Tech Landscape #423
One video model replaces another, a Nano Banana competitor, and Social Commerce is back!
Hello!
I can barely articulate how bored and disappointed I am by the discourse around AI. There’s so much hype and dis-/mis-information, with very little rationality on either side of the conversation. “RIP Hollywood!” is as uninformed and misleading as “using AI boils the oceans!”. Too many people have built an identity as cynic or zealot, when what we need is more skeptics.
Anyway, let’s get on with it. Hope you’re well!
Synthetic Audiovisual
Seedance 2.0 began its global rollout
It’s available through both CapCut and Dreamina in some countries, and CapCut only in others — including, as I write this, the UK.
capcut.com/newsroom, x.com/dreamina_ai
I wrote about Seedance 2.0 when it was announced six weeks or so ago [TL 417]. It’s very impressive, generating high-quality results that follow prompts exactly, but still suffers from some of the same issues as other video models. You can see in my initial tests ⬇️ that the detail, motion, and camera work are all excellent; but in the third clip it still doesn’t fully get cause-and-effect. That said, it’s still the best video model available. I’ve seen reports / complaints that many of its most blatant copyright- and likeness-infringing capabilities have been neutered since the initial launch, but of course they have.
Open AI is shutting down Sora
The app and API will be removed, although details of the timeline haven’t yet been shared.
This came somewhat out of the blue; new API features were announced last week, an article about its safe use was published the day before the shutdown, and Disney didn’t seem to have advance notice that its billion-dollar deal was off. There’s a lot of speculation about why it’s shutting down; the popular opinion is that it was incredibly expensive to run and very few people wanted to pay for it, and OpenAI wants to tighten its focus rather than trying to be all things to all people.
I think the positioning of the Sora app as a fun ‘slop’ social network was probably misguided, as it didn’t appeal to businesses and more intentional creators. Also it was expensive compared to rival models — over twice as expensive as Kling 3.0 on Freepik, for example. It was a very good model, though — remember how people freaked out about its effect on Hollywood? — and it’s a surprise that it’s gone now. But for those who think this somehow signifies the beginning of the end for AI video: I direct you to the story above this one.
Luma Labs launched Uni-1
It’s a unified image generation model that integrates reasoning to follow complex compositional instructions and spatial logic more accurately than most other image models.
It’s similar to Google’s Nano Banana in the way it uses reasoning, making it easy to prompt the exact outputs you want while staying consistent during editing. In my test ⬇️ I asked first for a sketch, then to render the sketch cinematically, then to change the icon on the superhero costume to be clearer; it did pretty much exactly what I asked for at every step.
Creative Tools
CapCut launched Video Studio, a “canvas-based AI production workspace” with agents for rapid creation. instagram.com/capcutapp
Ideal for using Seedance 2.0! This pattern of combining an infinite canvas with an agentic assistant is becoming really popular in creative tools.Freepik added Relight for images and videos, with controllable light sources, optionally from a preset or image reference. x.com/freepik
I’ve only had time for a quick test ⬇️ but it looks like it could be useful.
Freepik added 3D Scenes: turn any image into a 3D scene where you can move a virtual camera around to compose shots. instagram.com/freepik
Like OpenArt’s Worlds [TL 422] this is essential for generating consistent environments and a step towards the future of creator tools.Shopify launched Tinker, a free mobile app that gives access to AI tools for creating brand assets like logos, product photography, and videos. shopify.com
This includes image and video models from Google, Kling, and OpenAI (for now)… for free. I don’t know how the economics make sense to Shopify, but grab it while you can.
Music
Google released Lyria 3 Pro, the advanced version of its music generation model that can create tracks up to three minutes long with better understanding of musical composition for more creative control. blog.google
It’s available now in Gemini, Vids, Producer [TL 419] and other Google products. The extra ‘thinking’ power is better at writing lyrics and creating coherent song structure; you can hear it in my example ⬇️ (“an emo song about Lyria 3”). There are some useful tips for developers on how to incorporate it in apps.
Suno released its v5.5 model with more personalised features: Voices lets users capture their own voice and use it to generate new songs; Custom Models lets them tune the core model to their original style by uploading tracks; and My Taste combines stated (from a prompt) and implied (from listening history) preferences to create a taste profile that influences generations. suno.com
The Voice feature in particular is very impressive; this goes way beyond “prompt to song”. Don’t get too attached to this model, though; Suno says:
The capabilities we're putting in place today — voice fidelity, personalised sound, custom models — lay the foundation for the next generation of music models we’re launching with the music industry later this year.
Mureka released its v9 music model, with improvements to speed, prompt understanding, and mix and vocal quality. x.com/Mureka_AI
Social & Messaging
Instagram lets users reorder carousels after posting. A small but useful update. instagram.com/creators
WhatsApp added photo editing and suggested responses, powered by Meta AI, and several quality-of-life improvements including dual-account support for iOS and cross-platform chat transfers between iOS and Android. about.fb.com
Snap introduced AI Clips in Lens Studio, a tool for developers to add video generation capability to their Lenses for Lens+ subscribers. newsroom.snap.com
Reddit is adding anti-bot measures, including labelling “good” bots and requesting human verification from accounts showing “fishy behaviour”. reddit.com
Like it or not, age / identity verification is becoming a common feature on the internet; it’s even implemented in the latest version of iOS for UK users to prove their age when downloading certain apps.
Ads & Social Commerce
This week was the annual IAB NewFronts, where internet companies announce their latest advertising and commerce options. Among those shown:
More YouTube creators can join the Shopping affiliate program. Members of the YouTube Partner Program with at least 500 subscribers can earn commissions by tagging products in Shorts, long-form videos, and livestreams. blog.youtube
Facebook launched Affiliate Partnerships, letting eligible creators tag shoppable products directly in Reels and photo posts to earn commissions from partner brands. creators.facebook.com/blog
Meta AI is going to help customers make purchase decisions on Facebook and Instagram, and there’s a new one-click checkout option. facebook.com/business
Snapchat’s Total Snap Takeovers let advertisers show up as the first ad spot in each tab. forbusiness.snapchat.com
TikTok’s TopReach lets advertisers show up in the first content people see when they open the app and the first in-feed ad spot in the For You feed. ads.tiktok.com
TikTok made two additions to its Pulse Suite, which lets brands buy ad placement next to culturally relevant content and trends: Mentions places them next to the moments when people are actively talking about their brand and category; and Tastemakers places them immediately after videos from a hand-selected group of creators. ads.tiktok.com
Assistants & Search
Google released Gemini 3.1 Flash Live, a fast and high-quality audio/voice model for more natural real-time dialogue. blog.google
It’s already available in Gemini Live and in Search Live, which has expanded globally to over 200 countries and territories.
Gemini made it easier to import your memories and search history from other AI assistants. blog.google
ChatGPT improved its product shopping experience with richer and more visual results. openai.com
It’s powered by the expanded Agentic Commerce Protocol (ACP) — not to be confused with the Universal Commerce Protocol (UCP). [TL 413]
Genspark added Realtime Voice input to its Workspace tool. x.com/genspark_ai
Claude can now control your Mac as a research preview of computer use is available in the Cowork app. claude.com
It’s especially useful with Dispatch, the service that lets you run Cowork (and Code) from your desktop or mobile app; you can assign tasks on your morning commute, and have them ready to review when you get to the office. I still wouldn’t trust it yet, but YMMV.
MolmoWeb is a new open browser use agent that can autonomously navigate and interact with web interfaces. allenai.org
It’s free and small enough to be hosted locally, which makes it powerful and useful. I ran a couple of quick tests and it was fast and accurate, although only on sites which it has already been tested on before.