GPT Image 2.6 Native GIFs: What's New and What Day-One Users Say

On November 4, 2026, OpenAI rolled out ChatGPT Images 2.6 across ChatGPT and Codex, alongside two new API model variants: gpt-image-2.6-flare and gpt-image-2.6-sunburst. While the release includes faster image generation, reusable reference images, and SVG export, the headline addition is native animated GIF generation.

Instead of stitching frames together through Python scripts in Code Interpreter or exporting static assets to a separate video model, users and developers can now generate short, looping animations directly from a text prompt or uploaded photo. Within the first 24 hours of release, independent benchmarks and user tests on X showed a split reception: fast, charming results for stylized loops and cartoon gags, alongside visible loop seams, low-frame-rate choppiness in photorealistic scenes, and occasional refusals to animate at all.

Here is a breakdown of what OpenAI shipped in GPT Image 2.6, how the ChatGPT and API workflows differ, how the Flare and Sunburst models split the workload, and what early testers actually experienced on day one.

What Is New in ChatGPT Images 2.6

In its official announcement and X launch post on November 4, OpenAI summarized the update with a straightforward pitch: “Images, upgraded—introducing ChatGPT Images 2.6. Now with GIFs and reference images.”

According to OpenAI, the release arrives as image generation volume across ChatGPT, Codex, and the API reaches roughly 4 billion images per week—up from the roughly 3 billion per week figure OpenAI shared in September, as noted by The Verge.

Key capabilities announced by OpenAI include:

• Native GIF generation: Users in ChatGPT can create animated GIFs from text prompts or uploaded photos, adjust playback speed, inspect and edit individual frames, and share the resulting loops. Developers can generate GIFs directly via the Images API.
• 50% faster generation: OpenAI claims image generation speeds are roughly 50% faster compared to the previous version.
• Saved reference images: ChatGPT users can save character, product, or person reference images and reuse them across subsequent generations without re-uploading the files in every turn.
• Two API model tiers (Flare and Sunburst): The API splits GPT Image 2.6 into gpt-image-2.6-flare for low-latency layout work and gpt-image-2.6-sunburst for higher detail and multi-turn consistency.
• SVG export and markup-guided editing: The November 4 API changelog adds SVG output support, a new complexity parameter, and image markup capabilities for guided edits.
• Broad availability: OpenAI states that ChatGPT Images 2.6 is rolling out globally across all ChatGPT and Codex subscription tiers on web, desktop, and mobile, with Flare and Sunburst live in the API.

OpenAI also published a dedicated system card section covering GIF risks, safeguards, and evaluations, detailing how safety filters evaluate multi-frame motion and animated outputs.

How GIF Generation Works in ChatGPT vs. the API

Although both consumer ChatGPT and the developer platform run on the 2.6 family, the controls and endpoint rules differ depending on where you build.

Inside ChatGPT

In the ChatGPT interface (web, desktop, and mobile), GIF creation works through natural language prompts or by attaching a photo and asking ChatGPT to animate it. Once a loop renders, the UI provides built-in playback and editing controls:

1. Playback speed adjustment: Users can speed up or slow down the frame transition timing after generation.
2. Frame-by-frame inspection and editing: Users can step through individual frames to spot visual glitches or request targeted edits to specific frames.
3. Reference image reuse: Saved character or product references can be pulled into new prompts so the subject remains recognizable across static images and animated loops.

Via the OpenAI Images API

For developers, GIF output is handled through the POST /v1/images/generations endpoint by setting "output_format": "gif". According to OpenAI’s Image Generation Guide, the API exposes specific parameters to control frame count and timing:

• output_format: Set to "gif" (SVG is also newly supported on the 2.6 models).
• frame_count: Controls the number of frames in the loop. Note that OpenAI’s official guide lists a range of 2 to 64, whereas some third-party references and API documentation pages list 2 to 16—developers should treat current live API validation as the source of truth when configuring frame counts. The default is 9 frames.
• frame_duration_ms: Sets the display duration per frame between 20 and 2000 milliseconds (rounded to roughly 10ms increments). The default is 200 ms per frame.
• complexity: A newly introduced parameter in the November 4 changelog for tuning generation behavior.

