Projects

### Overview: This collection includes 252 curated Z-Image Turbo prompts across 9 categories, featuring preview images to help you see the results before running the prompts. ### Key features: - **Portrait & Avatar** - No. 1: Editorial Studio Portrait - No. 2: Candid Street Portrait - No. 3: Corporate Headshot - **Product & E-commerce** - No. 4: Product Detail - No. 5: Product Shot - No. 6: Product Illustration - **Poster & Typography** - No. 7: Poster Text - No. 8: Typography Design - **Infographic & Diagram** - No. 9: Infographic Element - No. 10: Diagram Illustration - **Anime & Illustration** - No. 11: Anime Character - No. 12: Illustration Scene - **3D & Material** - No. 13: 3D Model - No. 14: Material Texture - **Food & Lifestyle** - No. 15: Food Illustration - No. 16: Lifestyle Scene - **Architecture & Interior** - No. 17: Interior Design - No. 18: Architecture Sketch - **Nature & Landscape** - No. 19: Landscape Scene - No. 20: Nature Illustration ###

Overview: AUGUstus is an autonomous driving model designed for unified generation and understanding, utilizing 512 NVIDIA H800 GPUs and 318K+ video hours of training data. It achieved a deployment speed of 12 Hz on a real vehicle and demonstrated state-of-the-art performance on various benchmarks, including NAVSIM v1, NAVSIM v2, and nuScenes. Key features: - Unified Generation and Understanding Model - 512 NVIDIA H800 GPUs for ~2 days of training - 318K+ video hours of driving data - 513B+ training tokens Tech stack: Python Why it matters: AUGUstus is a significant advancement in autonomous driving, showcasing the potential of unified generation and understanding in achieving high deployment speeds and impressive performance metrics.

### Overview: MRGen: Segmentation Data Engine for Underrepresented MRI Modalities MRGen is a segmentation data engine designed for underrepresented MRI modalities. It aims to address the challenges in medical image analysis by providing a comprehensive framework for training, testing, and evaluating segmentation models on various MRI modalities. ### Key Features: - **Data Engine**: MRGen serves as a data engine for MRI modalities, enabling efficient data processing and management. - **Modality Support**: Supports multiple MRI modalities including T1, T2, DWI, and more. - **Training Pipeline**: Includes three stages: latent encoding, text-guided generation, and mask-conditioned generation. - **Model Training**: Utilizes advanced architectures like VAE, BiomedCLIP, and controlnet for high-quality image synthesis. - **Privacy Protection**: Ensures data privacy by only releasing JSON files and providing download links for authorized users. ### Tech Stack: Python The project is implemented using Python 3.8, with dependencies on PyTorch, xformers, diffusers, accelerate, transformers, and triton. ### Why It Matters: MRGen addresses the limitations in medical image analysis by providing a robust framework for training segmentation models on underrepresented MRI modalities. This makes it invaluable for researchers and clinicians working with diverse patient populations and underrepresented datasets.

### Overview: EstateWise is a full-stack, monorepo AI/ML chatbot & data analytics project designed for real estates in Chapel Hill, NC and its surrounding areas. It leverages advanced machine learning techniques and comprehensive data analysis to provide personalized property recommendations. ### Key Features: - **Secure Authentication**: Ensures user privacy and data security. - **Conversation Management**: Facilitates natural language interactions for property inquiries. - **Personalized Recommendations**: Utilizes advanced algorithms to match user preferences with property details. ### Tech Stack: - TypeScript - Node.js - Express - React - Next.js - Tailwind CSS - Vue - Framer Motion - Three.js ### Why It Matters: EstateWise offers a seamless, AI-driven experience that helps users find their dream homes efficiently and effectively. Its integration of agentic AI, hybrid RAG, k-Means clustering, and other advanced techniques ensures accurate and personalized recommendations, making it a valuable tool for real estate professionals and enthusiasts alike. ### Repository: hoangsonww/EstateWise-Chapel-Hill-Chatbot ### GitHub Stars: 41 ### GitHub Description: 🏠 An AI real estate app featuring secure auth, conversation management, & personalized property recommendations. Powered by Agentic AI, RAG (w/ Pinecone), GraphRAG (w/ Neo4j), MCP, kNN, K‑Means Clustering, CoT, LangGraph, and a Mixture‑of‑Experts ensemble to deliver accurate, effective, and scalable AI insights

