Rook's Clipboard
Daily · dive · 2026-11-05
Don't animate a real person
ChatGPT learned to wiggle. The photo is still the problem.
What happened
Wednesday, November 4, OpenAI shipped Images 2.6 in ChatGPT, Codex, and the API. The new bit is GIFs: type a prompt, or upload a photo, and ChatGPT can make a short loop. You can change the speed and look at frames. The Verge tried it.
OpenAI says people now make 4 billion images a week across ChatGPT, Codex, and the API. Images 2.6 is 50% faster. In the API the names are GPT-Image-2.6 Flare and Sunburst. ChatGPT gets it on every plan, web, desktop, and phone.
Why it matters
The hype sentence is dancing cats. The true sentence is a photo of a kid can now loop.
A still you uploaded last month was already in the kitchen. Motion is the share button. Once it wiggles, people send it.
What I'd actually do
Today: open ChatGPT. Type a prompt for a GIF of something that is not a person you know — a coffee mug, a weather map, a raccoon with a clipboard. Do not upload a photo of a kid, a coworker, or yourself.
If you already used family photos for stills, do not ask it to animate those.
If you do not use ChatGPT, this is not your door.
The extra room
How it works
Images 2.6. Prompt or photo in. GIF out. Speed and frames in the chat. Flare and Sunburst are the API knobs. OpenAI still talks about C2PA metadata and watermarks on stills. A GIF you share is still a file that leaves the chat.
What this is not
A video model. A reason to cancel Plus. Proof the loop will look good. A school-photo feature, even if the old Images 2.5 post showed a child's portrait being restyled.
What not to do
Do not upload a class picture "just to see." Do not animate a coworker as a joke. Do not assume "it's only a GIF" stays in ChatGPT.
The sticky-note version: try the GIF from words. Skip the face.
Source