Downloads IKEA furniture as 3D models
You're about to drop $400 on an IKEA sofa and you still can't tell if it fits the wall.
IKEA already built a photorealistic 3D model of that exact sofa. It's loaded on the product page right now. They just never gave you a download button.
This adds one.
It's called IKEA 3D Model Downloader.
It's a tiny Tampermonkey script that puts a "Download 3D" button right next to IKEA's own "View in 3D" button. Click it and you get the real .GLB file IKEA uses to render the model, saved straight to your drive.
→ Grabs the actual production 3D model, not a screenshot or a render
→ Auto-names every file by product and color, like "BILLY - White.glb"
→ Works on every language version of the IKEA site
→ Opens directly in Blender or any home-planning 3D software
→ Drop the real furniture into your real room at exact dimensions before you buy
→ One script, no account, no signup, runs in your browser
Here's the clever part:
You're not generating anything. The model already exists. IKEA streams it into your browser every time you tap "View in 3D." The script just saves the file that's already sitting there.
So you can furnish an entire room in 3D out of real IKEA pieces, sized exactly right, before spending a single dollar.
1,437 GitHub stars. 60 forks. Still getting updates.
Free. Full source is public on GitHub.
TODO EL MUNDO DICE “PONELE UNA CACHÉ”.
NADIE TE MUESTRA EL ACANTILADO.
Breakscale no es un load tester.
Es un simulador de system design:
armás el sistema, subís el tráfico
y ves DÓNDE se rompe.
En el video:
2 req/s → todo verde
840 req/s → 99% failing
la API no es el cuello
las dos bases están al 99.9% busy
Eso es lo que un diagrama estático nunca te enseña.
Qué trae:
→ 33 componentes (LB, cache, queue, replicas, circuit breaker…)
→ 23 ejemplos (retry storm, cache aside, sharding, autoscaling)
→ p50 / p95 / p99 medidos, no inventados
→ el trabajo abandonado sigue ocupando slot
→ chaos: tumbar un nodo y ver si el diseño aguanta
No reemplaza k6 contra tu prod.
Te entrena el ojo antes de tocar prod.
Open source.
Demo: https://t.co/59CXx8t8rl
Repo: https://t.co/c7SciX4HKm
Guárdalo si diseñas sistemas
y estás harto de rules of thumb.
Holy shit... Someone built a free AI that watches your security cameras and describes what happened in plain english.
It's called Clearcam. It bolts onto any security camera you already own and turns it into a smart one.
No new hardware. No cloud. The AI runs on your own machine.
→ Spots and tracks people, cars, and objects the moment they appear
→ Saves a clip of the exact event instead of 24 hours of an empty driveway
→ Writes a plain english summary of each event with Qwen3 VL, so you read what happened instead of scrubbing footage
→ Makes every event searchable, so "the person at the door last night" actually pulls it up
→ Runs fully local. Your footage never leaves your machine.
Here's the part that stings.
Ring, Nest, and Arlo all charge a monthly fee to unlock person detection and saved event history on cameras you already bought. You pay rent to watch your own front door.
Clearcam gives you the AI brain for free and keeps it on hardware you control.
Runs on Mac and Linux. iOS and Android apps included.
1.2K stars. GPL-3.0. Open source.
Locally Uncensored is one desktop app:
- uncensored/abliterated chat
- This one ships abliterated weights,
- coding agent with plan + file edits
- images + video via bundled ComfyUI
- 12 backends auto-detected
- no Docker, no cloud required
- https://t.co/py7hjMg7iU
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
https://t.co/T457hVrQjz
You can now run Uncensored Qwen3.8-27B locally on 16GB cards that actually speed, no refusals.
- Most GGUFs kill MTP, This one didn’t,
- Architecture-aware quant + no refusals.
- This one keeps the hybrid stack, keeps multi-token prediction, and uses architecture-aware mixed precision instead of a blunt 4-bit coat.
- YMQ mixed-precision presets: S-Pro 12.5GB, M-TI 12.9GB, M 14GB
- YMQ-Compiler does AutoRound-style bit allocation
- more bits on reasoning/Mamba gates, fewer on dead weight.
- Keeping quality after quantization is not.
That’s the difference between “it loads” and “it stays coherent.”
