Amir Ferdos

Architectural design meets AI. Join me, a CG artist and AI enthusiast, for tutorials, insights, and experiments in CG art and cutting-edge architectural design.

[Jiraphon Yenphraphai @JYenphraphai]

[1/4] 🏛️ Building Rome from a Single Image 🏛️

We reconstruct 3D scenes from the input image, including geometry beyond the observed view across indoor, outdoor, and large-scale scenes.

👉 Project: https://t.co/ivayd3Pk8R

[Grounded Superintelligence @GroundedSI]

Today, we are announcing GroundedSLAM.

LaMAria is the most difficult egocentric SLAM benchmark to date. We are the first to close the gap on Aria’s SLAM and eclipse all other methods by a wide margin.

SLAM is the foundation of all 3D estimation. We think this is a big unlock for better robotics 👇🧵

[Victor M @victormustar]

Thanks Google, Embedding Gemma 2 is a big deal 🫶

Multimodal embeddings (text, images, video, audio) can run on-device in the browser. No server, no API key, ~20–70 ms per query on WebGPU. Everything stays on your machine (ofc).

Now go build (new) things with it 🚀

[Maziyar PANAHI @MaziyarPanahi]

DINOv3 + SAM 3.1 is a ridiculous combo 🔥

i turned a surgical tray into a visual search engine.
One instrument as the query. The rest ranked by visual similarity. ALL LOCAL!

specialized vision models are still cooking!

[Azlen @azlenelza]

Continuation of my old video-as-3d-object demo

Always wanted to extend it further with segmentation so you can search for any type of object within the scene

[Stav Zilbershtein @mightyking]

Ok guys this is crazy

FLUX 3 image precision edit is real!

GPT 2.5 is dead in the water

[Ryohei Sasaki@engineer @rsasaki0109]

AMB3R-SLAM: Kilometer-scale SLAM
with Hierarchical Backend
https://t.co/9iCtzB9ijY
TL;DR: 1) A training-free, model-agnostic SLAM system that scales to kilometer-scale trajectories in real time on one consumer GPU; 2) A hierarchical backend that enforces local, mid-level, and global consistency without bundle adjustment, so dynamic scenes work out of the box; 3) One framework that extends to stereo, RGB-D, and LiDAR.

[Junhyeong Cho @JunhyeongCho]

Where on the floorplan was this photo taken? 📍
Excited to share SceneAligner, accepted to #NeurIPS2026!
We tackle this by reconstructing the scene in 3D and aligning it to the floorplan. Joint work with @ruojin8 @ElorHadar [1/8]

[Ultralytics @ultralytics]

Make workplace monitoring more intelligent with Ultralytics YOLO26! 🔒

Detect people and faces in real time to support occupancy monitoring, access control, and security operations.

Discover more ➡️ https://t.co/S8NzfpX1B1

#Ultralytics #YOLO26 #Survillence

[Ultralytics @ultralytics]

LLM-powered auto-annotation is now in the Ultralytics Platform! ⚡

Use models from OpenAI, Anthropic, Gemini, Kimi, https://t.co/zYQQnji56V & DeepSeek to turn images into labeled training data.

Choose a model → Click Predict → Review annotations.

Try it today on the Ultralytics Platform ➡️ https://t.co/7djXRVUIv7

#Ultralytics #ComputerVision #DataAnnotation #LLM

[AK @_akhaliq]

LEGO-Anything

Coding Agents for 3D Scene Reconstruction

paper: https://t.co/lWpBbRNuP8

[mirrash.eth 🦇🔊 @mirrash7]

I just labelled over 1000 segmentation masks for utility poles in under 90 s 🤯

-GPT Astra for box detections
-SAM3 for masks on those boxes

How long would it take a human to do the same??

