OpenAI Images 2.6 adds animated GIF generation

• Animated GIFs now start from prompts or pictures, according to ChatGPT’s announcement, which also introduces reusable image references.
• Images 2.6 reaches ChatGPT and Codex across all subscription tiers on web, desktop and mobile, per the official rollout post.
• Developers get Flare and Sunburst models, separate planning and quality controls, markup-guided editing and SVG export in OpenAI’s linked announcement.

OpenAI’s launch examples specify a six-frame sprite sheet arranged in a 3-by-2 grid for a looping animation. Its API guide warns that increasing planning effort can introduce unwanted changes during editing.

OpenAI’s November 4 release includes turning simple sketches into animated characters. The system card confirms GIF generation in both ChatGPT and the API.

An OpenArt example published in the announcement gives its animation prompt three continuity constraints:

• Identical panel dimensions, with frames read left-to-right, then top-to-bottom.
• A fixed camera and consistent character, outfit, desk and background.
• Incremental movement, with the sixth frame transitioning smoothly toward the first.

People, pets and characters can be kept ready for reuse in new creations without uploading their reference pictures again, according to ChatGPT’s announcement.

OpenAI puts the generation speedup at 50% and offers two API variants:

• gpt-image-2.6-flare: Optimized for fast iteration. OpenAI reports generation in as little as six seconds at low quality, plus improved layouts for posters and product imagery.
• gpt-image-2.6-sunburst: Targets detailed visual work, with higher-quality generation and stronger consistency across edits at medium or high quality.

Both models default to high complexity, an easy-to-miss latency gotcha in the API guide.

The two controls affect different parts of generation:

• complexity: Sets planning effort before rendering. Supported values are low, medium and high; adjusting it carries no separate fee.
• quality: Controls rendering quality, visual detail and fidelity independently of planning effort.

OpenAI recommends low complexity for most edits, paired with the required image quality. Its medium and high complexity settings target scenes with many elements or constraints.

Both models support additional editing and output features through the Image API:

• Markup-guided edits: Region-specific change requests, bringing image markup previously available in ChatGPT to the API.
• SVG output: Scalable exports for websites, apps and print, as described in OpenAI’s announcement.

The guide’s existing code examples still use GPT Image 2.5, despite its new section documenting 2.6 support.

The rollout is global, according to OpenAI’s launch post. Flare and Sunburst are both available in the API.

Flowith said its own Images 2.6 integration was “coming soon” in its announcement.

OpenAI documents motion-specific safeguards and offline safety results in its system card:

• Motion moderation: GIFs can convey sexualized movement or harmful meaning across frames that individual still images do not capture. GIF-specific rules block some content categories more broadly than still-image rules, and OpenAI acknowledges that motion violations remain difficult to detect.
• Still-image safety evaluation: On 3,077 adversarial requests, Sunburst’s unsafe, unblocked output rate fell from 1.92% to 1.72%; Flare’s rose from 1.49% to 2.14%, a 0.65 percentage-point increase. Neither overall difference was statistically significant, and these offline results do not measure production delivery rates.
• Provenance: OpenAI pairs C2PA metadata with Google DeepMind’s invisible SynthID watermarking across ChatGPT, Codex and the API.