### Overview: JAV Trans是一款为JAV日语视频生成日文、中文或双语字幕的工具。它支持本地音视频识别、时间轴对齐及API或本地模型翻译；网页界面操作，视频与音频留在本机。适用于Windows + NVIDIA显卡的环境。 ### Key features: - 本地音视频识别 - 时间轴对齐 - API或本地模型翻译 - 支持日文、中文或双语字幕 - 网页界面操作 - 仅日文、双语字幕或质量报告模式 - 仅日文模式不发送字幕 - 本地翻译模型支持 - 本地翻译模型下载路径为`models\` ### Tech stack: Python ### Why it matters: JAV Trans简化了JAV日语视频的字幕生成过程，适用于需要高质量字幕的用户，特别是对日语学习者和翻译爱好者来说是一个非常实用的工具。

### Overview: This project focuses on soft value-based decoding in diffusion models, specifically designed for generating natural sequences like DNA (enhancers) and RNA (5'UTRs). The implementation includes pre-trained models and datasets, aiming to enhance the generation of biological sequences for controlled diffusion processes. The algorithm is summarized in a table and figure, and the code is designed to be easily installable and reusable. ### Key features: - **DNA Enhancers**: Generates natural enhancers with high activities in HepG2 using soft-value based decoding. - **RNA 5'UTRs**: Generates natural 5'UTRs with high activities in HepG2 using soft-value based decoding. - **Pre-trained Models**: Utilizes masked diffusion models to generate sequences. - **Dataset**: Dataset from [Gosai et al., 2023](https://pubmed.ncbi.nlm.nih.gov/37609287/) and [Sample et al., 2019](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE114002). ### Tech stack: Python ### Why it matters: This project is significant in the field of computational biology and machine learning, as it provides a method for generating high-quality sequences that can be used in various applications, including drug discovery and personalized medicine. The use of soft-value based decoding ensures that the generated sequences are more realistic and closer to natural sequences, which is crucial for applications in biology and medicine.

Overview: TemporalBench is a benchmark designed to evaluate fine-grained temporal understanding and reasoning for multimodal video models. It consists of ∼10K video question-answer pairs sourced from ∼2K high-quality human-annotated video captions, capturing detailed temporal dynamics and actions. Key features: - Fine-grained temporal understanding - Multimodal video models - Detailed temporal dynamics and actions Tech stack: Python Why it matters: TemporalBench provides a comprehensive benchmark for evaluating the capabilities of video understanding models, particularly in handling complex temporal sequences and actions. This is crucial for advancing research in video analysis and multimodal AI. Repository: mu-cai/TemporalBench Primary language: Python GitHub stars: 41 GitHub description: TemporalBench: Benchmarking Fine-grained Temporal Understanding for Multimodal Video Models README excerpt: TemporalBench is a video understanding benchmark designed to evaluate fine-grained temporal understanding and reasoning for multimodal video models. It consists of ∼10K video question-answer pairs sourced from ∼2K high-quality human-annotated video captions, capturing detailed temporal dynamics and actions.

