In two days, three ad platforms moved the ad closer to the machine: OpenAI will place image ads beside ChatGPT image generation, TikTok put AI and creator content in one approval queue, and Meta turns catalog photos into video. All three run on your product image. This week’s job is a master pack file and a rule sheet.
The ad platforms now make the variants
OpenAI will show visual ads alongside images people generate in ChatGPT, from later this month, in the US only, clearly labeled (OpenAI, 5 Oct; TechCrunch, 5 Oct). TikTok’s Smart+ Creative Selection puts brand assets, Symphony AI creative and creator content in one approval workflow (TikTok, 5 Oct). Meta made video from static assets generally available in Advantage+, and turns catalog photos into product video for beta advertisers (Meta, 6 Oct).
What it changes: On the range reel eight categories share one grid, one timing and one tile movement, so they read as one group. Feed a platform one styled hero and you get 200 versions of one shot. Feed it a clean, front-on, true-colour pack and a rule sheet — and check every end scene.
Edit tools lock the logo. Nobody locks the pack
Ideogram 4.5 promises images that stay clean “edit after edit”, by its own comparison (Ideogram, 30 Sep; model page). FLUX 3 Image edits inside bounding boxes, at $0.10 per 2K image and $0.607 at 4K (Black Forest Labs, 1 Oct). Luma Variants takes one approved static ad into five placements and other languages with the “logo, headline, and CTA” intact; its examples are English and Spanish (Luma, 1 Oct).
What it changes: Every tool this week locks something — the logo, the headline, the CTA, a box on a grid. None locks the pack, and for an FMCG brand that is the one lock that matters. On Sky Red Grape the purple and gold come from the grape and the bottle label. One shade off and the film no longer matches the shelf. Run five edits in a row, overlay the result at 100% and look only at the pack. Proof at 2K. 4K costs six times more.
Previz got cheap. Delivery did not
Nano Banana 2.1 reached Google Ads and Flow, and a 2K API image now costs $0.0504, half of Nano Banana 2’s $0.101 (Google DeepMind, 6 Oct; Gemini API pricing). Its own card admits “poor text rendering in small text”. HeyGen Video went live on 30 September at $0.01 a second through October, and its docs name the weak spots: long on-screen text, “petals, paper and hair”, close hand work (HeyGen; HeyGen docs).
What it changes: Small print, petals, hands. That is the pack and half of food advertising. On the Christmas greeting I made the images in Nano Banana and built each sequence from them in Kling. Drift at the image stage costs every shot after it. Use the cheap tools to block timing and camera. Composite the real label in post.
A watermark will not prove who made it
OpenAI will watermark eligible ChatGPT and Codex text in the EU. Swapping 25% of words for synonyms cut detection to 17%, and OpenAI says the mark does not “measure human contribution” (OpenAI, 5 Oct; The Next Web, 5 Oct). In a Tavus test, 26 of 54 people took its Griffin model for a real person after a one-minute video call; disclosure features are still in the works (Tavus, 1 Oct).
What it changes: Who made your film will be settled by a record, not a detector. The range reel credit lists the idea, layout and type, product images, motion, music and edit. Keep that list for every job. If a synthetic face ever talks to customers live, say so in the first second.
What I’d do next week
Make one master file per hero SKU: front-on, true colour, label readable at 100%. Under it, one page — what any tool may change, what it never touches, who signs off. Then test that pack in FLUX 3 or Luma Variants at 2K. For a second pair of eyes on it, this is how I scope it.
Sources
- OpenAI — Building advertising for the way people use AI, 5 Oct 2026 — openai.com
- TechCrunch — OpenAI launches visual ads that appear alongside image generation results, 5 Oct 2026 — techcrunch.com
- TikTok Newsroom — TikTok unveils AI-powered updates for advertisers, Advertising Week New York, 5 Oct 2026 — newsroom.tiktok.com
- Meta for Business — Advertising Week New York 2026: new AI capabilities, 6 Oct 2026 — www.facebook.com
- Ideogram — Introducing Ideogram 4.5 (post on X), 30 Sep 2026 — x.com
- Ideogram — Ideogram 4.5 model page, accessed 7 Oct 2026 — ideogram.ai
- Black Forest Labs — release notes: FLUX 3 Image, 1 Oct 2026 — docs.bfl.ai
- Luma AI — Introducing Luma Variants, 1 Oct 2026 — lumalabs.ai
- Google DeepMind — Nano Banana 2.1 model card, 6 Oct 2026 — deepmind.google
- Google — Gemini API pricing (Nano Banana 2.1 and Nano Banana 2), checked 7 Oct 2026 — ai.google.dev
- HeyGen — HeyGen Video launch catalog (live 30 Sep 2026), accessed 7 Oct 2026 — developers.heygen.com
- HeyGen — HeyGen Video model docs, accessed 7 Oct 2026 — developers.heygen.com
- OpenAI — Our approach to EU text provenance rules, 5 Oct 2026 — openai.com
- The Next Web — OpenAI starts watermarking ChatGPT and Codex text in the EU, 5 Oct 2026 — thenextweb.com
- Tavus — Griffin, 1 Oct 2026 — www.tavus.io