📦 danmoreng/qwen-tts-studio

Qwen-TTS Studio

The local text-to-speech app that finally runs high-quality voice generation entirely on your own machine without a complex Python setup. Qwen-TTS Studio is a slick desktop interface for Windows and Linux that gives you a fast C++ speech engine powered by your CPU or Nvidia card. Instead of sending your voice data to some corporate cloud, you can generate natural speech, play with tone instructions like whispering, and even clone custom voices from a tiny audio clip, all completely offline. It is the ultimate playground for running advanced speech models locally without the headache. Give it a spin and take back your voice privacy.

🆔 @hackernewsgithubprojects

📦 pham-tuan-binh/leslider

LeSlider

You can now expand a standard six-axis robot arm into a seven-degree-of-freedom system using a single extra servo motor and some 3D-printed parts. The leslider project is an open-source hardware and software add-on designed for the SO-101 robot arm. By mounting the arm on a linear V-slot slider, it introduces a seventh travel axis that integrates with the LeRobot framework. It supports both velocity-based steering and a calibrated position mode, allowing you to control linear travel alongside the arm using a keyboard. It lets you teleoperate, record unified datasets, and train AI policies on the entire seven-joint setup.

🆔 @hackernewsgithubprojects

📦 taylorhuston/local-life-manager

Local Life Manager

Have Claude Code run your entire day by transforming your terminal into a personalized command center. This framework links your local notes and code repositories to custom AI agents, using dozens of tailored skills to orchestrate your daily routine. Instead of juggling generic prompts, you can trigger deep-dive research, manage daily journals, structure learning sessions, or even have the AI read and summarize your RSS feeds and YouTube subscriptions directly into your Obsidian vault. It acts as a clever coordinator that remembers your preferences, plans your tasks, and acts as a dedicated study partner. It is a brilliant way to build a private, localized second brain.

🆔 @hackernewsgithubprojects

📦 gabrieletrovato01/koreshelf

KoreShelf

Organize your digital books in a stunning virtual room and turn them into custom brain food for artificial intelligence. By dragging your electronic documents into this local environment, you get a beautiful virtual bookshelf complete with moving bookmarks and a built-in reader. But the coolest trick is the export button, which strips away noisy formatting from your books to create a clean, structured document. You can drop this directly into your favorite artificial intelligence companion to chat with the text, ask about the plot, or get quick summaries. It is the ultimate bridge between your personal library and smart tech.

🆔 @hackernewsgithubprojects

📦 capitan01r/comfyui-krea2t-enhancer

ComfyUI Krea2T Enhancer

Make AI images follow written prompts with much higher accuracy. A new custom extension called comfyui-krea2t-enhancer hooks directly into the text-fusion path of Krea2 diffusion models to solve the common issue of ignored details. Instead of messing with text prompts or adding complicated extra nodes, this extension applies a controlled conditioning adjustment directly inside the model during sampling. You simply drop the node between your model loader and the sampler to blend the enhancement strength to your liking. It is a clean, lightweight way to get precise control over your image generation pipeline without cluttering your workflow.

🆔 @hackernewsgithubprojects

📦 gazingstars123/anima-standalone-trainer

Anima Standalone Trainer

Train personalized AI art styles on your own computer with a dedicated, lightweight environment that focuses strictly on the Anima model. This training tool packages everything you need into a single desktop interface, solving the headache of complex command lines and messy dependencies. It gives you a clean web control panel to manage your training jobs, adjust model paths, and fine-tune your parameters. What makes it really special is how it optimizes multi-GPU setups on Windows and Linux, letting multiple graphics cards share the heavy lifting to speed up your training. Now you can easily create custom styles without fighting complicated software configurations.

🆔 @hackernewsgithubprojects

📦 avadiii/rewardspy

rewardspy

rewardspy is the plug-in debugger that catches reinforcement learning reward hacking before it ruins your model's training run. When you are training a model, the reward curve might look like it is climbing straight to success, but your model might actually just be exploiting a loophole, like over-writing its own tests or inflating response lengths. With one decorator import, this tool observes your reward function without modifying it, tracking statistical anomalies like variance collapse or component dominance in real time. It displays everything on a live terminal dashboard and can even automatically halt training if it detects cheating.

