Projects

Overview: Holo-World is a unified camera, object, and weather control system designed for a video world model, aiming to enhance the realism and functionality of virtual environments. Key features: - Unified Camera Control - Real-time Object Tracking - Weather Simulation Integration Tech stack: Python, OpenCV, PyTorch Why it matters: Holo-World provides a comprehensive solution for creating more immersive and interactive virtual environments, enhancing the user experience in applications such as video games, simulations, and virtual reality. Repository: XiangchenYin/Holo-World Primary language: Unknown GitHub stars: 50 GitHub description: Unified Camera, Object and Weather Control for Video World Model

Overview: FFHQ-Text is a small-scale face image dataset designed for text-to-face generation and manipulation, focusing on the attributes of 760 female images from the FFHQ dataset. Key features: - 13 multi-valued facial element groups - 9 descriptions for each selected image in FFHQ - 760 images in PNG format - 766 KB of text annotations Tech stack: Python Why it matters: This dataset provides a controlled environment for exploring facial attributes and text-to-image manipulation, making it valuable for research in vision-related tasks. Repository: Yutong-Zhou-cv/FFHQ-Text_Dataset

Overview: Gemini Nano Banana 2.1 API provides a comprehensive set of image generation and editing capabilities, including text-to-image and reference-image editing, with a focus on Google's official and extended routes. This API is compatible with OpenAI, offering a single API key for all models and routes, and flexible pricing based on usage. Key features: - Compatible with OpenAI, offering a single API key for all models and routes - Supports text-to-image and reference-image editing - Flexible pricing based on usage, with real-time pricing updates - Supports multiple output resolutions and image URLs as reference - Asynchronous processing, with task IDs for easy monitoring Tech stack: Python Why it matters: This API simplifies the process of generating high-quality images and editing them, making it ideal for developers and content creators looking to integrate advanced image processing into their projects without the complexity of multiple APIs or expensive subscriptions. Repository: a-kiecshv/gemini-nano-banana-2-1-api-id GitHub stars: 45 GitHub description: Gemini Nano Banana 2.1 API (`gemini-nano-banana-2.1`): model id, pricing, parameters and copy-paste examples for the OpenAI-compatible APIMart endpoint.

Overview: This repository contains tested image prompt recipes for the GPT Image 2.5 Sunburst model, including 8 recipes with examples, ratios, and costs. Key features: - 8 tested image prompt recipes - Example renders, ratios, and costs for the GPT Image 2.5 Sunburst model - Model: `gpt-image-2.5-ext` - Effective price: $0.0106 per image Tech stack: Python Why it matters: This repository provides practical tips for crafting effective image prompts, ensuring high-quality results and cost efficiency.

### Overview: X-LXMERT is a multi-modal transformer model designed to generate images from text, aiming to bridge the gap between image captioning and visual question answering. This project introduces X-LXMERT, an extension of the LXMERT model with improved training techniques, achieving state-of-the-art performance in generating rich and semantically meaningful imagery. ### Key features: - **Image Generation**: X-LXMERT generates high-quality images from text inputs, rivaling state-of-the-art generative models. - **Question Answering**: The model's question answering abilities remain comparable to LXMERT, demonstrating its effectiveness in downstream tasks. - **Training Refinements**: The model's image generation capabilities are enhanced through specific training refinements, addressing the limitations of LXMERT. ### Tech stack: - **Python**: The primary programming language used for the project. - **GitHub Stars**: 50, indicating the project has gained significant traction and interest among developers. ### Why it matters: X-LXMERT represents a significant advancement in the field of multimodal transformers, offering a new approach to generating images from text inputs. This technology has the potential to revolutionize applications in computer vision, image generation, and natural language processing, making it a valuable contribution to the AI community. ### Repository: The project is hosted on GitHub under the repository `allenai/x-lxmert`. The primary language used is Python, and the project has gained 50 GitHub stars, indicating its popularity and interest among developers. ### Install: To get started with X-LXMERT, you need to install the

Overview: AionDB is a Rust database that supports PostgreSQL wire protocol, disk storage, and advanced features like vector and graph access paths. Key features: - PostgreSQL compatibility - Disk-based storage - Vector and graph access paths Tech stack: Rust Why it matters: AionDB offers a flexible and powerful database solution that can be used in various applications, providing a PostgreSQL-compatible interface and advanced querying capabilities. Repository: ayoubnabil/aiondb GitHub stars: 50 GitHub description: PostgreSQL-compatible SQL, graph, and vector database built from scratch in Rust.

