Images 2.6 speeds up GIFs and editing in ChatGPT
OpenAI is rolling out Images 2.6 in ChatGPT, with GIFs created from a sentence or a photo and faster generation announced. The launch is accompanied by increased usage and advertising now integrated into the display.
OpenAI began rolling out Images 2.6 in ChatGPT on November 4, on the web and in desktop and mobile apps. This version allows creating GIFs from a phrase or a photo, with generation announced to be 50% faster by the company. The topic matters as the tool is becoming central to mainstream use, while monetization through advertising appears right alongside the generated images.
GIFs from a phrase, then frame by frame
The logic is simple: the user describes a scene or sends a photo, and ChatGPT generates a looping animation. OpenAI cites the in-house example of the "dancing cat," but the mechanism also applies to more practical content, such as an illustration to animate or a sequence to retouch. Each image in the sequence can then be inspected and corrected individually. A GIF, but with more fine-tuning than a simple visual copy-paste.
Version 2.6 also adds reference images retained by the model, to reuse a face, a pet, or a recurring character without resending. OpenAI also announces image tagging to guide retouches, as well as export in SVG format. The rollout, however, spans several days: not all accounts see these options yet, and the company does not specify any particular limit for free accounts.
La cadence d’OpenAI s’accélère, et les chiffres suivent
Images 2.6 follows Images 2.0 in April and Images 2.5 on September 8. In seven months, OpenAI has thus released three major versions of its image generator. At the same time, claimed production has risen from 3 to over 4 billion images per week, including API usage, an increase of one billion in less than two months according to the company.
This growth says two things at once: usage volume is rising, and the company seeks to maintain control over the quality of the experience. The feature may seem anecdotal at first glance, but it touches a sensitive point for generative assistants: the faster they produce, the harder they become to distinguish from a traditional creation tool. And the more they attract electricity costs, which is rarely the preferred marketing angle.
La publicité se glisse à côté des images générées
Since October, OpenAI has also been deploying visual ads in the United States displayed alongside generated images, provided by an initial test group of advertisers and presented as clearly labeled. The move extends an advertising offensive launched earlier this year, then expanded to our region at the end of summer. For free accounts, the message is now quite clear: refusing ads without paying will soon mean losing image generation.
The Plus subscription, sold for 20 euros per month, remains the official way out. In this context, Images 2.6 is not just a technical update. It is also a retention lever, a conversion tool, and another piece in ChatGPT’s monetization strategy.
Canvas chez Google montre une autre voie, plus orientée prototype
Meanwhile, Google published on November 5, 2026, a demonstration of Gemini in Canvas, where a photo of a paper sketch becomes an interactive application. Canvas, launched on March 18, 2025, as an interactive space for documents and code, is used to generate and preview HTML and React prototypes. The demonstration shows a very direct sequence: image import, code generation, testing in a preview window, then adjustments by conversation.
The desktop documentation for Canvas specifies that the user can directly modify the code, consult the console, review recent changes, and request targeted retouches with Select & ask. On mobile, Preview and Show console controls are also available, and Android allows adding a shortcut on the home screen. Google also reserves access with Gemini 3 and a one-million-token context window for heavier projects to certain Google AI Pro and Google AI Ultra subscribers.
Both announcements head in the same direction: reducing the time between an idea and a first usable result. At OpenAI, this goes through images and GIFs; at Google, through application prototypes. In both cases, the real question is no longer whether AI can produce content. It already can. The issue now is who retains control over edits, access rights, and usage behind the shared link.
Key points
- OpenAI has been rolling out Images 2.6 since November 4.
- Generation is announced to be 50% faster.
- Over 4 billion images per week according to OpenAI.
- Google Canvas appeared on March 18, 2025.
- Gemini 3 is reserved for certain Google AI Pro and Ultra subscribers.
- OpenAI charges 20 euros per month for Plus.
By the numbers
- 4 billion images per week — volume claimed by OpenAI, including API.
- 50% — speed gain announced for Images 2.6.
- 20 euros per month — price of the Plus subscription cited by OpenAI.
- 1 million tokens — context window announced for Gemini 3 in Canvas.
- March 18, 2025 — launch date of Canvas by Google.