🆔 @hackernewsgithubprojects

📦 spikelab-jhu/isaac-sim-event-camera-plugin

isaac-sim-event-camera-plugin

isaac-sim-event-camera-plugin is the physics simulator extension that generates high-speed event camera data without slowing your computer to a crawl. Standard virtual event cameras require rendering hundreds of frames a second, which completely tanks simulation performance. This clever tool solves that by rendering only a few key frames and using smart motion-vector math to fill in the gaps. It gives you hyper-realistic, thousand-hertz event data while cutting down actual rendering work by up to eight times. It even features a live preview window so you can watch simulated robot eyes react in real time. Download this plugin to instantly add fast, efficient event-camera sensing to your robotics projects.

🆔 @hackernewsgithubprojects

📦 artokun/comfyui-mcp

comfyui-mcp

comfyui-mcp is the Claude Code plugin and Model Context Protocol server that lets you edit and run your live ComfyUI graph directly from an AI chat session. Instead of tangling with spaghetti nodes by hand, you can just tell the AI to build workflows, swap models, or generate art. It packs over a hundred tools and specialized skills for heavy-lifters like Flux and WAN video, meaning the AI actually understands sampler settings and resolutions without breaking your setup. It even embeds an autonomous agent panel right into your sidebar. It is the ultimate way to orchestrate complex AI generation with pure, conversational ease.

🆔 @hackernewsgithubprojects

📦 fredantb/spec-driven-development

Spec-Driven Development

spec-driven-development is the Claude skill that finally stops your AI coding assistants from contradicting each other. If you have ever used Claude Code, Cursor, and Copilot on the same project, you know the pain of one agent completely overwriting another agent's progress because they do not share a single source of truth. This clever tool fixes that by interviewing you first, generating three unified markdown files for your requirements, design, and tasks, and then writing matching configuration files for all your favorite AI editors. Now, every single assistant reads the exact same playbook, stopping architectural drift and wild code hallucinations instantly.

📰 https://news.ycombinator.com/item?id=48221805

🆔 @hackernewsgithubprojects

📦 harshuljain13/llm-inference-at-scale

llm-inference-at-scale

Serving a large language model is completely different from traditional software because every single response has a highly unpredictable cost. If you are trying to understand how to handle this, the project llm-inference-at-scale is a comprehensive, open-source handbook designed specifically to teach you how to serve these models in production. It breaks down the complex mechanics of GPU hardware, memory limits, and optimization strategies into easy-to-digest chapters with practical labs. You will learn how systems manage memory during conversations and how to handle massive traffic. Check out the repository to master the engineering behind running models efficiently.

📰 https://news.ycombinator.com/item?id=48424467

🆔 @hackernewsgithubprojects

📦 jia-lab-research/unityshots

UnityShots

Most video generators struggle to keep characters and settings consistent when switching between different camera angles, but unityshots fixes this by generating an entire multi-shot film with synchronized audio in one go. It uses a smart memory system to remember what characters look like and how the environment looks across up to nine different camera cuts. This means a character keeps the same clothes and face from shot to shot, while the background music and ambient sounds transition smoothly. It turns a single-shot video generator into a complete, multi-shot storyteller that keeps everything in perfect harmony.

🆔 @hackernewsgithubprojects

📦 porunc/codewiki

CodeWiki

You can now transform any messy codebase into an interactive, structured developer wiki with a single command. CodeWiki is a local intelligence tool that scans your repository, builds a complete code graph of definitions, imports, and calls, and uses a local database to generate deep, source-grounded documentation. It completely bypasses manual writing by mapping your systems, workflows, and API layers into a visual graph explorer and a searchable web UI. It even has a lightweight mode that runs entirely without AI, feeding fast symbol searches and call traces straight to your favorite editor or assistant. It is like having a sharp developer who already knows your entire system inside out.

