📦 zihan-su/self_gradient_forcing_plus

Self Gradient Forcing Plus: 24-Hour Videos From 5-Second Clips

A machine learning model trained on just five seconds of video can now generate twenty-four hours of continuous footage. That is the core capability of Self Gradient Forcing Plus. The project solves a specific problem where the math used to learn context and clean up noise fought each other, causing errors to pile up over time. By separating these conflicting updates, the model keeps its footing, allowing it to roll out for days without falling apart. This matters because it proves that short training data can support massive, long-form generation. The takeaway is simple: smarter math lets small models dream for hours.

🆔 @hackernewsgithubprojects

📦 shadcn-labs/mcpcn

mcpcn: The UI Toolkit for AI Apps

Stop wrestling with generic form libraries when you need to build interfaces for AI agents. mcpcn gives you ready-made, customizable React components specifically designed for ChatGPT and Claude applications. It works like that famous copy-paste component library, but it is built for the unique needs of conversational interfaces. Think message bubbles, payment flows, and status badges that just work out of the box. No complicated setup, no config files, just clean code. If you are building tools for machine learning models, this saves you from reinventing the wheel every time you need a chat UI. Grab it, paste it, and start building.

🆔 @hackernewsgithubprojects

📦 atlascloudai/awesome-minimax-h3-prompts

20 languages of video magic in Awesome MiniMax H3 Prompts

One specific prompt in this collection can instantly generate a finished brand film, a complex motion graphic, or an entire game trailer, all in twenty different languages. This is not a coding tool or a software library; it is a carefully curated library of text instructions designed to teach the MiniMax H3 video engine exactly what you want to see on screen. It solves the frustrating problem of guessing how to word requests for complex video scenes by providing official, tested examples from six major creative industries. You can copy these precise phrasings to get professional, polished results without spending hours experimenting.

🆔 @hackernewsgithubprojects

📦 wentingw/awsm

AWSM: The 3D World Builder

AWSM uses AI agents to turn flat video clips into fully editable three-dimensional worlds that robots can actually use. Instead of just guessing what a room looks like, it measures the space and builds a digital model with real geometry, which is a big deal for testing how machines interact with their surroundings. The system acts like a careful builder, checking its work against camera data before finalizing the scene. This creates a reliable map that simulators can use to practice tasks without needing the physical space. It is a fascinating step toward giving software a true understanding of physical reality.

🆔 @hackernewsgithubprojects

📦 zju3dv/supernav

SuperNav: The AI Agent That Navigates 3D Worlds

SuperNav is the agentic navigation system that lets a general-purpose AI agent walk through complex 3D environments without needing any specialized training. Instead of teaching a robot how to walk from scratch, this system gives a standard multimodal model a set of tools and skills to figure out its way on the fly. It handles everything from finding a specific object to visiting multiple locations, all while keeping track of its progress and context. The coolest part is that you can watch the agent think and move in real-time through a built-in web viewer, seeing exactly how it decides where to step next.

🆔 @hackernewsgithubprojects

📦 atameric/ps5-portal-high-bitrate

PS5 Portal High Bitrate

You can temporarily force your PlayStation Portal to stream at 100 or even 200 megabits per second, bypassing the standard speed limits that usually cap the picture quality. This experimental tool, called PS5 Portal High Bitrate, uses a computer to intercept the initial handshake between your PlayStation five and the handheld console, tweaking a single number to request a faster data pipe before the video session begins. It is not a permanent hack or a jailbreak, but a temporary tweak that restores the normal network path once you are ready to play.

🆔 @hackernewsgithubprojects

📦 tencentcloud/octop-gateway

Octop Gateway: One Bot, Every App

Octop Gateway is the universal translator that finally stops you from rewriting your bot code for every single messaging app. Built by the folks at Tencent Cloud, this tool connects your AI agent to almost every major chat platform at once, including Telegram, Discord, and even WeChat. The coolest part is that you write your logic only once, and the gateway handles all the messy, platform-specific formatting automatically. It turns a dozen different chat systems into one clean, unified stream, so you can focus on making your bot smart rather than fixing its syntax.

