A hand-drawn giraffe can become a character that moves its head, winks, and smiles: that’s what Marketing4eCommerce got when it asked ChatGPT to animate its drawing. The Images 2.6 update adds the ability to create and edit animated GIFs from a text prompt or an image, right in the conversation.
The feature is integrated into ChatGPT and Codex, and is also coming to the API. Instead of generating a still image and taking it to another app to add movement, users can request an animation directly. Digital Trends reports that OpenAI says image generation is 50% faster than with the previous model.
Images, upgraded—introducing ChatGPT Images 2.6.Now with GIFs and reference images.Plus, two new models are coming to the API with 50% faster image generation.
— OpenAI
From a prompt or photo to a GIF
Images 2.6 accepts both a written description and a starting image. The Decoder notes that the rollout reaches ChatGPT and Codex users on the web, desktop, and mobile apps. Marketing4eCommerce also reports that it has been available globally since launch and across all subscription tiers.
The example shown by OpenAI starts with a simple request: “Can you make a gif of a cat dancing?”. The Decoder describes the result as several small variations of a character linked together like a flipbook. The visual continuity between those images allows the cat to maintain its appearance throughout the movement.
You can also upload a photo or drawing and ask for it to be animated. In the test published by Marketing4eCommerce, all it took was attaching a drawing of a giraffe and writing: “please turn this drawing into an animation”. ChatGPT described the result as follows: “Your giraffe has come to life! 🦒🎀 I animated it while preserving the original drawing style: it gently moves its head, winks, and smiles, while wearing its pink bow.”
The feature can be used to animate characters, brand mascots, product photos, memes, and illustrations. Digital Trends also points to animated storyboards as a way to plan character movement before producing a video. According to the outlet, a simple sketch can be enough to get started; you don’t need a finished design.
Adjustable speed and frames
The GIF doesn’t necessarily have to be final as soon as it’s generated. The Decoder and SoftZone report that ChatGPT lets users change playback speed with text prompts and review individual frames. This makes it possible to inspect the sequence and adjust the pacing or a detail without redoing the entire animation.
SoftZone explains that before this integration, creating a GIF with ChatGPT usually involved several attempts or using an external app. With Images 2.6, the prompt and, if provided, the reference image are processed in the same exchange. Frame editing and speed adjustments also take place within the chat.
There are limits that OpenAI has yet to clarify: SoftZone reports that the company has not specified the maximum duration or number of frames. It has also not clarified whether it will expand output to other animated formats or apply specific restrictions to free accounts.
Flare and Sunburst come to the API
The update includes two variants for developers: Flare and Sunburst. Marketing4eCommerce describes Flare as the option geared toward speed and rapid iterations; Sunburst prioritizes detail, quality, and consistency, especially when editing images. Both variants are available through the API.
Digital Trends specifies that Flare can generate an image in as little as six seconds at the low quality setting. For Sunburst, it highlights sharpness and visual continuity at medium or high quality. The outlet also links the improved composition to posters and product images, while Marketing4eCommerce notes that the model distributes elements more effectively in advertising materials and marketing creatives.
OpenAI is adding image markup to the API, allowing users to indicate which area they want to modify, as well as SVG export. Marketing4eCommerce describes the latter format as an option for using vector designs on websites, in apps, and in print materials. For localized editing, the advantage is being able to change one part without redoing the entire image.
The model also includes a complexity control, with low, medium, and high levels, to regulate how much it plans before generating. OpenAI’s documentation explains that a low level can be suitable for specific changes, such as modifying a background or a label, while a high level can help with dense compositions, text, and relationships between objects. It also warns that greater complexity can increase latency and cause unwanted changes. The documentation says that if no level is specified, Flare and Sunburst default to high, and that this setting has no separate additional cost.
Marketing4eCommerce names PicsArt, Higgsfield, Figma, fal, and Sable among the first companies to get access to the tool. OpenAI’s documentation confirms that the models support animated GIF output, markup, and SVG through the Image API.
Four billion images every week
The launch comes as the use of OpenAI’s image generation grows. The company puts the number of images generated through ChatGPT and the API at 4 billion per week. SoftZone and The Decoder compare this volume with the 3 billion OpenAI reported in September, an increase of 1 billion. Digital Trends also describes it as more than 500 million a day.
SoftZone cautions that OpenAI has not said whether this growth occurred before or after GIFs were introduced, so the figure does not make it possible to attribute the increase to the new feature. The company does present faster generation as part of Images 2.6: the model generates images 50% faster than its predecessor, according to OpenAI and the outlets reporting the announcement.
GIFs thus join still images in the same creation and editing workflow. For now, the available information leaves two practical questions unanswered for those who want to use the feature: how long animations can be and whether different account types will have different limits.
Sources: Marketing4eCommerce, SoftZone, The Decoder, OpenAI Developers, Digital Trends
