Radar Log

Production-quality 4K 3DGS demo from a 360-drone-only capture. Pipeline: AVATA360 → Metashape → kotohibi tools → LichtFeld Studio (LFS, with MRNF). First documented success for AVATA360-class capture.

CesiumJS 1.140 + Cesium for Unreal v2.23.0 + Cesium ion ship native 3DGS with hierarchical LOD streaming. Open glTF KHR_gaussian_splatting + KHR_gaussian_splatting_compression_spz extensions co-developed via Khronos+OGC with Bentley, Autodesk, Esri, NVIDIA, Niantic, XGRIDS. Streams city-scale to sub-cm without performance collapse. WASM-powered sorting.

Google Earth now imports industry-standard SHP shapefiles as cloud-native layers, GLB 3D models for architectural mockups/massings, and elevation profiles in the measure tool. Live now on Google Earth Web (1B+ users, no install).

Claude now connects natively to creative tools. Blender connector live: debug scenes, build tools, batch-apply changes from Claude. Connector ecosystem announced spans Blender, Adobe Creative Cloud, Autodesk Fusion, Ableton Live, Splice, SketchUp, Resolume Arena/Wire, Affinity. @claudeai original post: 6.3M views, 14k bookmarks. CopyRebeldia frame: "this is Anthropic entering the entire creative industry in one stroke."

Most analytical of 4-post 360-drone-splat cluster (kotohibi 53ec3f8a, bilawalsidhu e986622c, asimahmed 23bb50cc, Azadux). Practical specs: statue in mins, building in 5 min, multi-block in 10 min. Can fly within streets (close to subject without losing FOV alignment). Omnidirectional obstacle avoidance. Trade-off: GSD lower than non-360 drone — mitigate with closer flight or paired non-360 detail pass. No DJI SDK / waypoint repeatability yet (manual only).

Browser-based, no-install, free 3DGS reconstruction. Ingests photos, videos, OR 360 captures. Outputs splat + point cloud, downloadable. By Over the Reality (overthereality.ai).

Co-founder of Over the Reality showing browser-native city-scale 3D map viewer in development. Companion to FreeGaussian.ai. Same OVR team building both capture AND viewer in-browser.

HKUST team releases purely autoregressive 3D foundation model. ~20 FPS on 518x378 over 10,000+ frame sequences. Beats DA3 / VGGT. End-to-end learnable, no optimization or post-processing. 1.35M views.

Beff (e/acc, 218k followers) reacts to Claude Autodesk Fusion connector. Companion to chesterzelaya post (2.19M views) showing same connector going viral in engineering community. Specific use case: "create and modify 3D models through conversation."

81-second 4K demo of virtual camera flying through tree canopy down to ground level on a 3DGS captured by a 360 drone. Tree canopies — the hard case for traditional aerial photogrammetry — reconstruct cleanly. Cross-link to parent AVATA360 entry (53ec3f8a). 4-post 360-drone-splat cluster in 2 days.

New 3D representation that ray traces AND rasterizes in real time at "no compromise" quality. Claims rasterization at 3DGS-class speed AND ray tracing at Radiant Foam speed. If it holds, sits on the Pareto frontier vs 3DGS. Project: powerfoam.github.io | Paper: arxiv.org/abs/2604.24994 | 10s demo.

Sneak-peek of upcoming QGIS OpenGeoAgent plugin. Run geospatial analysis + visualisation through natural-language prompts inside QGIS, backed by 480+ WhiteboxTools functions. By Qiusheng Wu (well-known geospatial-Python author). 53s demo, 246 likes.

Open-sourced reconstruction + auto-annotation of 3D scenes from internet videos. Massive dataset for 3D VQA, navigation, embodied AI. Solves data scarcity without manual scanning/labeling.

Claude Code now sends phone push notifications when long tasks finish or require input. "Walk away from the terminal."