🆔 @hackernewsgithubprojects

📦 guwidoe/pi-prompt-suggester

How pi-prompt-suggester writes your next command for you

pi-prompt-suggester is the coding agent extension that finally writes your next command before you do. Instead of just guessing based on your last message, it builds a deep understanding of your entire project first. It reads through your codebase to figure out your goals, constraints, and architectural rules, creating a persistent mental map of what you are actually trying to achieve. Then, after every assistant response, it uses that context to predict exactly what you want to ask next. The suggestion appears as faint ghost text in your editor, letting you just press space to accept it. This keeps your workflow moving without breaking your focus.

🆔 @hackernewsgithubprojects

📦 samuelit0davila/fifa-street-pc-port

FIFA Street 2012 on Windows

FIFA Street 2012 Recompiled is the experimental port that brings the beloved Xbox 360 title to Windows PC. This project takes the original game code and rebuilds it to run natively on modern computers, offering a graphical installer and a launcher with high-performance graphics settings. You do not need to install complex development tools; instead, you bring your own game disc to run the software. It includes a custom launcher that supports advanced graphics options and performance profiles for different hardware setups. This allows players to enjoy the fast-paced street soccer experience on their desktop without needing an Xbox console.

🆔 @hackernewsgithubprojects

📦 enntity/sparkglm

SparkGLM: The Two-Spark AI Workstation

SparkGLM is the local AI setup that runs a massive language model on just two small computer boxes. The real surprise is that it connects these two Sparks with a single cable to act as one giant machine, allowing them to process huge conversations without crashing. Instead of forcing you to restart or wait for slow responses, it remembers past chats on the hard drive so you can pick up right where you left off in under a second. It turns two modest devices into a reliable, fast AI server that actually works for long, complex tasks.

🆔 @hackernewsgithubprojects

📦 reardencode/rbitcoin

rbitcoin: The Bitcoin Node That Skips the UTXO Set

rbitcoin is a full Bitcoin node written in Rust that stops maintaining a live list of every available coin and instead stores the entire chain history in a relational archive. By dropping the traditional UTXO set, it avoids the massive memory spikes typical of standard nodes, allowing a full sync to complete in under thirty hours on a regular laptop. It also runs wallet backend services directly inside the process, meaning you get a complete, self-contained server without stitching together separate indexing tools. It is a fascinating look at how much overhead can be stripped away from the standard model.

🆔 @hackernewsgithubprojects

📦 humanive/cogfree

CogFree: Thinking Tools for AI Agents

Free your brain from the headache of juggling every detail in your head. CogFree is a fresh collection of twenty-eight tiny commands you can drop into AI tools like Claude or Codex to help you think better. Instead of the AI just answering questions, these skills let you pause, compress your thoughts, check your assumptions, or save your progress so you can come back to it later. It turns the chat into a real thinking partner that helps you untangle messy ideas without losing your place. It is a clever way to keep your focus sharp and your logic clean, right where you are working.

🆔 @hackernewsgithubprojects

📦 byhowiecreations/fileapex

Move files without the cloud

Send files from your phone to your laptop without uploading them to a cloud server first. FileApex creates a direct, private link between your own devices, letting you browse folders and move files over your home Wi-Fi instantly. It also lets you read text messages or copy links on one device and see them on another. Since you control the connection and the data stays on your hardware, there is no third party holding your files. It is a simple, local way to keep your digital life organized and connected. Pair your devices once and forget about the cloud.

🆔 @hackernewsgithubprojects

📦 ubc-vision/solv3r

Solv3r: One Solver for 3D

Solv3r unifies how computers understand three-dimensional space by acting as a single, flexible solver for multi-view reconstruction. Instead of forcing you to pick between offline batch processing and real-time streaming, this new tool handles both seamlessly. It also solves a common headache: it works even when your camera views arrive out of order, making 3D modeling far more forgiving. Built by researchers at UBC and partners, this project promises to simplify complex geometric tasks. Code and models are dropping soon, so keep an eye out for the release. It is a clever step toward making 3D vision easier to build and use.

