Tech Landscape #438
Real-time video edits, Gemini comes to Chrome in the UK, and another powerful Chinese model hits the market
Hello!
Britain’s second heatwave is over and the weather is back to being season-appropriate, so I’ve been outside making the most of it — hence I don’t have time to write more of an intro.
So, let’s get on with it. Hope you’re well!
Synthetic Content
Decart AI released Lucy 2.5, a real-time video model that supports live edits at 30fps and 1080p quality, with character and background swap, physically-aware visual effects, and object removal, with a high level of consistency. decart.ai
It’s impressively stable. You can see a quick test here ⬇️; note that there’s a slight ~1sec delay that isn’t shown when the source and output are comped together like this.
VEED launched Lip Sync 2.0. It can re-sync videos of speakers with new speech, matching the mouth movements and tone, in clips up to 10 minutes long at 4K quality. It’s available via API, apparently in partnership with Magnific. veed.io
Viggle launched a text-to-motion model, which lets users animate rigged 3D characters from a prompt, in its PINOC app. x.com/Viggle_PINOC
Creative Tools
Reve had a week of feature launches: Reframe can adjust an image to different aspect ratios, either by preserving its original layout (Reshoot) or adapting it (Relayout); Color uses a picker or slider to adjust the colour of any element in an image; Shareable references let users share characters and visual styles; and Layer Groups are for moving and resizing multiple elements simultaneously. It also launched a Teams tier, and made the 2.1 model available through the Reve API, either as straightforward image generation and editing or with full layout control.
You can see the Reframe and Color tools used in my example ⬇️, with the initial generated image shown inset. Reve 2.1 is an impressive model, and the granular editing tools are really good — hopefully good enough to stand out from the (plentiful) competition.
OpenArt added Brand Kit, a way to keep logo, colours, tone, and rules consistent across generations. threads.com/openart_ai
Google Flow Music now lets users upload songs (instrumental) to use them as the foundation of new tracks. x.com/googleflowmusic
Avatars
Google Vids added personal avatars, enabling users to create digital likenesses of themselves (from a selfie and voice recording) that can be used to narrate videos. It’s powered by Gemini Omni, which is also available in Vids to generate and edit clips. blog.google
Gemini Avatars now work with Nano Banana, letting users create images with their own face without uploading a new reference each time. instagram.com
This was previously only available in Gemini Omni videos.
Neither of these features is available in the UK and EEA, I suppose due to stricter privacy requirements.
Tavus’ Magic Canvas is a dynamic visual interface for its PAL video avatars that displays interactive elements (such as calendars, charts, and text fields) directly within a conversation. tavus.io
Assistants & Search
Gemini in Chrome came to desktop users in the UK, with multi-tab context, Google app integrations, repeatable skills, content summarisation, and image transformation. An iOS release is scheduled for next month. blog.google
I tested it out quickly, asking it to look up online reviews of a pair of shoes I was looking at, and to generate an image in the style of an Instagram page I visited, and it all checks out. But this is very new, and I think I need to consider my needs and workflow a little to get the best out of it.
Google’s Gemini Spark agent is rolling out to more countries, with a speed improvement and deeper Workspace integration. x.com/GeminiApp
Still only for Ultra subscribers, still not in the UK and EU.
Moonshot AI released Kimi K3, a huge (2.8-trillion-parameter) multimodal model with native vision capabilities and a one-million-token context window, designed to handle complex coding, reasoning, and knowledge tasks.. kimi.com
I’m trying not to report every new model release (there are so many) but this is notable because it’s competitive with the very best proprietary models available but it’s released under an open license (or will be soon) meaning anyone — or any organisation — with enough compute can run their own version. While it’s apparently not the best all-rounder it currently sits at the top of the Arena WebDev leaderboard, above even Claude 5 Fable but at potentially a fraction of the cost. In the face of this, and stronger competition from GPT 5.6 Sol, Anthropic is already cutting Fable’s rates.
On the surface it seems counterintuitive that a company can train a first-class multi-trillion parameter foundation model and give it away for free. Even when charged for, the cost is significantly lower than Anthropic and OpenAI’s flagship models. But there’s a degree of geopolitics at play here. Chinese President Xi Jinping gave a speech in which he directly promoted China’s open source model approach as a challenger to the US’s closed-model hegemony.
SpaceXAI’s Grok added Automations that can run jobs from a schedule or trigger. x.com/grok
Also, Grok 4.5 is now available in Europe. But please think twice before using Grok.
Work
Manus added Auto-Publish, an optional feature that when toggled on automatically deploys successful builds to a live public URL, removing the manual steps required to update sites during ‘vibe-coding’ iteration. manus.im
Manus can now generate PowerPoint slides natively (rather than converting), with editable charts, structured tables, and more. manus.im
Search and Research
Google Search introduced connected apps to AI Mode, letting users securely link and directly interact with services such as Instacart, Canva, and YouTube Music. It’s rolling out in the US now. blog.google
I’ve said it before and I’ll say it again: AI Mode is Gemini in all but name, and I think Google is going to get in trouble for this despite the benefits it offers to users.Google Images got a redesign for its 25th anniversary, featuring real-time personalised galleries in a Pinterest-like layout. It’s coming to desktop users in the US first. Google is also adding image generation to AI Overviews. blog.google
Google’s NotebookLM is now Gemini Notebook, and added a secure cloud computer in each notebook to write and execute code for tasks such as data analysis. Available today for Ultra subscribers, coming soon for Pro. blog.google
Social
Threads users can choose Flairs when making posts to a Community, letting them show their specific tastes. threads.com/@threads
It started as a Twitter clone, and is now a kind of mid-point between X and Reddit.Meta AI will alert parents if their teen discusses suicide or self-harm. The feature is live now for parents using Instagram parental supervision in the UK, US, Australia and Canada. about.fb.com
Gaming / Metaverse-ish
Roblox announced Build, which lets users create games from a prompt on their phones, that they can share with friends, export to Studio for development, or publish to the Store. about.roblox.com
Fortnite creators can publish islands with LLM-powered NPC conversations, including 36 Fortnite IP characters available with locked (non-editable) voices and personas. fortnite.com