6 PM PDT · Best of the Day · Mon Oct 19, 2026 Report
View on GitHub ↗

AI Design Best of the Day · Monday, Oct 19, 2026

Best of the Day header

3D worlds you can step inside are having their moment

The design story that broke out today is Viggle's JST-2, a "physics-native world model" that builds scenes, characters and motion directly in 3D instead of generating flat video frames. Viggle says a scene takes about one second of model time and costs under a cent, that a single photo becomes a 3D character with motion built in and no rigging step, and that you can try it free today. The launch post passed 670K views with more than 800 saves, and it landed on the same day as the most-liked design post of the day: a WebGPU demo of jumping through portals between Gaussian-splat worlds in real time. It matters because 3D has been the slowest, most specialist part of a design pipeline. If photo-to-character and photo-to-scene really work in a browser, you can put real 3D in front of clients without a rigging artist or a game engine. Treat the hype carefully, though. Much of today's JST-2 chatter came as a wave of near-identical "free today" posts that read like a coordinated launch push, and Viggle itself says object physics is still rule-based until JST-3.

Engagement figures were read from the X API, GitHub, Hacker News and YouTube search on the evening of Oct 19 (PT) and will keep changing. Reddit blocked our tools this evening, so Reddit sentiment is not counted.

⭐ Pick of the day

🧊 3D and worlds

  • JST-2, Viggle's physics-native world model (Viggle Research), model/tool. One model that generates 3D scenes (and keeps extending them as you move to new viewpoints), turns a single image into a 3D character with motion built in, and pulls full-body motion from text or ordinary phone video, filling in limbs the camera missed. Why people are losing it: the launch post has 1,037 likes, 828 bookmarks and about 679K views; creators are already comparing it with static splat tools, and Viggle says it beats rival 3D tools, including Meshy and Tripo, in its own blind tests. How to try it: open the free tier, feed it one photo of a room and one photo of a person, and see whether the character and scene are usable as a pitch-stage 3D mockup.

The ranked picks

🧊 3D

  • Portal jumping through Gaussian-splat worlds in real time with WebGPU (@XRarchitect), showcase. Walk through a doorway and you're in a different photoreal captured world, live in the browser, with sound. The creator's bio says he designs at World Labs. Why people are losing it: the most-liked design post of the day at 4,487 likes, 2,039 bookmarks and about 137K views, up from about 1,200 likes at noon. How to use it: a reference for immersive portfolio or launch-site transitions, where each "room" is a captured place instead of a rendered set.
  • SolidCraft, free open-source parametric CAD with a built-in MCP server (GitHub), repo/tool. A pure-Rust CAD app with sketches, constraints and a feature timeline that runs on desktop or in the browser, and that AI agents can drive directly. Why people are talking: 7,286 stars since it appeared on Oct 7, and askNK's video asking whether it ends paid CAD reached 21,624 views in about 7 hours. How to try it: open the browser version, connect Claude through its MCP server, and ask for one small product part you can then edit by changing a dimension.
  • SKY-001, a world model aimed at engineering and CAD (@testingcatalog), news. Skyfall AI's model breaks an object into parts and rebuilds it in CAD as a visual build plan. Skyfall's own tests say it beats GPT-6 Astra and Opus 5.5 on CAD and BIM tasks at a fraction of the cost; nobody independent has checked that yet. Why it's here: 108 likes and 47 bookmarks on a niche engineering launch. How to use it: one to watch if you do product or architectural visualisation.

💻 Code and UI

🖼️ Image

  • GPT-Image-2.5 Sunburst is #1 on three Arena image leaderboards, and free to test for 14 days (Arena), news. OpenAI's latest image model leads Arena's Text-to-Image, Image Edit and Multi-Image Edit boards, and can be picked by name in Direct Mode until Nov 2 at 8 AM PT. Arena doesn't say whether it's related to the "aurora" test models spotted over the weekend. Why it's here: 231 likes and about 21K views on the announcement. How to try it: run your hardest edit (text in the picture, or several reference images) in Direct Mode before access closes.
  • A small custom model that makes exact 16×16 pixel art, where frontier models don't (@GrilliotTodd), showcase. Asked for "exactly 16x16 pixels", his trained model hits the grid every time while GPT Image 2.5 and Nano Banana come out at whatever size they like. Why people are losing it: 216 likes and 162 bookmarks, nearly one save per like. How to use it: for game sprites, a dedicated pixel model (or a converter that forces a real grid) beats prompting a general image model.

🎬 Video and animation

🎮 Games

🧪 Mashups and agent studios

  • Runway's generation tools now plug straight into Grok Bot (Runway), tool. Runway's MCP connector lets you generate images and video from inside a chat instead of switching apps. Why it's here: 398 likes, 159 bookmarks and about 32K views. How to try it: connect it and ask for a storyboard and a first clip in the same conversation where you wrote the script.

🗣️ What people are talking about

Method: starts from today's morning and noon digests, then a fresh evening check of X, YouTube, GitHub trending, Hugging Face trending and Hacker News. Ranked by engagement and velocity, how many creators are copying it, and how new it is. Reddit blocked our tools this evening.

Follow every creator we feature on X: Daily Design Creators on X

Want a deeper dive on any of these, or more of one medium tomorrow? Say the word and the next edition will steer there.

← Back to the Daily Design index