### Overview: Botality II is a Telegram bot built with AIogram, designed for local inference of Stable Diffusion, text-to-speech, and large language models. It supports remote service support and integrates with multiple speech and text recognition engines. ### Key Features: - **Stable Diffusion**: Uses the Stable Diffusion WebUI API for image generation. - **Text-to-Speech**: Supports TTS (VITS) and OS voices, as well as integrated speech recognition engines. - **Speech Recognition**: Compatible with various speech recognition tools. - **Large Language Models**: Supports LLMs like LLaMA, GPT-J, and GPT-2, including assistant mode via Lora models and multimodal support via LLaMA-Adapter. - **TTA (Text-to-Audio)**: Experimental support for text-to-audio conversion. - **Memory Management**: Tracks and manages models to optimize performance on M1 Macs. - **User Queues and Delayed Processing**: Ensures smooth user experience with user-based queues and delayed task processing. - **Multiple Access Scopes**: Supports WL, BL, Both, and Admin-only modes. ### Tech Stack: Python ### Why It Matters: Botality II offers a comprehensive solution for local inference of complex models, making it ideal for developers and enthusiasts looking to integrate advanced AI capabilities into their projects without the need for cloud-based services. ### Repository: remixer-dec/botality-ii GitHub Stars: 41 GitHub Description: Telegram bot for local inference of Stable Diff

### Overview: This utility evaluates the performance of Automatic Speech Recognition (ASR) systems by comparing the transcriptions generated by the system against a human-generated reference. It calculates metrics such as Word Error Rate (WER), Word Information Loss (WIL), Levenshtein Distance, and more, providing a quick assessment of the system's accuracy. ### Key Features: - **Word Error Rate (WER)**: Measures the percentage of words that are incorrectly transcribed. - **Word Information Loss (WIL)**: Approximates the proportion of word information lost in the transcribed text. - **Levenshtein Distance**: Compares the number of word-level insertions, deletions, and mismatches. - **Phrase Level Metrics**: Compares the number of phrase-level insertions, deletions, and mismatches. - **Color Highlighted Text Comparison**: Visualizes the differences between the original and generated files. - **Pre-Processing**: Normalizes the text by removing speaker names, custom annotations, extra white spaces, quotes, dashes, and punctuations. ### Tech Stack: Node.js ### Why It Matters: This utility is essential for developers and researchers who need to evaluate the performance of ASR systems quickly and efficiently. It helps in identifying areas of improvement and validating the effectiveness of different speech recognition techniques, making it a valuable tool in the field of speech processing and natural language understanding.

### Overview: CasTex is a novel approach for text-to-texture synthesis, specifically designed to generate realistic texture maps under various lighting conditions. This project addresses the limitations of existing methods by proposing a cascaded diffusion model architecture that directly synthesizes textures without the need for implicit texture parameterization, leading to improved quality and efficiency. ### Key Features: - **Cascaded Text-to-Texture Synthesis (CasTex):** A cascaded diffusion model architecture that directly synthesizes textures without the need for implicit texture parameterization. - **Explicit Texture Maps:** The use of explicit texture maps to generate high-quality textures. - **Physically-Based Shading:** Incorporation of physically-based shading to ensure realistic appearance under varying lighting conditions. ### Tech Stack: - Python ### Why It Matters: CasTex significantly outperforms state-of-the-art optimization-based solutions on public texture synthesis benchmarks, offering improved quality and efficiency in text-to-texture synthesis tasks.

Overview: This project is a portable speech recognition (ASR) tool for Windows, specifically Microsoft VibeVoice, that provides high-quality text recognition, speaker diarization, and temporal segmentation. It is fully translated into Russian and supports NVIDIA GPUs, including the ability to emulate 4-bit quantization for memory savings. Key features: - High-quality text recognition - Speaker diarization - Temporal segmentation - Emulation of 4-bit quantization for memory savings - Fully Russian interface - Dark theme interface - Automatic browser opening - Portability - all files in one folder - Support for multiple audio and video formats via FFmpeg - Faster model loading via HuggingFace Xet Tech stack: Python Why it matters: This project is essential for developers and users who need a portable, offline, and high-quality speech recognition tool for Windows, especially for those with NVIDIA GPUs, and it is fully translated into Russian, making it accessible to a wider audience. Repository: timoncool/VibeVoice_ASR_portable_ru Primary language: Python GitHub stars: 41 GitHub description: Portable speech recognition (ASR) for Windows — Microsoft VibeVoice, one-click install, 100% offline, NVIDIA GPU.