- https://t.co/CCwH4lLv27
The Best LTX-2.3 Uncensored NSFW video generation model you can run locally on 15GB,
- 258,775 Downloads
- Supports both text-to-video and image-to-video
- focused on smoother motion and better output
- with GGUF, fp8, bf16 and int8 versions available.
- Updated workflow included for better negative prompting.
- Multiple quantized versions ready for download,
- Sharing in case it helps your creative process.
- https://t.co/zuRQQ6mA5E
J'utilise beaucoup Obsidian avec Hermes / Claude Code pour stocker toute la partie mémoire permanente, veille et autres sujets à ne pas oublier
Du coup je m'autorise quelques tips :
- ne pas le mettre sur un dossier synchro iCloud ça le fait déconner tout le temps (filesystem)
- avoir des jobs récurrents qui cleanent le vault, mettent à jour les pages en auto ou sous validation ça évite de rendre tout has been
- forcer la prise de note à ajouter des liens entre les fiches du vault, les agents sont un peu flemmards nativement mais c'est tout l'interet du vault
- avoir un job qui transforme le vault en action dans asana ou autre outil de tracking des actions ça évite le côté PNJ de Notion où tu travailles à documenter le travail de documentation
- ajoutez y les transcripts de vos réunions, les prépas de vos weeklies et rattachez ça à des fiches "people" et la magie de la mémoire va opérer
- utiliser QMD pour l'indexer et les recherches c'est vraiment plus performant que laisser l'agent galérer https://t.co/TDZCtIrc3Q
- utiliser https://t.co/M7mNq7vnja sur le vault permet d'accéder de partout au vault comme si c'était un site web (mettre le vault sur Tailscale pour ne l'ouvrir qu'à vous) et surtout ça permet d'éditer les pages et donc les .md à la volée au lieu de devoir aller les ouvrir
- le design de silverbullet est un peu rebutant et des features sont pas dingues (recherche), votre agent fera un truc adapté en 5 minutes, la beauté de l'open source
Sn1per is a platform for automated penetration testing and attack surface management that orchestrates over 90 third-party tools for reconnaissance, scanning, and exploitation.
https://t.co/rLgJQlnEzt
Turns YouTube videos into structured, Obsidian-ready Markdown notes with transcripts, metadata, and chapters instead of raw text.
https://t.co/eoLcECFcDN
Your dad shot a video on a camcorder in 2003.
It is 480p. Grainy. Blurry. On a 4K TV it looks like the screen is covered in fog.
It is the only video of your sister's first steps. The only recording of your grandmother's voice. 480 pixels wide. That is all the camera captured.
You cannot reshoot these moments. They are gone.
Topaz Video AI charges $299 a year for a subscription that fixes this. In October 2025, Topaz retired the one-time license. Cancel the subscription and lose access to the app.
A developer in Montréal who goes by K4YT3X has been building the free version since February 2018. It started as a student hackathon project at Hack the Valley II. Eight years later it has been downloaded over 1.1 million times from GitHub. AGPL-3.0 licensed.
It is called Video2X. It takes your 480p video and upscales it to 1080p or 4K. Frame by frame. The AI studies each frame and generates the detail that was never captured. Faces sharpen. Text becomes readable. Textures appear.
Not stretched pixels. AI-generated detail that was never in the original footage.
Four models built in.
Real-ESRGAN for real-world video. Faces, landscapes, home footage.
Real-CUGAN for anime and animation. Sharper edges. Cleaner cel lines.
Anime4K v4 for real-time upscaling through MPV shaders.
RIFE frame interpolation. This is the part nobody expects. Video2X does not only upscale resolution. It upscales time. Your 24 fps home video becomes 60 fps. Smooth. Fluid. Like a completely different camera recorded it.
480p at 24 fps becomes 4K at 60 fps. Resolution and motion transformed together.
Runs on Windows, Linux, and Docker. Vulkan GPU acceleration on NVIDIA, AMD, and Intel Arc.
Topaz Video: $299 a year, forever.
Video2X: $0. Local. No watermark. No cloud. No subscription.
21,030 stars. 1.19 million downloads. One primary maintainer.
AutoClip finds highlights in your video. UVR5 separates instruments in your music. Video2X brings your old footage back to life.
That 480p video from 2003 has been waiting for this moment.
(Link in the comments)
Self-hosted Clearcam turns any security camera into an AI monitor that detects and tracks objects, sends mobile alerts, and writes plain-language summaries of each event so you can search for what happened.
https://t.co/pzdwatS4sG