🆔 @hackernewsgithubprojects

📦 thumai-lab/llava-uhd-v4

LLaVA UHD v4: Smarter Vision AI

LLaVA UHD v4 is the vision model that finally stops wasting energy on blurry details. It changes how computers see by squeezing information down inside the image processor itself, not after the fact. This means the AI can read tiny text and understand complex diagrams without needing a massive amount of computing power. It cuts the required math by more than half while keeping the same sharp accuracy. It is a clever way to make high-resolution vision models fast and cheap. If you want to see how AI can be both smart and efficient, this is the one to watch.

🆔 @hackernewsgithubprojects

📦 sarpstar/cinecode

Films Made Entirely With Code

A single AI agent can write the script, build every visual frame from code, design the audio, and export a finished movie file for your computer. This project, called cinecode, turns a simple text prompt into a complete video, handling everything from animation to sound mixing. It runs locally on your machine using standard tools, so you keep total creative control without relying on cloud servers. You choose the visual style, from realistic 3D to hand-drawn ink, and the agent handles the technical heavy lifting. The result is a fully mastered video file, ready to share.

🆔 @hackernewsgithubprojects

📦 truongfelix/workflow

Workflow: The AI Workflow That Actually Remembers

Workflow is the Claude Code setup that finally stops your AI from forgetting what it did yesterday. It solves the headache where an agent starts every session with amnesia, blindly trusting old notes about work that is already done or broken. The magic is its ground truth script, which checks the actual git history and running jobs to verify reality before the AI acts. Instead of repeating lies, it compares its notes against the facts and flags the contradictions. You get a smart assistant that knows exactly where your project stands, keeping long-term builds consistent across weeks and machines.

🆔 @hackernewsgithubprojects

📦 cloudopsbench/cloudopsbench

CloudOpsBench Tests AI Cloud Skills

CloudOpsBench gives AI agents forty five real world cloud tasks to complete, like fixing broken services or cleaning up unused resources, to test if they can actually operate infrastructure safely. Each task comes with a specific starting state and a strict checker that verifies the final result, ensuring the AI didn't just guess but truly solved the problem. This is useful for anyone building automated cloud tools who needs to know if their AI assistant is ready for real work. It turns vague promises of AI capability into concrete, measurable results.

🆔 @hackernewsgithubprojects

📦 0xs4n4tg/papercut-explainer

Turn Any Topic Into a Paper-Cut Video

Turn any topic into a snappy, vertical video that looks like a 3D paper diorama, without touching a single line of code. The Papercut Explainer project uses Claude Code to handle the entire workflow, writing the script, planning the camera moves, and generating the voiceover for you. It then builds the scenes using a library of paper textures and sound effects, so you just describe the video you want and let the AI do the heavy lifting. It’s like having a mini production team that speaks code. If you want to make stylish explainer reels without hiring an editor, this is your new shortcut.

🆔 @hackernewsgithubprojects

📦 leag1234/elixiroffice

ElixirOffice: The Open Source Word That Actually Feels Like Word

Edit documents with the exact same ribbon, menus, and keyboard shortcuts you know from Microsoft Office, but on free software. ElixirOffice is an open source office suite built for people who are used to Word, Excel, and PowerPoint. It solves the problem of switching away from Microsoft Office by making the new tool feel familiar from day one, so your muscle memory still works. The most surprising part is its speed: it uses a special WebAssembly converter that opens large files ten times faster than a standard build.

🆔 @hackernewsgithubprojects

📦 jesdga95/basalt

Basalt: Run Massive AI Models Locally

Basalt is the local engine that lets massive language models run on a single computer without cloud dependence. It solves the problem of models that are too big to fit in your graphics card memory by cleverly parking the parts that do not fit in your regular system memory, allowing a workstation to handle complexity that usually requires a server farm. It also supports image analysis, turning your hardware into a fully private, capable assistant. The result is high performance without the overhead of external data centers. If you want to own your AI infrastructure and run powerful tools locally, this is the project to watch.

🆔 @hackernewsgithubprojects