai-resources

100,000+ AI briefings in 41 categories: models, dev tools, agents, research, funding. New briefings are fact-checked against their source post and link to it. Plain Markdown, CC BY 4.0. MCP server + Agent Skill let Claude, Cursor or VS Code ask for the latest. Read at blogs.drix10.com

### Overview: AI Resources is a comprehensive collection of 100,000+ AI briefings categorized into 41 distinct topics, including models, development tools, research, funding, robotics, climate, quantum, security, and crypto. The briefings are meticulously checked against their sources, ensuring accuracy and reliability. ### Key Features: - **Checked, Not Just Summarized**: Every claim in the briefings is verified against the original source post. - **Every Claim Links to Source**: The original posts and pages are linked directly, ensuring credibility. - **41 Categories**: Covers models, development tools, research, funding, robotics, climate, quantum, security, and crypto. - **Ready for Your Agent**: Includes MCP server, Agent Skill, JSON API, and an llms.txt file for easy integration into AI agents. ### Tech Stack: - **JavaScript** ### Why It Matters: AI Resources provides a reliable and comprehensive source of information on AI, making it an invaluable tool for developers and researchers. The detailed and verified briefings help ensure accurate and up-to-date knowledge in the field of AI. ### Repository: Repository: Drix10/ai-resources Stars: 122 Description: 100,000+ AI briefings in 41 categories: models, dev tools, agents, research, funding. New briefings are fact-checked against their source post and link to it. Plain Markdown, CC BY 4.0. MCP server + Agent Skill let Claude, Cursor, or VS Code ask for the latest
Unknown 126 stars

EvelynFan/AWESOME-MER

🔆 📝 A reading list focused on Multimodal Emotion Recognition (MER) 👂👄 👀 💬

### Overview: AWESOME-MER is a comprehensive reading list dedicated to Multimodal Emotion Recognition (MER). This project aims to provide insights and resources for researchers, developers, and enthusiasts interested in the field of emotion recognition using multiple modalities such as audio, facial expressions, and text. ### Key Features: - **Datasets**: Contains a variety of datasets including CMU-MOSEI, ASCERTAIN Dataset, EMOTIC Dataset, BP4D+, Multimodal Spontaneous Emotion Database (BP4D+), EmotiW Database, LIRIS-ACCEDE Database, CREMA-D, SEMAINE Database, MAHNOB-HCI, IEMOCAP Database, and eNTERFACE Dataset. - **Challenges**: Covers challenges related to multimodal emotion recognition, including MMER, EmotiW, AVEC, OMGR, MEC, and EmoPain. - **Projects**: Lists projects such as CMU Multimodal SDK, Real-Time Multimodal Emotion Recognition, MixedEmotions Toolbox, and End-to-End Multimodal Emotion Recognition. ### Tech Stack: Python, TensorFlow, PyTorch, and OpenCV - **Python**: Used for data processing and analysis. - **TensorFlow**: For deep learning models. - **PyTorch**: For building and training neural networks. - **OpenCV**: For image processing and computer vision tasks. ### Why It Matters: This reading list and projects are crucial for researchers and developers looking to advance the field of multimodal emotion recognition
Python 126 stars

LIYUESEN/druggpt

DrugGPT: A GPT-based Strategy for Designing Potential Ligands Targeting Specific Proteins

### Overview: DrugGPT is a GPT-based strategy designed to design potential ligands targeting specific proteins. Leveraging the power of deep learning language models, it aims to accelerate the drug development process by optimizing ligand design. ### Key Features: - Ligand Energy Minimization - Multiple Ligand Energy Minimization - Protein-Ligand Binding Data Learning - Novel Protein Scaffold Design - Repurposing Existing Ligands ### Tech Stack: Python ### Why It Matters: DrugGPT offers a novel approach to ligand design, significantly improving the efficiency and effectiveness of the drug development process. Its ability to discover new molecules and repurpose existing ones makes it a valuable tool in the pharmaceutical industry. ### Repository: - **Repository**: LIYUESEN/druggpt - **GitHub Stars**: 126 - **Primary Language**: Python - **GitHub Description**: DrugGPT: A GPT-based Strategy for Designing Potential Ligands Targeting Specific Proteins - **README Excerpt**: ```shell # Clone ```shell git clone https://github.com/LIYUESEN/druggpt.git cd druggpt ``` > Or you can just click *Code>Download ZIP* to download this repo. ```shell ### Create Python virtual environment ```shell conda create -n druggpt python=3.8 conda activate druggpt ``` ### Install PyTorch and other requirements ```shell pip install torch torchvision torchaudio --index-url https://download.pytorch.org/
Jupyter Notebook 126 stars

