Fast general-purpose agents could replace specialist robotics models
After trying GPT 6 Astra Ultrafast controlling a robot in simulation, the prediction is that very fast general-purpose agents could replace specialist robotics models.
AI could revive virtual reality and the metaverse
AI could boost independent production of videos, games and virtual reality modifications. Making games easier to adapt for VR could attract new players and renew momentum behind metaverse development.
A transformation may only just be beginning
The old world may be ending and everything may be about to change. The transformation may only just be beginning.
Our Picks
Ideas, videos and reflections that caught our eye. Every card links to the original content and its author.
A browser-based 3D world built with AI in twelve hours
The claimed result is a browser-based 3D world built in twelve hours without a coding background. Opus 5.5 reportedly handled mechanics and integration, Sonnet 5.5 some objects, and Meshy additional models.
Successive edits to an Olipop advert with Ideogram
A sequence of precise edits to an Olipop advert puts Ideogram’s new model to the test. The images appear to hold up across edits; Ideogram’s 4.5 announcement is cited.
Two AI models take on the same launch-film brief
Two unofficial launch films made with GPT 6.1 Sol and Sonnet 5.5: the same brief, music and Gemini 3.8 TTS voiceover, with the model names revealed at the end.
Ten ideas for building small gadgets with Schematik
Schematik is presented as an assistant for choosing components, writing code and assembling devices. Ten suggested projects to customise include a plant with celebratory lights, a gesture-controlled lamp and a weather display offering clothing suggestions.
A rainbow animation drawn frame by frame in code
How the Rainbow Got Its Colors is an animation whose frames were reportedly drawn one by one in code by Claude Sonnet 5.5, rather than generated directly as video.
Ideogram 4.5 available on the Leonardo platform
Ideogram 4.5 is available on the Leonardo.Ai platform and is presented as a model for successive image edits, from product positions to colours and labels, while preserving the parts users want to keep.
Chess with a paper-doodle look
A chess project designed to look like doodles on paper, made with Claude Sonnet 5.5.
Fast general-purpose agents could replace specialist robotics models
After trying GPT 6 Astra Ultrafast controlling a robot in simulation, the prediction is that very fast general-purpose agents could replace specialist robotics models.
AI could revive virtual reality and the metaverse
AI could boost independent production of videos, games and virtual reality modifications. Making games easier to adapt for VR could attract new players and renew momentum behind metaverse development.
Ideogram 4.5 preferred to GPT Image 2.5, with hopes for local use
Ideogram 4.5’s image editing is rated above GPT Image 2.5. There is also enthusiasm for the prospect of the model’s weights being released and for running it locally.
When will AI-generated 3D assets start appearing in games?
A strawberry attributed to Claude Opus 5.5 could fit into a farming game.
A transformation may only just be beginning
The old world may be ending and everything may be about to change. The transformation may only just be beginning.
Ideogram 4.5 for repeated image edits
Ideogram 4.5 arrives on Krea’s platform with the stated aim of limiting visual defects, pixel shifts and colour drift during successive edits to the same image.
Far too much “SI” lately
Far too much “SI” lately, with a link to Soap Bubble and no expansion of the abbreviation in the text.
An AI assistant designs editable circuit boards
HeyPCB connects AI assistants to a circuit-board design tool. It produces an editable KiCad project, taking a request into work on components and connections.
Recurring negative thoughts and memory: a link to understand
A 2020 study of people over 55 found an association between repetitive negative thinking, cognitive decline and proteins linked to Alzheimer’s. This is an association, not proof that worrying causes the disease.
Image: Nephron, Wikimedia Commons · CC BY-SA 3.0.
Generate and edit images, including transparent backgrounds
Qwen-Image-2.1 is a model with available weights that combines image generation and editing. The announcement showcases transparent layers and compositions guided by multiple reference images.
Bingo, a small robot imagined as an imaginary friend
Moving Parts gives its digital companion an expressive body. Called Bingo, the robot is being developed by Joe Tsao with engineering support from Gentle Systems.
Sell the finished work, not just the software
To build services around AI: choose a specific task, automate part of it and keep human review where it matters. A perspective on organising work, not a promise of returns.
GUIDE · The author’s analysis, with examples and opinions.
Explore a video like a sculpture in space
Frames become a volume to explore, with time as the third dimension. The Video Summagator technique comes from research published in 2012 and is resurfacing in these demonstrations.
A technique from 2012; demonstration shared by @poetengineer__, credited to Bradley Tangonan.
Alma: five minutes in an unsettling doll shop
Rodrigo Blaas’s animated short follows a girl drawn to a doll that looks like her. A small story from 2009 to rediscover.
A camera built from simple materials
A short video shows a photography experiment with a homemade camera, returning to the simplicity of a pinhole without a lens.
A portrait comes to life in a Solaris demo
Solaris, Runway’s project for interactive experiences, brings a portrait to life in a visual demo. This is a demonstration of a product in development.
A living-room wall becomes a game you control with your hand
A game projected onto a living-room wall uses a camera to track hand gestures. The demo includes an account of how it was built with Codex.
Impossible scenes, generated on a home computer
Everyday places interrupted by something surreal, in images reportedly created with Qwen on a home computer. The prompts are also available.
Ask an assistant to use services on your behalf
Muse connectors let developers connect their services to the assistant, turning chat requests into actions in apps.
