Projects

### Overview: Seedance 2.5 Prompts is a comprehensive resource for developers looking to enhance their image generation tasks. It provides tested recipes, example renders, aspect ratios, and the cost per image, making it a valuable tool for optimizing image prompts. ### Key features: - **Model ID:** `seedance-2.5` - **Aspects:** 16:9, 16:10, 16:17, 16:20, 16:25 - **Cost per Image:** $0.0961 per second (480P), $0.216 per second (720P) ### Tech stack: Python ### Why it matters: Seedance 2.5 Prompts offers optimized image generation recipes, ensuring high-quality results at a competitive price. It simplifies the process of creating professional-looking images by providing clear instructions and examples, making it an essential tool for developers looking to enhance their projects.

### Overview: This repository is a curated collection of AI image generator prompts, text-to-image workflows, and AI image editing guides, designed to help creators, marketers, designers, e-commerce sellers, and agencies enhance their image creation and editing skills using AI. ### Key Features: - AI image generator prompts - Text-to-image workflows - AI photo editor ideas - AI art prompts - AI logo generator workflows - AI profile picture prompts - AI product image prompts - Image editing checklists ### Tech Stack: HTML, Markdown, and GitHub Pages. ### Why it Matters: This repository provides a practical and SEO-friendly guide for creators looking to improve their image creation and editing skills using AI, including text-to-image prompts, realistic image generation, and various AI art and logo generation workflows.

Overview: Mia is a lightweight, cross-platform inference framework built in Rust, designed to efficiently execute machine learning models across various hardware platforms. Key features: - Type-safe operations enforced at compile-time - Shape-aware runtime validation for operation compatibility - Zero-cost abstraction for building computation graphs - Lazy evaluation with a tensor tape for deferred execution - Ownership-driven memory reuse to eliminate manual management - Decoupled backend architecture for platform-specific implementations Tech stack: Rust Why it matters: Mia addresses model deployment challenges by providing a high-performance, type-safe, and cross-platform solution for efficient inference, leveraging Rust's strengths in type safety and cross-platform compatibility. Repository: cryscan/mia GitHub stars: 10 GitHub description: A Lightweight, Cross-Platform Inference Framework in Rust (WIP) README excerpt: ...

### Overview: This repository contains eight tested Wan 2.7 Image prompts, each with an example render, ratio, and cost on the APIMart API. The prompts are structured to ensure the material is named rather than the adjective, set the light explicitly, and pin the framing. ### Key Features: - **Model ID:** `wan2.7-image` - **Prompts:** Eight prompts in JSON format and a table in Markdown. - **Cost:** $0.0216 per image, estimated cost for the set: $0.1728. ### Tech Stack: Python ### Why It Matters: The prompts are designed to ensure the material is named rather than the adjective, set the light explicitly, and pin the framing to achieve the best possible results. This ensures that the output is tailored to the desired aesthetic and context, leading to more effective and consistent image prompts.

### Overview: This repository contains tested image prompt recipes for the Grok Imagine Image model, including examples of renders, aspect ratios, and the cost per image. ### Key features: - **Model ID**: `grok-imagine-image` - **Aspect Ratios**: 1:1, 3:4, 4:3, 8:5, 16:9, etc. - **Cost**: $0.016 per image ### Tech stack: Python ### Why it matters: This repository provides a concise guide to crafting effective image prompts for the Grok Imagine Image model, ensuring high-quality renders at a reasonable cost.

### Overview: Nano Banana 2 Prompts: 8 Tested Recipes With Outputs ### Key Features: - **Model:** `gemini-3.1-flash-image-preview` - **Recipes:** 8 (machine-readable in [`data/prompts.json`](data/prompts.json), table in [`PROMPTS.md`](PROMPTS.md)) - **Effective Price:** $0.015 per image — estimated cost of the whole set: $0.12 ### Tech Stack: Python ### Why It Matters: - **Key Features:** Nano Banana 2 Prompts offers 8 tested recipes with outputs, including the model ID (`gemini-3.1-flash-image-preview`), aspect ratios, and the cost per image. It provides a concise and effective way to create professional-looking images without requiring a subscription or monthly plan. - **Model ID:** `gemini-3.1-flash-image-preview` is compatible with OpenAI and offers a low-cost endpoint at $0.015 per image, making it ideal for developers and creatives looking to experiment with image generation without financial constraints. - **Prompt Structure:** The prompt structure is designed to be clear and effective, focusing on the subject, light, framing, setting, and finish. This ensures that each recipe is both descriptive and actionable, helping developers create high-quality images quickly and efficiently. - **Prompt Cheat Sheet:** A cheat sheet is provided to help developers quickly understand the structure of the prompts, making it easier to create effective and consistent recipes. ### Repository: Repository: d63hy9