### Overview: This repository is a collection of papers, documentation, and codes related to diffusion distillation, focusing on the Diffusion model. It aims to provide a comprehensive resource for researchers and developers interested in this field. ### Key features: - Papers: A diverse range of papers on diffusion distillation methods. - Papers (2023): - "On Architectural Compression of Text-to-Image Diffusion" by Nota-NetsPresso - "Consistency Models" by OpenAI - "Accelerating Diffusion-based Combinatorial Optimization Solvers by Progressive Distillation" by jwrh - "Towards Safe Self-Distillation of Internet-Scale Text-to-Image Diffusion" by nannullna - "Accelerating Diffusion Sampling with Classifier-based Feature" by tcl9876 - "On Distillation of Guided Diffusion" by ruiqixu37 - "SnapFusion: Text-to-Image Diffusion Model on Mobile Devices within Two Seconds" by SnapFusion - "Diff-Instruct: A Universal Approach for Transferring Knowledge From Pre-trained Diffusion Models" by Diff-Instruct - "Fast Sampling of Diffusion Models via Operator" by neuraloperator - "BOOT: Data-free Distillation of Denoising Diffusion Models with Bootstrapping" by bootstrapping - "TRACT: Denoising Diffusion Models with Transitive Closure Time-Distillation" by transitive_closure - "

### Overview: InFi is a library designed for building input filters in resource-efficient inference, specifically for improving the performance of speech recognition models. ### Key features: - **Inference Efficiency**: Enhances the efficiency of inference by reducing unnecessary computations. - **Resource-Efficient Models**: Optimizes the deployment of trained models to minimize resource usage. - **Modular Architecture**: Provides a modular API for building and training input filters. ### Tech stack: Python ### Why it matters: InFi simplifies the process of deploying models for speech recognition tasks, making them more efficient and resource-friendly, thereby improving overall performance and reducing computational costs.

Overview: This repository demonstrates the integration of BGE-M3 and Qdrant for advanced hybrid search and reranking, showcasing the benefits of dense and sparse retrieval combined with ColBERT reranking. Key features: - Use of an all-in-one embedding model (BGE-M3) to generate dense, sparse, and ColBERT token vectors. - Multi-vector approach for superior search quality. - Integration of Qdrant for efficient vector database setup and indexing. Tech stack: Python 3.9+, Docker for Qdrain, Jupyter Notebook. Why it matters: This sample provides a comprehensive demonstration of how to build an advanced search system, highlighting the effectiveness of combining different types of embeddings for optimal search results.

Overview: This repository contains a collection of deep learning projects and theories, including applications in face recognition, neural style transfer, autonomous driving, sign language reading, music generation, translation, speech recognition, and natural language processing, as well as detailed explanations of neural networks and deep learning concepts. Key features: - Deep Neural Network for Image Classification - Initialization - Regularization - Gradient Checking - Optimization Methods - Convolutional Neural Network Tech stack: Jupyter Notebook Why it matters: This repository provides a comprehensive overview of deep learning, covering both theoretical foundations and practical applications, making it an excellent resource for anyone interested in deep learning and neural networks.

### Overview: This project is a TensorFlow implementation of end-to-end Automatic Speech Recognition (ASR) based on the Listen Attention and Spell (LAS) model. It aims to provide a comprehensive and efficient solution for ASR tasks, focusing on the development of a model that does not rely on the `tf.contrib.seq2seq` API. ### Key Features: - **Char/Subword Text Encoding**: Utilizes character and subword text encoding for improved performance. - **MFCC/fBank Features**: Incorporates MFCC/fBank acoustic features with per-utterance CMVN for robust audio representation. - **LAS Training**: Evaluates performance on LibriSpeech dev/test sets. - **Tensorboard Visualization**: Loss, sample text outputs, features, and alignments are visualized using tensorboard. - **TFrecord Dataset Pipeline**: Efficiently processes audio and text data using TensorFlow datasets. - **Batch Testing with Greedy Decoder**: Evaluates model performance on training data. - **Beam Search Decoder**: Provides high-quality text outputs using beam search decoding. - **RNNLM**: Implements a Recurrent Neural Network Language Model for improved performance. ### Tech Stack: - Python ### Why It Matters: This project is an essential resource for developers looking to implement end-to-end ASR models using TensorFlow. It provides a solid foundation for understanding and improving ASR systems, particularly focusing on the LAS model and its implementation without the use of `tf.contrib.seq2seq` API. The inclusion of performance evaluation on LibriSpeech datasets and visualization tools make it a valuable reference for researchers and developers alike. ###

