From late September to mid-November I wrote a short AI brief three times a week. Most of it was news that expires in a fortnight. This page keeps the part that still matters for a brand brief, grouped by what it changes. Each line links to the original brief, where the dated sources are.
Read the bold line under each heading if you only have a minute.
Video models
Takes are longer now, 15 to 30 seconds, so the storyboard is the brief again. Check a model’s retirement date before you build a workflow on it.
- OpenAI shut down Sora’s video API on 24 September with no successor. Read the brief
- Kling 4.0 takes up to 10 keyframes and 30-second shots, so the storyboard becomes the brief. Read the brief
- Runway now works inside OpenAI’s dots: brief at night, a finished spot by morning. Read the brief
- Veo 3.1 previews and gpt-image-1 leave the APIs on 22–23 October, voice models follow. Which IDs break, what replaces them, and the check to run this week. Read the brief
- Kling 4.0 is live on the web with 30-second takes and 10 keyframes; API, HDR and price still missing. Read the brief
- Gemini Omni 1.2 Flash makes 15-second clips at $0.07 a second. What a usable take really costs, the free Vids limit, and Google voice remixing. Read the brief
Images and editing
Editing one boxed area and holding characters across images is now normal. Keep the real pack as a separate layer anyway.
- Adobe put Photoshop and Express editors inside ChatGPT: describe the edit, then finish it by hand. Read the brief
- Ideogram 4.5 keeps an image clean edit after edit, even translating text in place. Read the brief
- FLUX 3 Image edits one boxed region and leaves the rest alone, packshot included. Read the brief
- Google’s Nano Banana 2.1 costs half as much per 1K and 2K image and holds 4 characters. Read the brief
- Qwen’s Turbo draws 2K images in eight steps under a research-only licence, Kandinsky 6.0 gives open video with sound, and GPT-6 turns ChatGPT answers into UI. Read the brief
- Gemini 4 Argon drives Blender from one image, half price until 30 November. What it means for 3D pack shots, plus Play private apps and Character.AI. Read the brief
- Adobe MAX: outside agents can edit layered Photoshop and Premiere files, ChatGPT lets brands claim a profile, and Avid puts Veo inside Media Composer. Read the brief
New formats
Code-built motion, playable games and AI series are cheap to try. Getting anyone to watch them is the hard part.
- China’s AI short-drama minute fell from 5,000 yuan to a few hundred; under 0.5% find an audience. Read the brief
- Anthropic’s Claude Motion builds animations as editable code and hands them to Luma or Higgsfield. Read the brief
- Google’s Playground turns a prompt into a playable game, Nikon pulls a winner over AI post-processing, and DistroKid’s takedowns hit licensed music. Read the brief
- Svedka offers $500,000 for a fan-made Super Bowl spot after its AI ad. What the brief gets right, plus the AI licensing fight and shared avatars. Read the brief
Ad platforms
Platforms now read, rewrite and remix your ads by default, and the legal risk stays with you. Check the defaults before launch.
- Luma Variants turns one approved static ad into five placements and several languages, logo and CTA locked. Read the brief
- OpenAI will test labeled visual ads during ChatGPT image generation. Read the brief
- Higgsfield’s Marketing Studio builds ads from a product page, about $4.45 a spot by its own maths. Read the brief
- From 29 October ChatGPT Ads rewrite, translate and redirect your ads by default, with no preview. How to opt out, and what 1,000 shoppers say about AI ads. Read the brief
- Meta’s new ad terms make you solely responsible for what its AI changes in your ads, even features it turns on. Read the brief
- Apollo’s chief economist says AI has paid off nowhere outside big tech. The brand budget answer, plus Grok’s AI Odyssey and Walmart’s shopping agents. Read the brief
- Advertisers made 1.5 billion campaign assets with Gemini in one quarter. Why choosing now matters more than making, plus Genie 4 and ByteDance’s drama studio. Read the brief
- Meta’s new ranking model reads the text and images in every ad. Why the frame is now your targeting, plus Mosseri on AI content and Runway App Studio. Read the brief
Rules, rights and trust
Disclosure rules, public AI detectors and training opt-outs arrived in the same season. Decide your labelling policy before a client asks.
- OpenAI says 53 images users gave ChatGPT ended up on image hosts. Check one setting today. Read the brief
- Google opened its SynthID Detector to everyone: 180 billion marked images and videos. Read the brief
- Google and OpenAI will honour one shared robots.txt signal for AI training. What a brand site should open and close, plus AI-reworked art and iQIYI. Read the brief
- Saudi Arabia’s AI Principles in Media say who must disclose AI use and when. A checklist for Ramadan briefs: labels, voices, faces and the archive. Read the brief
These briefs are paused; news now goes into a weekly digest. If you want this applied to your own brand, send the problem.
