Industry News
Upon the release of GPT-Image 2.6, it was evident that significant improvements had been made in AI technology. This update focuses on the support for GIF files and the provision of reusable reference images.
On November 5th, it was noted that the GPT-Image 2.6 entry page had been added to the AI Navigation section and listed under the “Most Popular AI” category on the homepage. This model was released by OpenAI on November 4th, and the access link for this entry page directs users to the GPT Image 2.6 page on the PixPix platform.
OpenAI stated that Images 2.6 is now available to all subscription levels of ChatGPT and Codex, including the web version, desktop version, and mobile version. Developers can use APIs to access two models: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst.
GIFs and reference images constitute the main aspects of these updates.
The most notable feature of this update is the ability to generate GIFs. Users can use prompts or images to create short loop animations, and they can adjust the playback speed or modify individual frames within ChatGPT. By default, the API produces animations consisting of 9 frames, each lasting 200 milliseconds, resulting in a total loop duration of approximately 1.8 seconds.
ChatGPT has also added a feature that allows users to save reference images. Once pictures of people, pets, or characters have been saved, they can be directly reused in new creations without the need for re-uploading them.
For developers, the API now includes features for image annotation, SVG export, and the use of a “complexity” parameter. The annotation function is used to identify areas that need to be modified, while the complexity parameter controls the level of pre-planning undertaken by the model before rendering. In line with OpenAI’s design philosophy, Flare focuses on speed—resulting images can be generated in as fast as about 6 seconds for the lowest-quality setting; Sunburst, on the other hand, places greater emphasis on detail and consistency across multiple editing rounds.
Officials claim that the new model is approximately 50% faster than GPT-Image 2.5. The pricing of the API remains unchanged: $5 per million tokens for text input, $8 for image input, and $30 for image output.
Early testing results: The unit price remains unchanged, but the costs associated with certain price ranges have increased.
On the day of its release, several developers already shared the results of third-party tests. One set of tests involving icons with dimensions of 1024×1024 showed that for the low-quality tier, the cost of using Flare increased from $0.0061 to $0.0250; for the medium-quality tier, it rose from $0.0134 to $0.0437. For the high-quality tier and above, the costs remained roughly the same as those in the previous version. The reason for this difference is that the 2.6 version consumes more output tokens at lower quality settings.
In the same test, the execution times of various Flare versions were approximately half those of their predecessors, while Sunburst was 17% to 27% slower in certain tasks compared to its previous generation. These data sets were relatively small, and the testers also noted that the results obtained on the night of the release did not reflect long-term performance.
Consistency remains the main limitation in commercial applications. Another set of tests revealed that the structure of the images changed when used with different visuals, and the seamless transition between the beginning and end of a GIF file was also insufficient. Furthermore, SVG exports can only generate new graphics based on textual descriptions; it is not possible to convert existing logos into vector formats.
Third-party platforms and relevant case studies will be released subsequently.
PixPix currently offers two options: Flare and Sunburst. These tools focus on ensuring visual consistency during image generation, reference image editing, and multiple rounds of editing. They are suitable for use in scenarios such as portrait images, product visuals, posters, and brand branding. After users enter the required prompt words and log in, Sunburst is selected by default on the workspace interface. The specific amount of credits consumed will be displayed on the workspace. Currently, the GIF functionality is not listed on this page.
It has been reported that the image case platform under AI has also launched a special section dedicated to GPT Image 2.6, which includes examples and guidelines for creating portraits, e-commerce visuals, posters, character designs, and animated images. The recommendations suggest that when producing animations, it is advisable to break down each movement into individual frames, clearly indicate the parts that remain unchanged, and ensure that the last frame seamlessly transitions back to the first one. When creating a series of images, it is best to start by producing a clean base image and then explicitly specify the elements that must not be altered.
Illustration note: The images used in this article were sourced from the Discover AI Image Examples website and created using Codex’s built-in image generation tool. This tool does not display the actual names of the models used; the animated images are composed of frames generated by the models’ rendering capabilities.