Overview: FLUX 2 Max API, or `flux-2-max`, is a model id, pricing, and parameters for the OpenAI-compatible APIMart endpoint. It provides a base URL and a single key for access, offering a cost-effective solution for image generation tasks. Key features: - **One Key, Multiple Models**: A single API key grants access to FLUX 2 Max (`flux-2-max`) and 300+ additional models for images, videos, and language, without the need to change the `model` field. - **Minimal Cost**: No need for monthly fees; pay as you go with a minimum of $1 per task. - **Summaries Returned**: Each completed task returns the sum of the cost (`cost` or `credits_cost`), and failed tasks are returned to the balance. - **Asynchronous by Default**: Sends tasks and retrieves `task_id` for polling, with a packageable processing and retry feature. Tech stack: Python Why it matters: This API is particularly useful for teams that need a single key to access multiple models for image generation, reducing the need for complex integration and avoiding the hassle of managing multiple APIs or subscriptions. Repository: d79bcrw6n9kd/flux-2-max-api-ru GitHub stars: 10 GitHub description: FLUX 2 Max API, or `flux-2-max`, is a model id, pricing, and parameters for the OpenAI-compatible APIMart endpoint. It provides a base URL and a single key for access, offering

Overview: This starter application allows developers, artists, or anyone interacting with the Dall-E 2 / Dall-E 3 API from OpenAI to quickly and easily generate images without writing code. Key features: - Access Dall-E 3 generation using API with Generation functionality (Variation and Edit currently unsupported by OpenAI for Dall-E-3). - Access Dall-E 2 generation using API with Generation, Variation, or Edit endpoints. - Generate images without corner logo. - Generate images using API pricing. - Mask images for Edit endpoints in-browser using canvas editor. - Save image generations locally, automatically. - Adjust generation quantity and resolution (style and quality options available for Dall-E-3). Tech stack: Node / Express, Jade, TailwindCSS / Flowbite, Fabric.JS (canvas element). Why it matters: This application provides a quick and easy way to interact with the Dall-E 2 / Dall-E 3 API through a web browser, without the need to write code, making it suitable for developers, artists, or anyone regularly working with the API.

### Overview: This repository contains annotated Jupyter Notebooks that dive into various advanced AI engineering techniques, including Self-Attention, In-Context Learning, LLM-Augmentation, Retrieval Augmented Generation, Knowledge-Graphs, Fine-Tuning, and Model Optimization. The notebooks are designed to provide best practices and practical insights for building AI models using Large Language Models (LLMs). ### Key Features: - Self-Attention Mechanism - In-Context Learning - LLM-Augmentation with Tool Integration - Retrieval Augmented Generation - Knowledge Graphs - Fine-Tuning BERT - Model Optimization: Quantization ### Tech Stack: Jupyter Notebook ### Why it Matters: These notebooks offer practical, hands-on examples that help developers and engineers understand and implement advanced AI techniques effectively. They are particularly useful for optimizing model performance, improving training efficiency, and enhancing the ability of LLMs to perform tasks based on contextual information. ### Repository: Repository: dcarpintero/ai-engineering GitHub Stars: 10 GitHub Description: AI Engineering: Annotated NBs to dive into Self-Attention, In-Context Learning, RAG, Knowledge-Graphs, Fine-Tuning, Model Optimization, and many more.

Overview: Seedance 1.5 Pro API (`seedance-1.5-pro`) provides access to the Seedance 1.5 Pro model via the APIMart endpoint, offering model id, pricing, and parameters for OpenAI-compatible API usage. Key features: - Model access via `seedance-1.5-pro` model id - Pricing per generated second, with tiered pricing based on resolution - Model parameters including prompt, duration, resolution, size, and image URLs - Quickstart example using `curl` and `python` scripts Tech stack: Python Why it matters: Seedance 1.5 Pro API provides a convenient and cost-effective way to integrate the Seedance 1.5 Pro model into projects, offering precise control and smooth motion for video generation.

