ChatGPT Images 2.6 Is Here! GIF Generation and 50% Speed Boost
機械翻訳 / Machine-translated
A
AI Friends
3日前・17分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
Have you ever wished you could make the images you create in ChatGPT move? In Images 2.6, announced by OpenAI on November 4, 2026, GIF animation generation has been added. This article walks you through the new features, the API, differences from competitors, and use cases.
On November 4 (local time), US-based OpenAI announced a new version of its image-generation AI: "ChatGPT Images 2.6." According to ITmedia NEWS, it is the successor to "Images 2.5," which was released in September.
This update has three main pillars: generation speed improved by 50%, the ability to create GIF animations, and the ability to save reference images.
That said, the 50% figure is OpenAI's own claim. No details have been provided about what it was compared to or under what conditions it was measured.
If you ask ChatGPT to "make a GIF of a dancing cat," it returns an animation made up of several slightly different images strung together — like a flipbook.
You can also upload a hand-drawn sketch and have it transformed into a waving animation. The Decoder reports that after creation, you can adjust the playback speed and edit the animation frame by frame.
GIF animations fall apart if a character looks like a completely different person from frame to frame. The Decoder explains that Images 2.6 has improved image consistency, allowing characters to maintain the same appearance across frames. The view is that this approach was difficult with previous models.
Anticipated use cases include company mascots, product animations, and reaction images for social media. Since the GIF feature is "rolling out gradually," it may not be immediately available in your account.
You can now name and save photos, then call them up in other chats. For example, you save a photo of your dog under the name "Pepper." From then on, you simply ask, "Draw a watercolor of @Pepper."
This eliminates the hassle of re-uploading a photo every time. It's a convenient feature for anyone who repeatedly uses photos of people, pets, or characters.
Editing precision has also improved. Background changes now preserve fine details in clothing, and character, product, and brand characteristics are said to be easier to keep consistent across multiple rounds of editing.
Imagine the owner of a general goods store swapping out product photos with seasonal backgrounds. If the product's shape or color changed every time, it would be unusable. The improvement in consistency is helpful for tasks like these.
For developers, two models are available: "GPT-Image-2.6 Flare" and "GPT-Image-2.6 Sunburst." Both names were first introduced with Images 2.5.
Flare prioritizes speed and can generate an image in "as little as 6 seconds" at low-quality settings. Layout quality for posters and product images has also improved. Sunburst is designed for medium-to-high quality, emphasizing more detailed images and consistency during editing.
Additional features include the ability to annotate specific areas of an image to direct edits, SVG export, and a "complexity" setting that determines how much the AI "thinks" before generating.
According to Kingy AI's explainer, GIFs are output at 2–64 frames, and SVG produces one static image per request. The "as little as 6 seconds" figure is also OpenAI's claim under optimal conditions and does not represent typical response times.
It is available across all ChatGPT and Codex plans. The free plan is included, and it can be used on web, desktop, and mobile.
API pricing remains unchanged from Images 2.5. Rates per 1 million tokens (the unit by which the AI processes text and images) are as follows:
Kingy AI estimates a combined cost of approximately $0.095, based on assumptions of 1,000 tokens of text input and 3,000 tokens of image output. Actual costs will vary depending on the number of retries and input images.
Kingy AI's comparison lists the following services as competitors in image generation:
Images 2.6's distinguishing features are SVG and GIF output, as well as annotation-based editing. Kingy AI notes that this combination is difficult to find among similar competitors. However, output quality cannot be judged without actually trying it.
In preview scores from the evaluation site Arena, Sunburst ranked 2nd–3rd and Flare ranked 4th in the text-to-image category. In multi-image editing, the difference between Flare and Nano Banana 2.1 was small, making a clear ranking difficult.
Since conditions differ between providers, price alone cannot be used to compare quality. The most reliable approach is to test with the same images you use in your own work and compare the cost per usable image.
The service is available to all ChatGPT plans worldwide. Japanese users on the free plan are expected to be able to use it, but none of the sources reviewed for this article provided details about Japan-specific terms of service or Japanese-language support.
The following are some use case examples. On Japanese social media, LINE stamp-style reaction images and animated icons of "yuru-chara" mascots are popular. Individual creators could save their hand-drawn characters and generate GIFs with different poses each time.
Small and medium-sized businesses could save their company mascot and create GIFs for seasonal greetings or sale announcements. It's also suited for creating rough drafts before commissioning a designer.
According to AI Weekly, OpenAI states that image generation across ChatGPT, Codex, and the API has surpassed 4 billion images per week. That's up from 3 billion in September, and image generation is becoming an increasingly everyday tool.
Safety measures have been added to GIFs to detect harmful motion that is harder to identify in static images.
Additionally, as a way to verify that an image was AI-generated, Google DeepMind's watermarking technology "SynthID" will be introduced to ChatGPT, Codex, and the API, in addition to the existing C2PA metadata (information that records the origin of an image).
A. Yes, it is included in ChatGPT's free plan. However, since the GIF feature is rolling out gradually, the timing of availability will vary by account.
A. OpenAI has announced a 50% improvement. The API's Flare model can generate images in as little as 6 seconds at low-quality settings. The comparison conditions have not been disclosed.
A. Yes, you can adjust playback speed and share GIFs. They can be used as reaction images on social media, among other things.
A. No, it remains the same as Images 2.5. Per 1 million tokens: $8 for image input, $30 for image output, and $5 for text input.
A. In addition to C2PA metadata, SynthID watermarking will be introduced. This makes it easier to verify the provenance of images.
Start by saving a character or a favorite photo of yours, and try creating a short GIF.
This article is a cross-post from AI Friends.
