Olst

3D artist/Freelancer/Blender3d user

[AI at Meta @AIatMeta]

Introducing SAM 3D, the newest addition to the SAM collection, bringing common sense 3D understanding of everyday images. SAM 3D includes two models:

🛋️ SAM 3D Objects for object and scene reconstruction

🧑‍🤝‍🧑 SAM 3D Body for human pose and shape estimation

Both models achieve state-of-the-art performance transforming static 2D images into vivid, accurate reconstructions.

🔗 Learn more: https://t.co/yXcvts8Ogc

[Cartesian Caramel @Cartesian_C]

Cat gaussian splat atomization #b3d #geometrynodes #blender

[MrNeRF @janusch_patas]

First Corporate Sponsor: Core11 GmbH Supports LichtFeld Studio!

Today marks an exciting milestone for LichtFeld Studio: its first official corporate sponsor. Core11 GmbH, a graphics technology company based in Vienna, Austria, has committed €16,000 to the project over the next eight months.

“We believe LichtFeld Studio has the potential to become the fastest splat reconstruction on the market.”
— Dr.-Ing. Gernot Ziegler, CEO of Core11 GmbH

The company offers remote services worldwide, including GPU programming education, performance analysis, and contract-based GPU development. Beyond financial support, Core11 GmbH will also provide valuable engineering expertise in CUDA and graphics programming to further advance the project.

Core11 GmbH’s CEO, Dr.-Ing. Gernot Ziegler, shared:

“Our partner geofront e.U. relies on Gaussian splatting for Web3D applications and requires a performant and versatile reconstruction pipeline for content creation.

We are impressed by LichtFeld Studio's recent evolution and potential, based on a persistent search for higher performance and quality in Gaussian splatting reconstruction methods. Our analysis makes us believe that LFS has the potential to become the fastest splat reconstruction on the market.

It is therefore consistent to support the LFS project with funding. As LFS runs its core algorithms directly in CUDA for NVIDIA GPUs, we will assist with our knowledge in GPU-based optimization and algorithmics.”

LichtFeld Studio is dedicated to developing a high-performance, GPLv3 open-source 3D Gaussian Splatting solution. Our goal is to support both the training and editing of Gaussian Splats while building a vibrant, growing community.

We are actively seeking sponsorship to ensure the project's sustainability. If you're interested in financing this initiative, supporting open-source innovation, and advancing 3D reconstruction technology, please reach out via direct message.

Find the link to our sponsor, Core11 GmbH, in the comments, as well as the link to the full @RadianceFields article online!

[MrNeRF @janusch_patas]

LichtFeld Studio is cooking!

We'll have an exciting announcement tomorrow that will help us shape the future of 3D Gaussian Splatting open source software!

[PlayCanvas @playcanvas]

PlayCanvas Engine v2.13.0 is here 🚀
Your free and open source 3D engine for the web
✅ Streamed LOD system for 3D Gaussian Splats
✅ Splat shader effects framework
✅ Globally sorted splats
[1 / 2]

[Victor M @victormustar]

New: Qwen-Image-2509-MultipleAngles. Very solid model and probably a lot of creative use cases to find with it.

⬇️ Free demo available on Hugging Face

[KIRI Engine - 3D Scanner App @KIRI_Engine_App]

Our 3DGS😤 Now Turbo-Fast & Photo-Real ⚙️✨

High-fidelity 3D Gaussian Splatting used to mean slow convergence, bloated point clouds, and overheating GPUs🐢🔥

So we rethought the whole pipeline.🥰
Instead of growing splats politely, one by one,
we now spawn them aggressively and prune with weighted filtering⚡

Result?
✨2× faster convergence
📦~50% smaller files
🎬and smoother, cleaner, cinematic-grade splats

No black magic. No hallucinated details. Just math, caffeine, and very stubborn engineers😪☕

[Google AI Developers @googleaidevs]

Build an on-device RAG system for private, fast, and offline-capable AI apps.

Learn how to implement Retrieval-Augmented Generation (RAG) using EmbeddingGemma and Gemma 3 1B models with this guide from Google Developer Expert @GSoloupis ↓
https://t.co/3u73iDHdMw

[Historic Vids @historyinmemes]

The day Maikel Melero pulled off one of the most jaw-dropping tricks ever seen in freestyle motocross.