Overview: DiffusionGuard is a robust defense mechanism designed to protect images from diffusion-based image editing, as presented in the ICLR 2025 paper "DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing". Key features: - Protection against diffusion-based image inpainting - Utilizes adversarial perturbations for image enhancement - Compatible with various models and configurations Tech stack: Python Why it matters: DiffusionGuard offers a significant defense against malicious image editing, ensuring the integrity and authenticity of digital content. Repository: choi403/DiffusionGuard Primary language: Python GitHub stars: 50 GitHub description: DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025) Dataset: Available at [this Google Drive link](https://drive.google.com/file/d/1EDVdqkdSY98xAGPJP0YDDC7SY25lX4qs/view)

Overview: The Friendli Client is a Python library designed to simplify interaction with endpoint services provided by Friendli Suite, the ultimate solution for serving generative AI models. It supports both synchronous and asynchronous operations, making it easy to integrate powerful AI capabilities into applications. Key features: - Synchronous and asynchronous operations - Support for Friendli Suite's serverless and dedicated endpoints - Easy integration with environment variables or personal access tokens - Provides access to popular open-source models like Llama 3.1 - Offers pay-per-token billing for serverless endpoints Tech stack: Python Why it matters: The Friendli Client simplifies the process of interacting with generative AI models, making it accessible and efficient for developers looking to leverage cutting-edge AI technologies.

Overview: Strata Manager simplifies the setup and management of Niko1221/Strata, a local large language model inference engine, by providing guided setup, model management, service controls, chat, API monitoring, and tuning features. Key features: - Guided setup - Model downloads and configurations across Strata versions - Service controls - Chat and API connections with online verification - Usage statistics and monitoring - Tuning and benchmarking - Opt-in public leaderboard Tech stack: Rust Why it matters: Strata Manager streamlines the process of setting up and managing Strata, making the local inference engine more accessible and easier to use.

### Overview: This project focuses on the development of a Time-Series Hybrid Autoencoder (TSHAE) model for estimating the remaining useful life (RUL) of equipment. The TSHAE is designed to predict degradation trajectories and generate synthetic sensor data using a combination of autoencoder-based and diffusion models. The project includes contributions to the field of RUL prediction, particularly in the context of Remaining Useful Life estimation and sensor data generation. ### Key Features: - **TSHAE Model**: A novel autoencoder-based model capable of estimating differentiable degradation trajectories in the latent space. - **Trajectory Extrapolation**: An unsupervised method for extending initial degradation trajectories using latent spatial properties. - **Sensor Signal Generation**: Utilizing extrapolated trajectories to generate realistic input sensor signals for synthetic data generation. ### Tech Stack: - Python - PyTorch ### Why it Matters: The TSHAE model and its contributions to RUL prediction and sensor data generation are significant advancements in the field of equipment maintenance and reliability engineering. By providing accurate and interpretable degradation trajectories, the TSHAE model helps in predicting the remaining useful life of equipment, thereby enhancing the reliability and efficiency of maintenance strategies. Additionally, the ability to generate synthetic sensor data from these trajectories can be crucial for training and validating models in various applications, including predictive maintenance and reliability engineering.

### Overview: This project evaluates the performance of the RAG (Retrieval-Augmented Generation) model enhanced with a Cross Encoder Reranker, using the Zephyr 7B Alpha 5 bit Quantised model (BAAI/bge-small-en-v1.5). The model was tested with a comprehensive document titled "Brains in Space: Effects of Spaceflight on the Human Brain and Behavior" spanning 311 pages. ### Key features: - Advanced RAG with Cross Encoder Reranker - RAG without Cross Encoder Reranker ### Tech stack: Jupyter Notebook ### Why it matters: This evaluation highlights the trade-off between speed and depth of understanding in AI models, underscoring the importance of choosing the right configuration based on the application's requirements.

