Google released Gemini 4 Argon on 13 October. It is not an image model. In Google’s own demos it takes one Nano Banana frame and builds a Blender scene around it, camera included — the slowest part of a 3D product film. The API costs half until 30 November, then doubles. The other two stories are about where your work lives: on one phone, or on a platform that can go under.
Gemini 4 Argon: from one image to a Blender set
Google made Gemini 4 Argon available on 13 October through the Gemini API and AI Studio (model gemini-4-argon), to Google AI Ultra subscribers in its Antigravity agent platform, and to Pro and Ultra subscribers in the Gemini app and AI Mode in Search (Google, 13 Oct; Gemini app, 13 Oct). The price is $2 per million input tokens and $10 per million output until 30 November, then $4 and $20, with a 1M-token input-output limit. Google says the gains are in spatial reasoning and “visual coding”. Its demos: starting from a single Nano Banana image, Argon drives Blender through MCP in an autonomous loop to build a Gothic scene; a pagoda complex refined by “multi-agent visual critique loops”; and photos of a real desk turned into a playable racing game, with textures from Nano Banana and engine sound from Gemini Omni Flash. Google also says it trained the model to ask for human confirmation in agent work (Google safety post, 13 Oct).
My read: I built Festival Toffee exactly this way, by hand. The sets and characters were painted with Kling from my sheets and placed in a Blender scene, each painting where a projector camera sees it, so it never distorts while the packs, fruit and ribbons stand in real 3D in front. My first version, modelled entirely in code as low-poly, looked too simple and I rebuilt it. Lining up those cameras was the slow, unglamorous part — and it is what Google is demonstrating. Two things the demos do not show: whether the loop keeps a pack’s artwork untouched (in that film the AI never redraws the packs), and what a full scene costs in tokens. Price it on one real shot before you promise a client anything, and budget at the December rate.
Google Play now runs your private vibe-coded app
Also on 13 October, Google Play started letting people build an Android app in Google AI Studio and run it on their own device as a private app, outside the public catalogue and with no developer account (Google Play, 13 Oct). It is rolling out in select markets, including the U.S., for users aged 18 and over. Sharing with friends and family is promised “in the months ahead”; going public still means a developer account and Play’s review (TechCrunch, 13 Oct).
My read: Private is the point and the limit. For an in-house team it is a real phone app for one person: a shelf-check tool for the field team, a shot-list app for a shoot day. It is not a campaign. The café games I built for Buzzout work because anyone at a table can scan a QR code and play in the browser — no install, no store. Use the Play route to put a working tool in a manager’s hand on Monday. Keep consumer-facing builds on the web, where the link is yours.
Character.AI files for Chapter 11; Disney buys assets
Character.AI filed for Chapter 11 bankruptcy on 13 October and chose a court-supervised sale, expected to close in late November or early December, TechCrunch reports (TechCrunch, 13 Oct). The service keeps running during the sale. Its CEO said that if no bid keeps it going, “the platform will be wound down.” Disney will buy some assets for $20 million and gets a non-exclusive perpetual licence to the technology. The report does not say what happens to user-made characters.
My read: Brands that built a talking mascot inside someone else’s app should read that sentence twice. For Popel I designed two knitted characters — a grey cardigan and beanie for him, a pink scarf for her — and the pack is the only thing that is not yarn. That design lives in our files, not in any tool’s account. If your character exists only as a profile on a platform, export the prompts, the voice settings and every approved image this week. A character you cannot move is rented.
What I’d do today
If you make 3D product films, give Argon one approved key frame and one pack, and ask for a Blender scene with a locked camera. Check the pack, the cost and the hours saved. Then list every brand character that lives only inside a vendor’s app and pull its files out. Want the pipeline built? This is how I work.
Sources
- Google — Gemini 4 Argon: Built to solve your hardest problems (API availability, $2/$10 introductory, $4/$20 after 30 Nov 2026, demo credits), 13 Oct 2026 — blog.google
- Google — Gemini 4 Argon comes to the Gemini app (Google AI Pro and Ultra), 13 Oct 2026 — blog.google
- Google — Developing Gemini 4 Argon responsibly and safely (human confirmation in agent work, 0.7% indirect prompt-injection success), 13 Oct 2026 — blog.google
- Google Play — More ways to create, discover and play (private apps from Google AI Studio, select markets incl. U.S., 18+), 13 Oct 2026 — blog.google
- TechCrunch — Google Play now lets anyone build private, vibe-coded apps, 13 Oct 2026 — techcrunch.com
- TechCrunch — Character.AI files for bankruptcy (Chapter 11, court-supervised sale, Disney $20M asset deal), 13 Oct 2026 — techcrunch.com