Echoes Anthropic creative-connector launch with strategic framing: "From 2026 onward, apps that can't connect to Claude might not survive." Demo shows Blender + Polyhaven assets + natural language for terrain generation.

9.3k likes, 2.19M views — viral in engineering+drone community. Same Fusion connector content as @beffjezos + parent Anthropic-connector entry.

Native equirectangular panorama generation in ChatGPT web with built-in interactive 360 viewer ("enter 360 world" button) — no Three.js wiring required. Prompt recipe: 'equirectangular panoramic image, high detail, immersive, cinematic lighting, 2:1 aspect ratio'. Author also ships Codex+Three.js HTML viewers as fallback.

Higgsfield ships an MCP server for agentic content generation. Replaces web-only path. Usable from OpenClaw, Hermes Agent, NemoClaw. Adjacent to Anthropic's creative-connector ecosystem (a3483511) — 2026 = year every creative platform ships a connector.

Real artist gives Claude a donut reference image. Claude autonomously matches colours, sprinkles, frosting, lighting in Blender. 19s demo, 90k views, 1.2k likes, 734 bookmarks. Cross-link parent a3483511 (Anthropic creative connectors). Most concrete demo in connector cluster — image-conditioned modeling with material+geometry+lighting matching, not just text-prompted scenes.

Tool that takes hours of raw mp4 and lets you search for any scene/movement by description, then instantly cuts and exports the matching clip. 100% open-source, runs locally. Tool name not given in tweet — likely a CLIP/SigLIP frame-embedding indexer with ffmpeg cut output. 212k views, 3.4k likes, 234 retweets.

David Hollin (Niantic Spatial) demos robot localization against a high-fidelity 3DGS of Mount Tamalpais captured "in just a few minutes" with an Insta360 X5. Robot-localization quality is a stricter bar than visualisation — confirms 360-camera splat quality is real. Cross-link parent AVATA360 cluster (53ec3f8a / 8c525ba4 / e986622c).

Google released DESIGN.md as open-source spec for portable interface rules across AI agents and platforms. Critique frame: Anthropic Claude Design has restrictive rate limits; same devs can do it via Claude Code without those limits.

Prototype where the entire visual surface is rendered by a model frame-by-frame — no HTML, no layout engine, no code. By Zain Shah (@southpkcommons). Linus (@thesephist) frame: "hard to unsee the future from demos like this."

30+ min video by Anthropic's Head of "Coding Agents" research. Frame: "more about vibe coding than 100 paid courses." Direct from team building the underlying capability.

fal.ai engineer notices Google Earth GLB import. Cross-link to parent Google Earth radar entry. Builder-side validation that GLB-into-Earth is on the AI-tooling community's radar.

Free authoring app for iPhone + Apple Vision Pro that lets AR creators build and share scenes. 52s demo. Author is iOS/visionOS engineer (ARKit/RealityKit/SharePlay). Small reach so far (4 bookmarks, 6 likes), early signal not viral launch.

Apple released a 400k-image dataset for training text-guided AI image editors. First large open dataset of real images + complex editing instructions (single-turn, multi-step reasoning, preference data for reward training). Apple used Google Gemini 2.5 Pro and Alibaba Qwen to generate edit instructions and verify quality, then open-sourced.

Roblox extends earlier 4D-Generation (mesh-based, e.g. car-5 schema) with 30+ new schemas powered by Procedural Model Generation. Output is procedurally editable AND functional Roblox asset (animated/interactive) — submarines that dive, jet planes that fly. Schema-driven so each category gets predictable rigging + parameters.

Companion bookmark to AVATA360 pipeline post. Deep Track = depth/tracking feature in kotohibi's 3DGS toolset (or SuperSplat). YouTube demo: JNd96j9VA24.

Earliest viral signal of Anthropic's Fusion connector arc. el.cine (128k followers) reacting to Opus 4.7 doing CAD on April 16 — 12 days before the formal Fusion connector launch on April 28. Embedded video by @adamdotnew.