PouriaRouzrokh/LatteReview

This is the repository for the LatteReview python package. LatteReview is a low-code Python package designed to automate systematic literature review processes through AI-powered agents.

Overview: LatteReview is a Python package designed to automate systematic literature review processes through AI-powered agents, making the review of numerous research articles a pleasant, efficient experience. Key features: - Decision reviewers run on Perplexity's pplx-decider and OpenAI's gpt-6-luna through their new Decisions APIs. - Decision models compared on 11,793 articles: pplx-decider ranked best, followed by Jev and v1 LLM reviewers. - Perplexity's models as LLM reviewers: `LiteLLMProvider` reaches Sonar and other vendors' models through Perplexity's new Agent API. - Correct costs for per-request fees: `LiteLLMProvider` now uses the actual API cost reported. Tech stack: Python 3.10+ and Jupyter Notebook. Why it matters: LatteReview streamlines the literature review process, making it more efficient and less time-consuming, allowing researchers to focus on their core work while benefiting from the expertise of AI-powered agents.
Python 126 stars

amazon-science/mix-generation

MixGen: A New Multi-Modal Data Augmentation

Overview: MixGen is a joint data augmentation technique designed for vision-language representation learning, enhancing the efficiency of training models with limited data. Key features: - Joint data augmentation for vision-language representation learning - Plug-and-play implementation, minimal code changes required - Adopted ALBEF as an illustrating example for integration Tech stack: Python Why it matters: MixGen improves the data efficiency of vision-language representation learning, making it a valuable tool for researchers and practitioners aiming to enhance model performance with limited training data.
Python 126 stars

anam-org/metaxy

Pluggable metadata management framework for versioned incremental multimodal data/ML pipelines.

Overview: Metaxy is a pluggable metadata management framework designed for versioned incremental multimodal data/ML pipelines. It tracks lineage and versioning across complex computational graphs for multimodal datasets, caching every sample and scaling to handle millions of them. Key features: - Tracks partial data dependencies - Detects prunable updates - Standardizes dependency specification, versioning, and metadata management - Integrates with popular tools like PyPI, S3, and DataFrame libraries Tech stack: Python Why it matters: Metaxy provides a standard instrument for any kind of multimodal or purely tabular incremental pipelines, standardizing dependency specification, versioning, partial data dependencies, and metadata manipulations. It is highly reliable and tested across all supported Python versions and platforms. Repository: anam-org/metaxy GitHub stars: 126 GitHub description: Pluggable metadata management framework for versioned incremental multimodal data/ML pipelines.
Python 126 stars

apimart-awesome-ai-api-proxy/nano-banana-pro-prompts

Nano Banana Pro prompts: tested prompt recipes with example renders, aspect ratios, the model id (gemini-3-pro-image-preview) and the cost each render reported.

### Overview: Nano Banana Pro Prompts: A Collection of 8 Tested Recipes with Outputs ### Key Features: - **Model ID:** `gemini-3-pro-image-preview` - **Aspect Ratios:** 1:1, 16:9, 3:4, 8:5, 2:3, 1:2, 1:1.5, 1:1.33 - **Costs:** $0.03 per image ### Tech Stack: - Python ### Why It Matters: Nano Banana Pro Prompts offer a comprehensive set of 8 recipes tested with outputs, aspect ratios, and costs for the model `gemini-3-pro-image-preview`. This resource is valuable for developers looking to optimize their API calls and ensure accurate rendering.