Nothing new shipped for brand teams in the last day. Something more useful did come up: two of the models most brand pipelines call are leaving their APIs within ten days. Google lists 22 October as the earliest shutdown for the Veo 3.1 previews. OpenAI turns off gpt-image-1 on 23 October. Voice models follow in November and January. If your agency, plugin or in-house tool calls one of these IDs, the job breaks or quietly switches models. Here is what goes, what replaces it, and the check to run this week.
Video: the Veo 3.1 previews go on 22 October
Google’s Gemini API deprecation table lists veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview and gemini-omni-flash-preview for shutdown on 22 October 2026, with gemini-omni-1.1-flash as the replacement (Gemini API deprecations, checked 13 Oct). Google says these are the earliest possible dates. The replacement is itself marked Preview on the models page: fast video generation, editing, keyframe interpolation and extension with native audio (Gemini API models, checked 13 Oct).
My read: A preview replacing a preview means you can be here again next quarter. We have seen this film: the Sora video API went on 24 September and I wrote then that a shot list should never depend on one model ID. For a running campaign, the risk is not the outage. It is the new model giving you a different grade, different motion and different faces on shot 14 of 20. Re-render two approved shots on the replacement now, while the old ones still exist to compare against.
Images: gpt-image-1 ends 23 October; Nano Banana’s first model has a date too
OpenAI removes gpt-image-1 from the API on 23 October 2026 and points to gpt-image-2.5-sunburst or gpt-image-2.5-flare; gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest follow on 1 December (OpenAI deprecations, checked 13 Oct). Both replacements bill image output at $30 per million tokens, or $15 with Batch, and OpenAI’s own calculator does not estimate their token use yet (gpt-image-2.5-flare model page). On Google’s side, gemini-2.5-flash-image, the model behind the original Nano Banana, is listed for 15 March 2027, replaced by gemini-3.1-flash-lite-image (Gemini API deprecations).
My read: My Christmas greeting film ran on four tools: ChatGPT for the storyline, Nano Banana for the frames, Kling for each sequence, CapCut for the edit. A seasonal film is the kind of brief that comes back next year, and by then the frame model under that name may not be the same one. What survives a model change is not the prompt. It is the reference sheet: approved frames, colours, the product angle. Keep those in your own folder, not in the tool’s history. And because the new price is per token, run 20 real frames in a test batch and divide. That is your cost per image, not the price list.
Voice: Gemini TTS previews on 17 November, OpenAI tts-1 on 6 January
Google lists gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-tts and gemini-3.1-flash-tts-preview for 17 November 2026, replaced by gemini-3.8-flash-tts or its Lite version (Gemini API deprecations); the replacement is stable and supports regional dialects across 130 languages (Gemini API models). OpenAI announced on 1 October that tts-1, tts-1-hd and both gpt-4o-mini-tts snapshots leave the API on 6 January 2027, replaced by gpt-realtime-2.1-mini (OpenAI deprecations, 1 Oct).
My read: A voice is a brand asset. If your product explainer, IVR or kids’ series uses a synthetic voice, that voice ends with its model ID. The new model will not sound the same with the same settings. Render your three most-used lines on the replacement now and let the brand manager choose by ear before the old one is gone.
What I’d do today
Send one email to whoever runs your AI production, agency or in-house: «Which model IDs does our pipeline call, and what is the plan for 22 October, 23 October and 17 November?» Then re-render two approved frames, one shot and one voice line on the replacements, and put old and new side by side for sign-off. If no one can answer the question, that is the real finding. This is the kind of pipeline I set up and run.
Sources
- Google — Gemini API deprecations: Veo 3.1 previews and gemini-omni-flash-preview, earliest shutdown 22 Oct 2026; Gemini TTS previews 17 Nov 2026; gemini-2.5-flash-image 15 Mar 2027 (page checked 13 Oct 2026) — ai.google.dev
- Google — Gemini API models: gemini-omni-1.1-flash (Preview), gemini-3.8-flash-tts (Stable, 130 languages), gemini-3.1-flash-lite-image (checked 13 Oct 2026) — ai.google.dev
- OpenAI — API deprecations: gpt-image-1 shutdown 23 Oct 2026 (announced 22 Apr 2026); gpt-image-1-mini / 1.5 / chatgpt-image-latest 1 Dec 2026 (announced 2 Jun 2026); TTS models 6 Jan 2027 (announced 1 Oct 2026) — developers.openai.com
- OpenAI — gpt-image-2.5-flare model page: image output $30 / 1M tokens, Batch $15 (snapshot 2026-09-08) — developers.openai.com; sunburst: model page