Will Smith and spaghetti, this time drawn with code
The familiar Will Smith and spaghetti experiment returns as a code-driven animation. A creative test whose rough edges are part of the exploration.
Turn your pet into a little paper character
A prompt for imagining cube-shaped paper pets. These are AI-generated images; the thread explains the extra steps needed to make a foldable template.
A face engraved almost entirely with one line
In 1649, Claude Mellan made the Holy Face with a spiral starting at the nose. The line widens and narrows to create light and shade.
What if we looked for intelligence beyond the brain?
In this essay, Michael Levin argues for seeing intelligence in organised processes, rather than only in brains. A theoretical perspective to explore and debate.
The robot that drives from the floor onto a wall
VertiGo uses two tilting propellers to press its wheels against a wall. This 2015 project from Disney Research and ETH Zürich is still striking to watch.
A shoe takes shape under a robot’s spray
With LightSpray, On makes a shoe upper by spraying a filament onto a mould. The video shows the manufacturing process.
A flat sheet folds itself into a tiny robot
This MIT prototype, presented in 2015, folds when heated and moves using external magnetic fields. It can walk and swim.
A thin fabric that helps stop noise
In a 2024 study, MIT researchers and collaborators controlled a fabric’s vibrations to reduce transmitted sound. A laboratory demonstration, not a curtain already on sale.
Knitted fabric that changes shape and stays that way
Harvard researchers created textiles that snap between stable shapes. By adding conductive yarns, they also demonstrated small switches and sensors.
Old LCD handheld games return as a modern adventure
Future Knight borrows the look of classic LCD handheld games for a modern action game, created by Studio Koba and Aeternum Game Studios.
Design proteins with AI and test them in an automated lab
Adaptyv and Anthropic launch a protein-design competition with funded experiments and open results. The proposed designs still have to pass laboratory testing.
The Titanic’s voyage becomes a browser animation
A code-built animation credited to afosxone in the post sharing it. A creative experiment you can explore in your browser.
Robots try household chores in 30 unfamiliar homes
Figure shows its robots tidying, folding towels and making beds. In the company’s tests they complete 56% of tasks: a demonstration still far from a reliable household helper.
A PlayStation controller drawn with code
A vector-graphics recreation of a PlayStation controller made with AI assistance. A visual test, rather than a ranking of model capabilities.
Search the web as though it were the past
Exa Snapshot searches historical copies of web pages. Its launch video was made by Donald Jewkes, who also describes his work with AI tools.
Embroider with your mouse, right in the browser
Stitchcraft turns cursor strokes into digital stitches, knots and threads. A demonstration of the experiment includes a version to try online.
An AI assistant alongside your Word document
The demo of ChatGPT inside Word shows how to turn notes into drafts, review paragraphs and suggest edits without leaving the document.
An AI assistant designed for lawyers’ work
Astra for Law combines tools and context for legal practice. The video presents the offering as support for professional work.
A desktop aquarium whose fish follow your cursor
An interactive Mac wallpaper built with AI assistance: moving the cursor changes how the fish behave.
Web photographs become little animated doodles
An AI assistant chooses images and draws over them with JavaScript. The video collects the experiment’s results.
Pokémon comes to life on a Daylight screen
Pokémon adapted to a Daylight device with touch controls and a wider map view. The video shows a test session from this personal project.
A robotic hand walks on its own fingers
Fingers as Legs teaches a robotic hand to move and manipulate objects using the same fingers. The researchers demonstrate the prototype and describe their method in a paper.
Enhancing an image, between recovered and invented detail
In an upscaler demonstration, zooming reveals increasingly sharp details. Generative enhancement does not guarantee that added details existed in the original.
A conversation that also uses what the camera sees
A Gemini 3.8 Live demonstration combines voice assistance with a camera feed. The Extended Thinking variant is intended for multistep tasks.
Voice, music and video in the same assistant
The expansion of ElevenLabs’ MCP to voice, music, images and video brings media generation and transformation into the assistant used for work.
Twelve animation principles to give movement weight
An animation is refined through several exchanges with Astra. Cartwheel MCP helps with foot movement, while effects and impact frames are drawn in Grease Pencil.
Stalker sped up reveals the movement of the camera
A sped-up edit reveals the rhythm of Stalker’s long camera movements: another way to study composition and space in Tarkovsky’s film.
Designing hardware with a modular system
An atech.dev demonstration explores building hardware projects with less manual wiring. The link leads to the project’s original presentation.
An animated mockup for trying out a phone design
An interface is placed in Ultramock’s iPhone Duo mockup with a blur transition. The video documents the tool’s workflow.
An assistant for working inside After Effects
Motion Designer is a plugin connecting ChatGPT to After Effects. Its demo shows changes to animation and expressions inside an editable project.
A simulated crowd built from individual collisions
A Houdini demonstration combines running characters, collisions and ragdoll dynamics. Local interactions produce chaotic collective motion.
A butterfly animated through 600 Photoshop layers
A butterfly animation reportedly built with 600 editable Photoshop layers: an Astra-assisted production experiment.
A voice assistant that listens while speaking
GPT-Live-1 arrives in the API with a demonstration of two-way voice conversation for developers integrating these interactions into applications.
From a photograph to a short video inside a chat
Effects 2.0 turns a photograph into video through ChatGPT in the examples shown. A product demonstration, not a promise about results or reach.