### Overview: MoteDB 是一个基于 Rust 的 AI 原生嵌入式多模态数据库，专为具身智能和机器人应用设计。它集成了向量、时序、文本和二进制数据类型，提供单个嵌入库中的 ACID 事务、列存储引擎、矢量搜索、全文搜索和空间索引。 ### Key features: - **多模态数据存储**：支持向量、时序、文本和二进制数据类型。 - **ACID 事务**：确保数据的完整性和一致性。 - **列存储引擎**：高效存储和检索数据。 - **矢量搜索**：快速查找向量数据。 - **全文搜索**：支持全文检索。 - **空间索引**：实现高效的地理空间数据查询。 ### Tech stack: Rust ### Why it matters: MoteDB 是一个专为具身智能和机器人应用设计的 AI 原生嵌入式多模态数据库。它提供了一种高效、安全且易于集成的方式，用于存储和检索各种类型的数据，从而加速开发过程并提高系统的性能。

### Overview: An easy-to-use image editor extension for Stable Diffusion Web UI, designed to speed up workflows by integrating seamlessly with other core functions of SD Web UI. ### Key features: - Lightweight and easy-to-use interface - Native integration with Stable Diffusion Web UI (compatible with Forge) - Quick crop, rotate, enhance, and send to img2img - Compatible with A1111 Stable Diffusion Web UI ### Tech stack: Python ### Why it matters: This image editor extension streamlines the process of image manipulation, making it faster and more efficient, especially for users familiar with Stable Diffusion Web UI. ### Repository: sontungdo/sd-image-editor ### GitHub stars: 50 ### Primary language: Python ### GitHub description: An easy-to-use image editor extension for Stable Diffusion Web UI ### Installation: - Install through official index: Extensions -> Available -> Load from - Install with Github link: Extensions -> Install from URL -> Paste `https://github.com/sontungdo/sd-image-editor.git` in the first box ### Example usage: - Crop and adjust color, contrast, brightness - Send result to inpainting ### Todo list: - Add cropping - Add send to image editor buttons - Add Gradio built in cropping tool - Add resize - Add more adjust features and filters

Overview: Lightning ASR is a modular and extensible speech recognition library built on PyTorch Lightning and Hydra, designed to simplify and accelerate research in speech recognition. Key features: - Modular architecture - Extensible design - PyTorch Lightning for high-performance AI research - Hydra for configuration management Tech stack: Python Why it matters: Lightning ASR provides a streamlined solution for researchers looking to implement speech recognition projects, making it easier to scale and iterate on research without introducing bugs.

Overview: This project is an intelligent file organization system that uses computer vision, audio analysis, and adaptive learning to categorize and organize your files efficiently, making it particularly useful for individuals with ADHD. Key features: - Smart Staging: Files are observed for 7 days before being organized. - Confidence-Based Decisions: Modes for different levels of certainty. - Adaptive Learning: The system learns from user corrections. - Bulletproof Deduplication: Three-tier verification for file uniqueness. - Emergency Space Protection: Automatic monitoring and cloud offloading. - Rollback System: Allows undoing any organization action. - Google Drive Integration: Optional cloud storage support. - Plugin System: Extensible with custom classifiers. Tech stack: Python. Why it matters: This system is designed to be intuitive and efficient, making it a valuable tool for managing files in a way that caters to the needs of those with ADHD, ensuring they can organize their files effectively without the stress of constant decision-making. Repository: thebearwithabite/ai-file-organizer GitHub stars: 50 GitHub description: Intelligent file organization with computer vision, audio analysis, chunking, proactive AI-powered analysis, interactive classification, and adaptive learning.

Overview: MixDQ is a memory-efficient few-step text-to-image diffusion model that utilizes metric-decoupled mixed precision quantization for efficient model compression and acceleration. Key features: - Achieves W4A8 performance with negligible visual quality degradation. - Utilizes mixed precision quantization to reduce memory usage by 2x and latency by 3.4x. - Provides CUDA kernels for hardware acceleration, achieving 1.45x speedup and 2x memory savings. - Open-source Huggingface pipeline for efficient INT8 GPU kernel acceleration, saving 1.94x memory and 4.36x dynamic memory usage. Tech stack: Python, CUDA, Huggingface. Why it matters: MixDQ is a significant advancement in the field of few-step text-to-image diffusion models, offering a practical solution for memory-efficient and accelerated model compression, making it particularly valuable for applications requiring high performance and low resource usage.

