ChatGPT Images 2.6: GIFs, image editing and a practical creator checklist

Checked November 6, 2026. OpenAI announced Images 2.6 for ChatGPT, Codex and the API on November 4. The release adds GIF creation and changes to image generation and editing. This guide explains the announcement and provides a practical exercise; MSL has not independently measured its speed or tested the new features.

日本語版はこちら

What changed?

GIFs: OpenAI shows short looping animations, such as characters moving or reacting.
Editing: OpenAI says the model is better at preserving details of a person, product or scene across edits.
API: Flare is positioned for speed and Sunburst for detail and editing consistency. OpenAI also describes SVG output, markup-guided edits and configurable complexity.

OpenAI says image generation is 50% faster than before. That is the vendor’s claim, not an MSL benchmark: wait time depends on settings, demand and other conditions. Its roughly six-second Flare example is a low-quality, best-case example, not a typical-time guarantee.

ChatGPT versus the API

OpenAI announced availability across ChatGPT and Codex plans on web, desktop and mobile. Check your own interface for rollout and usage limits. Flare and Sunburst are API model names; do not assume that all ChatGPT users can select those names in the app.

In the API, quality and complexity can be adjusted separately. OpenAI suggests lower complexity for many edits and considering medium or high for more demanding compositions. SVG output is documented as an API feature. Review the official pricing and your current settings before use: cost varies with format and configuration.

Try it: a short announcement GIF

This is a fictional practice prompt written by MSL, not a demonstration of an actual output. Start with a simple, original character for which you hold the necessary rights.

Create a short looping GIF of an imaginary white cat mascot waving hello with one paw in the center of the frame. Use a plain background and a fixed camera. Make the first and last poses connect naturally. Do not add text or logos. Keep the composition simple enough for the expression to remain clear at small social-media sizes.

Watch the complete loop for changes in hands, limbs, face and background.
Compare the colors and markings with the reference and record what must stay unchanged.
When revising, separate the requested change (“brighten the background”) from what must be preserved (“keep the character’s shape”).
Test playback, dimensions, file size, quality and loading time on the actual destination and a mobile device.

Before uploading or publishing a person’s or product’s image, check permission, ownership and your organization’s rules. For an advertisement that must represent the real product accurately, compare the result with the product and approved copy. If the destination does not support GIFs, check its supported formats and prepare an appropriate alternative.

How should API users choose between Flare and Sunburst?

One starting point is to use the same source and prompt with both: consider Flare for quick concepts and Sunburst for detail-sensitive revisions. This is an editorial suggestion based on OpenAI’s descriptions, not an MSL comparison test. Before adopting either for work, record latency, cost, number of revisions, text accuracy and variation across repeated outputs.

Speed first: generate several rough concepts, then refine only the selected one.
Consistency first: provide a reference and check color, shape and composition one by one.
Do not use the output yet: if a product or person is misrepresented, rights are unclear, or corrections cost more time than they save.