Overview: This project is a Python-based AI virtual assistant that utilizes text-to-speech conversion and speech recognition libraries to provide a conversational interface. Key features: - Text-to-speech conversion using `pyttsx3` - Speech recognition using `SpeechRecognition` - Compatibility with Python 2 and 3 Tech stack: Python Why it matters: This project is designed to be a versatile AI assistant that can help automate tasks and provide a more engaging user experience through voice interactions. Repository: Aman22sharma/Python-AI-Virtual-Assistant Primary language: Python GitHub stars: 40 GitHub description: This is python AI Virtual Assistant. README excerpt:

Overview: FlaxDiff is a library designed to simplify the understanding and implementation of diffusion models, focusing on readability and ease of use. It includes detailed notebooks on DDPM, DDIM, and EDM, providing a comprehensive guide to diffusion models. Key features: - Comprehensive notebooks on DDPM, DDIM, and EDM - Detailed mathematical explanations for diffusion models - Easy-to-understand implementation in Jupyter Notebook - Support for various diffusion techniques and models Tech stack: Jupyter Notebook Why it matters: FlaxDiff aims to demystify the complex mathematics behind diffusion models, making it accessible to developers and researchers alike. By providing clear, step-by-step guides, it helps in replicating and experimenting with the latest research in generative AI, fostering a deeper understanding of the field.

### Overview: Keyaos is an edge-native AI API gateway designed for cost-optimized routing across multiple providers, built on Cloudflare Workers. It simplifies the management of multiple AI services by unifying their endpoints, ensuring the cheapest and most reliable provider is always used. ### Key Features: - **Cost-optimized routing**: Routes every request to the cheapest available provider. - **Automatic failover**: Handles quota exceeded or rate limited scenarios by switching to the next cheapest option. - **Zero-latency streaming**: SSE responses are tee'd and forwarded in real time, providing seamless streaming experience. - **Auto-synced catalog**: Models and pricing are kept up to date via a Cron job. - **Multi-protocol support**: Supports OpenAI Chat & Embeddings, Anthropic Messages, Google Gemini, AWS Event Stream, and more. - **Multimodal support**: Accepts image, audio, video, and PDF inputs via chat completions. - **Reasoning effort normalization**: Unifies reasoning effort across providers for consistent results. - **Circuit breaker**: Automatically detects and bypasses failed providers. - **API key permissions**: Validates and checks for model restrictions, quota, expiration, and IP allowlist. ### Tech Stack: Keyaos is built on Cloudflare Workers, TypeScript, and React. It leverages D1 for performance and Cron Triggers for scheduling. ### Why It Matters: Keyaos simplifies the management of multiple AI services by unifying their endpoints, ensuring the cheapest and most reliable provider is always used. This results in cost savings and improved performance, making it an essential

### Overview: This repository is a curated collection of Ideogram 4.0 prompts and image examples, designed for developers and designers. The repository includes typography, photorealistic portraits, product/UI mockups, and model comparisons, providing high-quality examples to help you enhance your creative workflows. ### Key features: - Typography poster design - Photorealistic portraits - Product/UI mockups - Model comparisons ### Tech stack: The repository is built using Markdown and GitHub Pages. ### Why it matters: Ideogram 4.0 is the best open-weight text-to-image model in the world, ranking #1 on the DesignArena Text-to-Image leaderboard. This repository offers a wide range of prompts and images to help you improve your design skills and creativity. Repository: EvoLinkAI/awesome-ideogram-4.0-prompts