📦 nutlope/bygga

Bygga: 3D IKEA Assembly Guides

Stop staring at confusing IKEA paper manuals. Bygga transforms flat assembly instructions into interactive, three-dimensional guides that look exactly like the official drawings. Type in your product name, and the app animates every single step, showing you precisely where each piece fits. But here is the real kicker. You can just scan any paper manual with your phone. The app uses artificial intelligence to read that physical instruction sheet and instantly converts it into those smooth, visual walkthroughs. It turns a stressful furniture build into a simple, visual puzzle you can actually follow. No more guessing which way the screw goes.

🆔 @hackernewsgithubprojects

📦 lovisdotio/video-to-game

Video to Game: AI Creates Playable 3D Games

Video to Game turns a ten second gameplay clip into a fully playable browser game. The system uses an AI to analyze the video frame by frame, generating all the necessary art, models, and sounds to match the original footage. The agent then writes the code to build the game and even plays it to verify it works. This solves the problem of agents creating games with poor visuals by letting a video model design the look first. It is a neat way to see how AI can handle the entire pipeline from concept to a real, playable experience.

🆔 @hackernewsgithubprojects

📦 tmcleroy/cps2recomp

CPS-2 Recomp: Native Arcade Ports

CPS-2 Recomp is the arcade collection that turns retro fighting games into native PC programs. Instead of using an emulator to mimic old hardware, this project rewrites the actual game code in C++ so it runs directly on your computer. It includes titles like Street Fighter Alpha 2 and Vampire Savior, all inside one handy launcher for Mac, Linux, Windows, and the Steam Deck. The coolest part is the full training mode, giving you hit boxes and frame data so you can practice your combos without guessing.

🆔 @hackernewsgithubprojects

📦 emmilco/us-tax-model

us-tax-model: Your Taxes, Verifiable by Machine

us-tax-model is the open-source project that turns federal tax forms into machine-checkable data so anyone can verify the math. Instead of trusting black-box software, it encodes every IRS instruction as simple, cited rules that computers can audit line by line. The coolest part is a guided interview that walks you through your return, calculates your refund, and shows you exactly where every number comes from, all without sending your private data to the cloud. It’s a rare piece of software that prioritizes your right to understand your taxes. If you want to see how your money is actually calculated, this is the tool to watch.

📰 https://news.ycombinator.com/item?id=50278678

🆔 @hackernewsgithubprojects

📦 benderandrew069/deepseek-api-pricing-2026

DeepSeek API Pricing 2026

DeepSeek API Pricing 2026 is the cost calculator that finally breaks down your bill by cache hits and peak hours. It solves the guessing game of API budgets by showing exactly how much you save when your prompts hit the cache versus when they miss, and how much more you pay during specific daytime windows. The surprising part is that output tokens often cost more than all input combined, so controlling length matters more than just hoping for a cache hit. This guide gives you a simple formula and real numbers to build a low, base, and high budget case before you commit.

🆔 @hackernewsgithubprojects

📦 muse-ai-lab/remcts

ReMCTS: The AI That Learns From Its Own Mistakes

A computer program that can actually look at its own code, spot where it broke, and fix it without you telling it exactly what to change. That is the core trick behind ReMCTS. It stops treating code generation as a single, lucky guess. Instead, it builds a tree of possible solutions, runs every branch, and turns the failures into specific lessons. These lessons get stored and shared so the next branch does not make the same mistake. It is like having a pair of eyes that never gets tired of checking your work.

🆔 @hackernewsgithubprojects

📦 muse-ai-lab/concept-brittleness

Concept Brittleness: Fixing AI's Blind Spots

Text-to-image models often ignore style prompts for specific objects, turning grapes into sketches while leaving cakes in full color, regardless of the seed. Concept Brittleness exposes these systematic blind spots rather than random errors. The project uses sparse autoencoders at every denoising step to separate abstract style concepts from object details. By comparing successful and failed generations, it locates missing dimensions and repairs them at inference time by nudging features toward a class-level prototype. This approach fixes failures without retraining the model. The takeaway is that AI style consistency can be diagnosed and corrected on the fly by understanding how concepts fail internally.

🆔 @hackernewsgithubprojects

📦 muse-ai-lab/statetrack

Why AI Forgets Your Rules

A chatbot can repeat your rules back to you and still ignore them in the next sentence. Researchers at Shenzhen University identified this specific glitch, which they call state integrity collapse. It is not simple forgetting. The model clearly remembers the constraint, but it stops applying it to the task at hand. They built a benchmark called StateTrack to prove this happens in nearly ninety percent of errors. By testing long conversations, they showed that giving the model more memory does not fix the problem. This finding matters because it reveals a deep flaw in how language models maintain instructions.

🆔 @hackernewsgithubprojects

📦 10291124jcjc-dotcom/ai-anime-series

AI Anime Series: One Sentence, One Episode

AI Anime Series is the creative workflow that turns a single sentence into a full thirty second animated episode. It is a complete open source system where you define a few original characters once and then use an AI assistant to write the script, build the visual storyboard, and generate the final video in one go. The most surprising part is how little effort it takes to start a recurring show. You simply describe the theme for the day and the system handles the complex prompts and timing to produce a coherent short film.

🆔 @hackernewsgithubprojects