Overview: This repository contains the official code for a paper presenting a powerful and flexible method for personalized text-to-image generation using reinforcement learning. Key features: - Personalized Text-to-Image Generation via Reinforcement Learning - Training using 'Look Forward' reward - Training using 'DINO' reward - Inference capabilities Tech stack: Python Why it matters: This project showcases innovative techniques in text-to-image generation, demonstrating the potential of reinforcement learning in personalized image creation. Repository: wfanyue/DPG-T2I-Personalization GitHub stars: 50

### Overview: ECG-Bench is a unified framework designed to benchmark and evaluate generative electrocardiogram-language models (ELMs). This project aims to provide a comprehensive and standardized approach for testing and comparing the performance of various ELM models across different datasets and tasks. ### Key Features: - **Unified Framework**: A single platform for benchmarking ELMs across various datasets. - **Cross-Dataset Evaluation**: Supports evaluation on multiple datasets such as ECG-QA PTB-XL, MIMIC-IV-ECG, and ECG-Byte. - **Open-Source Repository**: Available on GitHub, with a DOI and Zenodo badge for reproducibility. ### Tech Stack: - Python ### Why It Matters: ECG-Bench offers a standardized way to evaluate ELM models, ensuring that results are comparable across different datasets and models. This makes it an essential tool for researchers and developers in the field of electrocardiogram-language models, facilitating fair and accurate comparisons of model performance. ### Repository: - **Repository**: willxxy/ECG-Bench - **Stars**: 50 - **Description**: A Unified Framework for Benchmarking Generative Electrocardiogram-Language Models (ELMs) ### Primary Language: - **Primary Language**: Python ### GitHub Stars: - **GitHub Stars**: 50

Overview: 小优AIGC是抖音官方的AI影视创作工作流，提供从想法到完整故事、再到电影分镜和视觉方案的全流程服务。通过编剧、分镜导演和视觉导演的接力，将创意转化为电影级别的生图提示词。 Key features: - 编剧 Skill（story-architect） - 分镜 Skill（libtv-storyboard） - 视觉导演 Skill（cinematic-director） Tech stack: 使用Claude Code / Codex / Hermes等AI Agent。 Why it matters: 该工作流为创作者提供了一个完整的AI影视创作流程，从创意到最终呈现，大大简化了创作过程，有助于提高效率并确保高质量的电影制作。

### Overview: This project utilizes AI-driven Non-Player Characters (NPCs) that can control their actions and dialogue in a game environment. The NPCs are generated using a combination of Cohere's Command model, Unreal Engine, and other tools, enabling them to perform actions and generate dialogue based on user input. ### Key Features: - **NPCs with Dialogue and Actions**: NPCs can respond to questions and perform actions, such as casting spells or performing specific actions. - **Real-Time Interaction**: The NPCs can interact with the game environment in real-time, enhancing the gaming experience. - **Voice Acting**: ElevenLabs generates voice dialogue, which is then imported into the game using the Runtime Audio Importer. - **Animation and Music**: The MetaHuman SDK brings the characters to life with facial animations and AI-generated music. ### Tech Stack: - Python - Unreal Engine - Cohere's Command model - ElevenLabs - AssemblyAI - ChromaDB - Langchain - Unreal Engine - Quixel Bridge - Metahuman SDK - Mixamo - Runtime Audio Importer ### Why It Matters: This project demonstrates the potential of AI in enhancing the gaming experience by allowing NPCs to interact with players in a more engaging and interactive way. It showcases how AI can be integrated into game development to create more dynamic and immersive environments. ### GitHub Stars: 49 ### Repository: The repository contains the Python script and a demo file that can be imported into Unreal Engine to see the logic and blueprint of the NPCs. It also includes other files that can