In partnership with
Hi tech & AI enthusiasts, this week's highlights:
Runway Built Software That Draws Itself While You Use It
A Video Model That Makes Clips Faster Than You Can Watch Them
The Free Mystery Model Finally Has a Name
Learn in The Lab: Hand Your Content Repurposing to ChatGPT Work
AI Image of the Week: Turn a real photo into a travel journal cover
Let's get into it.
▲ AI INTERFACES
Runway released Solaris, a model that generates a website or app as a live video stream instead of running code underneath.
Every frame is drawn as you click. There is no app sitting behind it.
What you need to know
No code: Normally a design gets translated into code before anyone can use it, and detail gets lost on the way. Solaris skips that step. What you see is the software.
Drag & Drop: Drag a shirt onto yourself in a virtual shop and it goes on. Ask to move a table and the room re-lights. Nobody programmed those specific actions in advance.
Runway's own testing favours it: In a study of 250 people, Solaris beat a coded version built by a top model 71% to 21% on behaving naturally. Worth noting Runway ran the study itself.
It cannot do text yet: Readable words are still the hard problem, which matters a lot for interfaces. Long sessions drift too. Early access is invite only.
Key Idea
▲ Why it matters: Every website you have ever paid for is a fixed set of screens somebody designed. This is a new way. A scene that rearranges itself around each visitor while staying on brand. It’s not ready for businesses yet, but it is a glimpse of what future work interfaces could look like.
▲ AI VIDEO
MiniMax H3 Max generates video quicker than the video plays. This is a pretty big milestone. As in demand streaming is now possible.
What you need to know
5 seconds of video in under 3: Video can be produced live. Imagine watching your favourite TV show and being able to choose different episode directions. like this Big Bang Theory example.
People are already building with it: Like livestreams where the audience types what happens next, prompt-as-you-play games, and a TikTok-style feed that generates each clip as you scroll.
The launch discount just ended: Rates moved on 1 September to roughly $0.05 a second at lower quality and $0.08 at higher. A 10 second product clip lands under $1.
It reads a starting image: Feed it a product photo and a line of direction, and it animates from there rather than inventing the scene.
Key Idea
▲ Why it matters: At $0.08 a second video you can now experiment and test different ideas. 50x variations of a product ad costs less than one afternoon of a freelancer's time.
▲ OPEN MODELS
Last week we covered Ox Alpha, the anonymous free AI model.
This week Chinese lab Z.ai has now confirmed it was theirs, released it as GLM-5.3-Flash, published the weights free to download, and priced it at a fraction of its rivals.
What you need to know
The original suspect all along: Z.ai was the leading guess when it was still anonymous. It topped OpenRouter with roughly double the traffic of second place, the biggest debut the platform has recorded.
Around 1/10 of the going rate: It scores level with the current mid-tier models from OpenAI and Google on independent scoreboards, while costing about 1/10.
Free to run yourself: The weights are public under a permissive licence, so it can run on your own hardware if client data cannot leave your systems.
It ran on Chinese chips: Z.ai says the entire free week was served on roughly 100,000 domestically made chips, and that the setup now serves work about as cheaply as the American hardware everyone else queues for.
Key Idea
▲ Why it matters: The price story is the obvious one. This is the cheapest route to decent intelligence available right now. The other is the hardware. If China can serve a chart-topping model without Nvidia, the export controls that shaped the last two years of this industry matter less than everyone assumed. That reshapes who can afford to compete, and it will keep pushing prices down.
▲ EVEN MORE NEWS
Other big ideas from the past week.
▲ OpenAI Is Pulling Its Models Out of Cursor: After Cursor was acquired by SpaceX, OpenAI is winding down the contract that supplies its models, with a shutoff date of 12 November. Sidenote this is why the past few weeks has emphasised open source harness & models for work.
▲ Grok Bot Can Now Spend Your Money: xAI added shareable bots as well as agentic shopping, so an agent can complete a purchase using a single-use card.
▲ Anthropic Is Cutting Claude Code Limits on 14 September: The change drew criticism for being framed as an increase when the weekly allowance works out roughly 17% lower.
The Lab
▲ AI EDUCATION
You already have the content. The blog posts, the newsletters, the case studies.
What you do not have is the hour a week to turn them into social posts, so they sit there earning nothing after week one.
ChatGPT Work is built for exactly this type of job.
What you'll learn:
The difference between Chat and Work, and when the switch is worth making
When to run it in the cloud instead of on your machine
The brief that produces a usable calendar instead of generic filler
How to steer it mid-run rather than rewriting the output afterwards
How to set it to repeat every month on its own
🔬 Interested in learning about AI for business? Join The Lab waitlist.
▲ AI IMAGE SHOWCASE
Create this type of image with SlideDrift
Trying something different this week. Rather than a prompt to paste into an image model, this is a photo style in SlideDrift.
You upload a real photograph and it comes back redrawn as a quiet pen-and-paper journal page.
The focus is a real photo that becomes that is still yours and not a stock photo.
Try it here: Minimal Travel Journal in SlideDrift
Optional direction to add:
Keep the subject centred with clear space across the top third for a headline. Warm off-white paper tone, single accent colour only.Image Usage Suggestion: Carousel covers where you want a human face without a stock-photo feel, founder and team introductions, case study openers, travel and hospitality client work, event recaps, or any post where a real photo needs to look considered rather than snapped.
💌 Reply to this email with your own version and I'll feature it in a future newsletter.
▲ SHARE YOUR THOUGHTS
Help shape the content you see here by giving feedback
Have specific feedback or want to get in touch? Reply to this email and I’ll get back to you.
Know someone who’d love this newsletter? Forward it to a friend and have them sign up here.
Thanks for reading, until next time.
Stay curious,
Matt Lok, Editor
Brought to you by Metalabs. Digital marketing consulting specialising in AI. Carousels made with SlideDrift.