[ModelScope @ModelScope2022]

@InSpatio_AI releases InSpatio-World 1.5, turning images and videos into real-time, explorable 4D worlds.
🤖 https://t.co/o42sQcECQD
📄 https://t.co/12tiNfFKAx

🏆 The 1.3B model scores 68.72 on WorldScore-Dynamic, ranking #1 among evaluated real-time and interactive methods while reaching up to 24 FPS.
🌐 A single image, four images, a panorama, or a video can become a navigable scene with wide viewpoint changes and target camera trajectories.
🎥 Next-view prediction preserves motion, spatial structure, and visual style across space and time, with support for bullet-time creation.
📐 For video inputs, Depth Anything 3 automatically estimates missing depth and per-frame camera intrinsics and extrinsics.
🧠 Spatiotemporal autoregression maintains a persistent world state for consistent long-horizon exploration.
📜 Code: Apache 2.0. Dependencies retain their own licenses.

[InSpatio｜影溯 @InSpatio_AI]

A photo shouldn’t have to stay a photo.
A video shouldn’t have to stay a video.
They can become worlds.

🌍 InSpatio-World 1.5 is here.
Our real-time 4D world model now supports:
🖼️ Single-image input
🖼️🖼️ Multi-image input
🌐 Panorama input
🎬 Video input

From visual input to a dynamic, navigable 4D world — explore beyond the original viewpoint and move through the scene in space and time.
Open-source.Real-time. Ready to explore.

Live Demo:
🔗 https://t.co/Fn9OTgg4eG
Code:
🔗 GitHub - inspatio/inspatio-world-v1.5
Weights:
🔗 https://t.co/3ujZ4Z66na
Project Page:
🔗 https://t.co/ZvgeTmZEnP

[Yoland Yan @yoland_yan]

Please join me to welcome our Intrim Chief DevRel Officer, Deep, to the team!

[Sadao Tokuyama @tokufxug]

高反射オブジェクトの3DGS「フローター（ノイズのように浮く粒）」を抑えるDAVINCI

🔹Killian Rando氏が開発
🔹メッシュ＋拡散・反射成分を分離学習
🔹異方性・球面ボロノイ＋NASGaborで外観を表現
🔹3DGSより高PSNR、容量は4分の1
🔹ラスタライズでリアルタイム描画

情報元はリプ欄。

[ComfyUI @ComfyUI]

Recraft V4.1 Flash is live in ComfyUI

1.3 seconds per image. The fastest image model on the market.

Iterate at the speed you think, then Refine your keeper to 2048×2048 Pro quality.

Try it with the link below ⬇️

[ComfyUI @ComfyUI]

Comfy Router is live

One API for frontier image, video, 3D, and audio models.

Same model string. Same arguments. No new SDK, no new key, no redeploy.

What Comfy Router gives you:

→ Explicit routing. You name the provider, we call that provider. It's down? The request fails there. No silent fallback.

→ Every job returns the provider that ran it. Log it, bill it, debug it.

→ Async. submit() returns a request ID immediately. The queue retries 429s and transient errors until a slot opens. subscribe() submits and polls to completion.

→ Batch-friendly. Queue a few hundred jobs, hold the IDs, pull results as they land. Nothing blocking on a 5-min video render.

→ 24h retention on inputs and outputs, then deleted.

→ Comfy credits. No sub, no Router fee.

Providers at launch: Comfy. Runware, Wavespeed, Fal, Higgsfield. Multi-provider where the model supports it.

Get Your API Key with the link below. ⬇️

[Yoland Yan @yoland_yan]

OpenRouter for generative media is here 🦾

[milad_kambari @MiladKambari]

I remember one or two years ago, I was recording a course about preparing AI-generated 3D models for sale.

I was working on a model very similar to this one, and I spent hours trying to remesh and unwrap it. I had to switch between several different programs just to find and fix the problems in the model before I could even start the remeshing process, and then finally unwrap it.

At that time, one of the best options I had for automatic unwrapping was Substance Painter, but of course, the result was nowhere near what I actually needed. So I still had to spend hours in RizomUV just to get an acceptable result.

What I never imagined was that one day, I would be the person who actually solved this problem for myself and created an addon that can unwrap a model like this with just one click.

It’s strange how much our career paths can change over time. Sometimes they take us in directions we never expected, and that’s what makes the whole journey so interesting.

https://t.co/QfR6rJ9U1J

#bleder
#b3d