🆔 @hackernewsgithubprojects

📦 ch-bas/threejs-sims-house-builder

threejs-sims-house-builder

Design and build your dream multi-floor home directly in your web browser. This project brings a complete grid-based architecture and interior design system to life using three-dimensional rendering. You can construct up to four floors, paint walls with custom patterns, lay down wood or tile flooring, and arrange dozens of different furniture pieces. There is even a first-person walkthrough mode to explore your creations, wandering computer-controlled characters, and a dynamic day-and-night cycle that casts realistic shadows. It is the perfect playground for anyone who loves design, offering a fully interactive, lightweight home builder without needing to install a thing.

🆔 @hackernewsgithubprojects

📦 saaspegasus/django-boilerplate

Django Boilerplate

You can now boot up a production-ready web application in seconds without writing a single line of setup code. Under the hood, django-boilerplate bundles modern Django with a fast styling engine, clean database integration, and ready-to-go user accounts. The coolest part is that it is completely optimized for coding assistants, shipping with built-in guidelines and pre-configured skills so AI agents can write and refactor your code right out of the box. It solves the massive headache of spending days configuring databases, tasks, and styling pipelines just to start a project. Grab this template, run one quick command, and start building.

📰 https://news.ycombinator.com/item?id=48642564

🆔 @hackernewsgithubprojects

📦 bobvibes/awesome-vibecoding-guide

Awesome Vibecoding Guide

Build and monetize real-world projects by letting AI do the heavy lifting while you direct the flow. The awesome-vibecoding-guide repository shows you how to turn AI-assisted coding into actual income by building local business websites, custom workflows, and micro-SaaS tools. Instead of chasing expensive tools, this handbook teaches you how to combine budget-friendly models and smart prompting techniques to deliver professional results on a dime. You will learn the exact step-by-step playbooks for finding clients, pricing your work, and deploying with zero hosting costs. Grab your favorite AI assistant, check out the guide, and start building software that pays.

🆔 @hackernewsgithubprojects

📦 10x-app-builder/10x

10x App Builder

You can now build fully functional iOS apps on your Mac just by describing what you want. A new open-source project called 10x runs a local AI agent on macOS that writes clean SwiftUI code, automatically scaffolds an Xcode project, and instantly launches it in the iOS Simulator. Instead of just generating code snippets, it operates as a client-side tool loop that edits files, fixes build errors, and captures live screenshots back into a chat interface. It is like having a junior developer build and test your app on a real virtual iPhone while you watch, giving you a fast, native way to turn ideas into working prototypes.

🆔 @hackernewsgithubprojects

📦 nvidia-bionemo/bionemo-agent-toolkit

bionemo-agent-toolkit

The AI assistant plugin that turns your everyday coding agent into a brilliant life sciences researcher. If you have ever wanted to design custom proteins or predict how molecules bind together but did not know where to start, this project is your shortcut. It packages a decade of complex biology models into simple, ready-to-use skills that you can install with a single command. Your AI agent can suddenly run molecular docking, predict protein folding, and even design new backbones using simple prompts. Grab this toolkit to give your favorite AI agent some serious lab credentials.

🆔 @hackernewsgithubprojects

📦 matticusnicholas/kupkaprod-cinema-pipeline

kupkaprod-cinema-pipeline

You can now run an entire autonomous movie studio directly on your home computer with zero cloud fees or subscription accounts. A local project called kupkaprod-cinema-pipeline orchestrates multiple offline artificial intelligence models to transform a single written prompt into a finished film. It starts by running a local chatbot to draft your script, plan detailed camera angles, and outline consistent characters. From there, it generates storyboard keyframes for your approval before directing a local video generator to render multiple synchronized audio and video takes. Once the rendering is complete, a simple interface lets you review the different takes, choose your favorite shots, and automatically stitch them into a final movie.

🆔 @hackernewsgithubprojects

📦 yibie/awesome-autoresearch

awesome-autoresearch

Ditch the manual guesswork and discover how self-improving AI loops actually run in the wild. The awesome-autoresearch repository is a highly curated list of real-world projects that use feedback-driven iteration to improve themselves. Instead of tracking generic research bots, this project spotlights systems that actively write code, test their own performance, and discard their failures to keep only the absolute best results. It is the perfect place to see how developers are building self-correcting programs for trading strategies, AI safety testing, and automated software optimization. It is an incredibly inspiring showcase of programs that learn from their own mistakes.

🆔 @hackernewsgithubprojects