[Massimo @Rainmaker1973]

Microdispensing at the picoliter scale is redefining precision in science enabling ultra-accurate droplet placement for biotech, diagnostics and electronics.

A raindrop ≈ 50 µL is 50 million× larger than a picoliter.

[Blue Nile @blue_nile_3d]

I took some videos from a train in my bleak hometown.
Then brought it into Blender and hopefully turned the greyness into something more dreamy. #b3d #blender

[Olst @Olstblend]

@blue_nile_3d Hah, cool! I think the only question here is whether the right people will notice your work! Because I believe you would do a great job with this task. Your style and vibe are unique and amazing!! 🤟💪

[MrNeRF @janusch_patas]

Structure-aware Long-term Generalizable 3D Reconstruction from Unposed Images

Contributions:
• We introduce SaLon3R for pose-free, long-term generalizable reconstruction with structure-aware learning. This eliminates redundancy and refines artifacts, enhancing rendering quality and geometric consistency. To our knowledge, SaLon3R is the first online generalizable GS method capable of reconstructing over 50 views at a speed exceeding 10 FPS.

• We present a saliency-aware Gaussian quantization mechanism to encode dense pixel-wise Gaussians into sparse Gaussian anchors incrementally. A lightweight point transformer is employed to facilitate interaction between surrounding Gaussians to learn the 3D structure, allowing out-of-distribution prediction and artifact refinement.

• Extensive experiments demonstrate our superior performance in long-term generalizable online reconstruction from unposed and uncalibrated images. With 50% to 90% redundancy removal, our SaLon3R significantly outperforms pose-required methods in depth estimation and achieves state-of-the-art results in novel view synthesis.

[OlstFlow @OlstFlow]

✨ The day has finally come.
I’ve been building solo ng 😅
Prompt Wrangler — a local-first, lightning-fast prompt manager for ChatGPT, Gemini, ComfyUI & more.
Free & open source.
🌐 https://t.co/ZBHTkCOt7d
🛠 https://t.co/GAqigph9fB
❤️ https://t.co/3KxZU75HJ5

[Jukan @jukan05]

I can’t believe this is actually AI.

『Created a live-action Chainsaw Man film set with AI 』

[Qwen @Alibaba_Qwen]

Introducing Qwen3-VL Cookbooks! ��‍🍳

A curated collection of notebooks showcasing the power of Qwen3-VL—via both local deployment and API—across diverse multimodal use cases:

✅ Thinking with Images
✅ Computer-Use Agent
✅ Multimodal Coding
✅ Omni Recognition
✅ Advanced Document Parsing
✅ Precise Object Grounding (across formats!)
✅ General OCR & Key Info Extraction
✅ 3D Grounding
✅ Long Document Understanding
✅ Spatial Reasoning
✅ Mobile Agent
✅ Video Understanding

🔗Link: https://t.co/6ONowbWDTF

⚙️ API: https://t.co/XkK2ubI7Hz

💬 Qwen Chat: https://t.co/PKOQqHlAj4

Explore, experiment, and build the future of multimodal AI! 🚀

[AshutoshShrivastava @ai_for_success]

🚨Google just dropped the Genkit extension for Gemini CLI.
Genkit is an open-source full-stack framework from Google for building, deploying, and monitoring production-ready AI-powered applications.

Video source: Google

[Nikhil Keetha @Nik__V__]

Meet MapAnything – a transformer that directly regresses factored metric 3D scene geometry (from images, calibration, poses, or depth) in an end-to-end way. No pipelines, no extra stages. Just 3D geometry & cameras, straight from any type of input, delivering new state-of-the-art results 🚀

One universal model enables SoTA for:
🔥 Mono Depth Estimation
🔥 Multi-View SfM
🔥 Multi-View Stereo
🔥 Depth Completion
🔥 Registration

… and many more possibilities! – plus everything is metric 🎯

We release code for data processing, training, benchmarking & ablations – everything Apache 2.0!

Details & Links 👇

[Gerard Pons-Moll @GerardPonsMoll1]

Real time online 3D reconstruction of 3D scene and humans represented with SMPL.

https://t.co/SMsxP4iZhT

I don't get tired of looking at these results