Overview: Explore what Hacker News is really thinking by extracting deep concepts, tracking entities, and analyzing discourse patterns through LLM-powered semantic analysis and vector embeddings. Key features: - 🧠 Concept Explorer: See trending ideas and their sentiment. - 🏢 Entity Tracker: Track companies, products, and open-source projects. - 📊 Sentiment Analysis: Monitor community emotional temperature and controversy levels. - 💬 Discourse Patterns: Identify types of arguments people make. - 🔍 Semantic Search: Search for concepts, not keywords. - ⚡ Background Ingestion: Automatic polling and processing. - 🗄️ Smart Caching: Skip already-seen stories and comments. - 🔧 Admin Dashboard: Monitor worker progress and re-analysis triggers. Tech stack: TypeScript, Docker Compose, OpenRouter, ChromaDB, Express, Sequelize, PostgreSQL. Why it matters: Go beyond surface-level HN browsing and uncover the underlying ideas, community sentiment, and discourse patterns through semantic analysis and vector embeddings. Repository: devrupt-io/ethos GitHub stars: 10 GitHub description: Exploring what HN is really thinking.

Overview: Seedance 1.5 Pro API is an OpenAI-compatible endpoint offering model id, pricing, parameters, and copy-paste examples for generating videos and images. It supports 300+ models and is billed by resolution tier at $0.0104 per second. Key features: - Model ID and Pricing - Parameters for prompt, duration, resolution, and size - Asynchronous processing - Billing in the response Tech stack: Python Why it matters: Seedance 1.5 Pro API simplifies access to a wide range of AI models, providing precise control and smooth motion without the need for monthly subscriptions or pre-dimensioned plans.

### Overview: Grok Imagine Image Prompts is a collection of 8 tested image prompt recipes designed for developers and creatives. These recipes are optimized for OpenAI-compatible endpoints, offering a cost-effective solution with a price of $0.016 per image. The prompts are designed to guide the model towards the desired outcome, ensuring high-quality renders without unnecessary complexity. ### Key Features: - **Model:** grok-imagine-image - **Recipes:** 8 (machine-readable in `data/prompts.json`, table in `PROMPTS.md`) - **Effective Price:** $0.016 per image (estimated cost of the set: $0.128) ### Tech Stack: - Python ### Why It Matters: - **Cost-Effective:** The cost per image is $0.016, making it a highly economical solution for developers. - **OpenAI Compatibility:** Works seamlessly with OpenAI-compatible endpoints, ensuring compatibility with existing workflows. - **Minimal Setup:** No need for a subscription; simply use the provided API key to start using the model. - **Detailed Prompt Structure:** The prompt structure is designed to be easy to follow and understand, guiding the model towards the desired outcome. ### Gallery: Each row in the gallery represents a single recipe, including an example render, category, ratio, and the exact prompt used. - **Product / e-commerce:** A frosted glass perfume bottle on a wet black stone slab. - **Food photography:** A spicy miso ramen bowl with soft boiled egg, nori, and scall

### Overview: Tennis-Vision-AI-Tracking is an end-to-end tennis video analysis system that leverages Ultralytics YOLOv8 for player and ball detection, multi-object tracking across frames, and a custom CNN for court keypoint extraction. The system provides physics-aware metrics and tactical visualizations, making it suitable for both academic research and professional deployment in sports environments. ### Key Features: - **🎯 Dual-Model Detection**: Separate YOLOv8 (players) and fine-tuned YOLOv5 (ball) models for precise detection. - **🧭 Identity-Aware Tracking**: DeepSORT-inspired tracking with `persist=True` for stable track IDs. - **📐 Court Coordinate Regression**: ResNet50 backbone for precise court keypoint extraction. - **⚡ Trajectory Reconstruction**: Rolling-window smoothing + intelligent interpolation for ball path. - **🧮 Physics-Aware Metrics**: Shot speed, player distance, and court coverage from calibrated coordinates. - **🗺️ Tactical Visualization**: Live mini-court overlay, player heatmaps, and shot spray charts. - **📦 Multi-Format Export**: Annotated MP4, frame-by-frame CSV, structured JSON telemetry, static plots. - **🚀 High-Velocity Stub System**: Serialize/deserialize detection outputs for rapid iteration. ### Tech Stack: Jupyter Notebook ### Why it Matters: This system addresses the complex challenges of computer vision in sports analytics, providing actionable insights and strategic visualizations for both academic research and professional deployment in sports environments. ### GitHub Stars: 10

### Overview: 本项目旨在开发一款针对大学生迷茫问题的智能学习平台，通过结合LLM与回归分析算法，为大学生提供个性化学习、交流、资源共享等一站式服务，帮助他们实现自我提升和规划未来。 ### Key features: - **我的未来**：基于回归分析的薪资预测与技术差距分析功能。 - **小智同学**：基于ChatGLM-6B大语言模型的智能百科问答与心理疏导机器人。 - **社团论坛**：同学们可以自由创建论坛，发表观点，交流技术，切磋学问，了解大学生活。 - **资源天地**：同学们可以共享学习资源干货。 - **云端自习**：上自习，也可以适当放松听音乐。 - **目标清单**：个人主页，提供一些建议。 ### Tech stack: Java ### Why it matters: 本项目通过解决大学生迷茫问题，为他们提供个性化学习和规划未来的机会，有助于减轻学习负担，提高学习体验，从而促进他们的全面发展。

