This week showed both ends of AI advertising. CNN counted more than 8,000 Google ads from one supplement brand, most of them AI cartoon musicals, some of them openly racist. Prada went the other way and drew its holiday campaign by hand. A food brand’s AI work has to sit visibly closer to Prada. Meanwhile, two models that a lot of brand assets were made with reached their shutdown dates.
Your AI ad now runs next to AI slop
CNN reports that the supplement brand Resilia “has launched 8,000-plus ads on Google in the last year and a half, 7,000 of which appear on YouTube, often between Shorts”, with about 480 more on Meta. They are AI-made animated musicals, and some use racist and antisemitic stereotypes. Google said: “We removed the ads flagged to us for policy violations” (CNN, 28 Oct).
What it changes: This is the inventory your AI spot shares a feed with. Viewers are learning what cheap AI looks like, and they file anything that resembles it in the same drawer. So the craft bar is now a brand-safety question, not a taste question. Ask your media team which placements and exclusions they use for Shorts and Reels, and judge every AI take against that feed, not against the agency deck.
Prada drew its holiday campaign by hand
“Drawing Prada: Animated by Hand” mixes live action with hand-drawn animation by veteran animator John Pomeroy, with Sadie Sink, Maya Hawke, Chen Haoyu, Kelvin Harrison Jr. and Luther Ford. “I learned my craft with pencil and paper,” Pomeroy says in Prada’s behind-the-scenes film (Marketing-Interactive, 28 Oct; Communicate Online, 1 Nov).
What it changes: Prada does not mention AI in what I read; the craft is the message. I make AI films, so my answer is not “no AI”. It is to decide which parts carry the brand and keep those real. On Festival Toffee the packs keep their original artwork in every shot, and only the world around them is invented.
The model behind last spring’s assets may be gone
Google lists the Veo 3.1 preview models in the Gemini API with a shutdown date of 22 October, replaced by Gemini Omni Flash at “approximately $0.10 per second” for 720p video, with no free tier (Google deprecations; Google pricing). OpenAI shut down gpt-image-1 on 23 October (OpenAI deprecations), and Krea’s old ChatGPT Image route now returns “410 Gone” (Krea, 24 Oct).
What it changes: If your Ramadan re-cut or regional version depends on regenerating last spring’s look, check first whether that model still exists. Treat an approved AI frame as a master file, not a recipe you can rerun. Write the model and version into the delivery sheet, and ask for the master files, prompts and reference frames at hand-over.
A 15-second model with separate audio stems
Pika released H3 Flex on 29 October, its post-trained version of MiniMax H3. It makes clips “up to 15 seconds” with stereo sound, takes “up to 12 reference files”, and the API can return “five separate audio stems”. The weights are open on Hugging Face under the MiniMax H3 Community License, but they generate natively at 480p or 768p, the base components are about 210 GB, and the higher-resolution upsampler stays on Pika’s servers (Pika, 29 Oct; Hugging Face).
What it changes: Fifteen seconds is a full social cut-down, and separate stems mean a sound designer can rebalance the mix instead of living with it. “Open” is not “free to use” for a brand, though: read the community licence before any client work.
What I’d do next week
Add two lines to every AI brief: the model and version, and which shots are made only with the camera, the hand or the real pack file. Then ask for the master files of every approved AI asset. If you want each shot written as a decision rather than a wish, the Director mode in my video prompt tool puts fourteen of them on one sheet.
Sources
- CNN (via KESQ) — The new frontier for ads is racist cartoon AI musical microdrama, 28 Oct 2026 — kesq.com
- Marketing-Interactive — Prada trades pixels for pencils this holiday season, 28 Oct 2026 — www.marketing-interactive.com
- Communicate Online — Prada brings hand-drawn animation to its 2026 holiday campaign, 1 Nov 2026 — communicateonline.me
- Google AI for Developers — Gemini API deprecations (Veo 3.1 preview models, shutdown 22 Oct 2026), read 1 Nov 2026 — ai.google.dev
- Google AI for Developers — Gemini API pricing (Gemini Omni Flash video), read 1 Nov 2026 — ai.google.dev
- OpenAI API — Deprecations (gpt-image-1, shutdown 23 Oct 2026), read 1 Nov 2026 — developers.openai.com
- Krea changelog — ChatGPT Image retired after gpt-image-1 shutdown, 24 Oct 2026 — www.krea.ai
- Pika — Pika H3 Flex, 29 Oct 2026 — pika.art
- Hugging Face — Pika-Labs/Pika-H3-Flex model card, read 1 Nov 2026 — huggingface.co