Overview: This project aims to provide an efficient and scalable API for embedding and storing vectors, leveraging FastAPI, Langchain, and a vector database such as Chroma. Key features: - File and Text Embedding - Vector Storage - Similarity Search - Scalable and Fast Tech stack: Python Why it matters: This project offers a comprehensive solution for embedding and storing vectors, making it ideal for applications requiring high performance and scalability. Repository: ehsanghaffar/vector-store-api GitHub stars: 10 GitHub description: This project aims to provide an efficient and scalable API for embedding and storing vectors, leveraging the power of FastAPI, Langchain and a vector database such as Chroma.

### Overview: snap is a decision engine designed for LLM pipelines, focusing on extracting typed answers from a single forward pass over the model. It eliminates the need for parsing, retry loops, and token management, making it ideal for scenarios where generating text is not desirable. ### Key Features: - **Single-pass Neural Answer Probabilities:** The input is the data and every question about it, while the output is a set of typed answers with their probabilities. - **No Generated Text:** Snap reads the answer directly from the model, eliminating the need for parsing, retry loops, and token management. - **Local, Deterministic, Jev-Compatible:** The engine is designed to be local, deterministic, and compatible with the Jev wire protocol. ### Tech Stack: Rust ### Why It Matters: Snap simplifies the process of extracting typed answers from LLM pipelines, reducing the need for parsing, retry loops, and token management, making it a practical solution for various use cases such as routing and triage of support tickets, checking tool calls, and more. ### Repository: emnlmn/snap ### GitHub Stars: 9

### Overview: This repository provides a fine-tuning pipeline for BERT, focusing on extractive summarization tasks. It aims to address the lack of datasets for extractive summarization and offers a tool for model interpretability using gradient-based methods and an attention-based method. ### Key Features: - Fine-tuning pipeline for BERT - Conversion of abstractive summarization datasets to extractive format - Simplified HuggingFace pipeline for fine-tuning - Data preparation and training using HuggingFace's `trainer.py` and `metrics.py` files - Model evaluation using ROUGE metric ### Tech Stack: Python ### Why It Matters: This project is useful for researchers and developers looking to improve extractive summarization models and datasets, offering a comprehensive solution for extractive summarization tasks.

### Overview: `langmem-ts` is a TypeScript-first class memory primitive designed for AI agents. It provides a seamless way to store, retrieve, and embed data using Postgres and pgvector, backed by a Postgres database. The library is MIT-licensed and zero framework coupled, making it ideal for developers looking to integrate AI capabilities into their projects without the complexities of traditional frameworks. ### Key Features: - **Embedding**: Converts text into vectors for efficient storage and retrieval. - **Storing Memories**: Writes memories to a Postgres database. - **Retrieving Memories**: Searches for relevant memories based on vector similarity. - **Swappable Embedders and Stores**: Allows consumers to swap out the default implementations with their own configurations. ### Tech Stack: - **TypeScript**: The primary language used for development. - **Postgres**: The database backend for storing and retrieving memories. - **pgvector**: The extension for Postgres that supports vector storage. ### Why It Matters: `langmem-ts` fills a gap in the market by providing a MIT-licensed, TypeScript-native memory library for AI agents. It simplifies the integration of AI capabilities into projects by abstracting away the complexities of traditional frameworks, making it accessible to developers of all skill levels. ### Repository: Repository: ereztdev/langmem-ts GitHub Stars: 10 GitHub Description: TypeScript first class memory primitive for AI agents. Embed → store → retrieve → inject, backed by Postgres + pgvector. Swappable embedders, consumer owns the connection, zero framework coupling. MIT. ### Install:

Overview: This project is a minimal Node.js demo showcasing how to use ChromaDB as a local vector database. It stores a set of sample documents and performs a semantic search query using natural language, perfect for understanding the basics of vector search and how embeddings can be used to find meaning-based matches in text. Key features: - Stores sample documents - Performs semantic search query - Uses ChromaDB as a vector database Tech stack: Node.js Why it matters: This project serves as an excellent starting point for developers to grasp the fundamentals of vector search and how embeddings can be utilized to find semantic-based matches in text, making it a valuable resource for anyone interested in this technology.