OpenAI provides the following example payload in its developer documentation for generating a looping GIF with gpt-image-2.6-flare:

{"model":"gpt-image-2.6-flare","prompt":"A cat dances in a seamless loop","output_format":"gif","frame_count":9,"frame_duration_ms":200}

At the default settings (9 frames at 200 ms per frame), a standard API call yields a 1.8-second animation running at 5 frames per second.

GPT-Image-2.6 Flare vs. Sunburst: Choosing a Model

Rather than shipping a single monolithic endpoint, OpenAI divided the API release into two distinct model IDs: gpt-image-2.6-flare and gpt-image-2.6-sunburst. Teams evaluating how to choose an AI model for visual pipelines will need to match these variants to their latency and fidelity requirements.

According to OpenAI’s documentation:

• gpt-image-2.6-flare: Built for speed and layout structure. OpenAI claims that at low quality settings, Flare can generate outputs in as little as 6 seconds. It is positioned for rapid iteration, posters, product graphics, and quick animations where layout clarity and turnaround time matter most.
• gpt-image-2.6-sunburst: Tuned for medium-to-high quality settings, fine visual detail, and cross-turn editing consistency. When a workflow requires maintaining intricate textures or preserving character details across multiple editing passes, Sunburst is the designated option.

When planning production budgets or comparing endpoints in an AI API pricing comparison, note that multi-frame GIF generation processes multiple frames per request. Developers should consult OpenAI’s official pricing documentation for exact per-image and per-token rates across Flare and Sunburst quality tiers. (In an independent hands-on test, Kingy.ai self-reported a cost of roughly $0.056 and a generation time of about 17.6 seconds for a 10-frame, 512×512 GIF via the API.)

Day-One User Reactions and Real GIF Examples

Within hours of the November 4 launch, developers, AI researchers, and tech publications put ChatGPT Images 2.6 through day-one testing. The consensus from early adopters is that native GIFs are genuinely fun and faster to produce than old workarounds, but the output behaves much more like low-frame-rate flipbook animation than smooth AI video.

Early Community Demos on X

Several creators shared immediate successes with stylized loops, sprite-like animations, and classic AI benchmark prompts. For example, X user @SSHCodes tested the classic “pelican riding a bicycle” prompt as an animated loop:

GIF example by @SSHCodes on X (Nov 4, 2026). Source: https://x.com/SSHCodes/status/2118127375481504100 — used with attribution for news illustration; copyright belongs to the original author.

Other users experimented with character motion and creative prompts. @plasm0 shared an early animated sample demonstrating the stylized loop aesthetic:

GIF example by @plasm0 on X. Source: https://x.com/plasm0/status/2118177268870562241 — used with attribution for news illustration; copyright belongs to the original author.

Meanwhile, creators quickly began documenting prompt workflows. @Krouxzz posted a how-to guide accompanied by a generated GIF example:

GIF example by @Krouxzz on X. Source: https://x.com/Krouxzz/status/2118283796181053637 — used with attribution for news illustration; copyright belongs to the original author.

Additional positive takes highlighted convenience and humor:

• Low-frame-rate charm: X user @xinzhang322 shared multiple GIF samples and noted that the feature was “surprisingly good,” adding that the slightly choppy, low-frame-rate feel actually made comedic prompts funnier.
• Fewer iterations than legacy ChatGPT workflows: @Awe_automation posted a self-test GIF of a weekend skipping motion and noted that getting a usable loop took significantly fewer prompt iterations than older ChatGPT GIF methods.
• Speed improvements: @cellcogai reported running 80 generations comparing 2.6 against 2.5, claiming roughly 2× faster generation speeds and showcasing a 12-frame loop generated purely from text. Similarly, @VIBEQUIRKLABS observed faster generation speeds, while noting that adjustable moderation intensity settings felt largely unchanged in practice.
• Ecosystem integrations and workflows: Platform provider @fal shared a high-engagement launch demo video as the models went live, and @higgsfield_ai demonstrated a sprite-sheet-to-GIF workflow featuring a pencil-sketch fight scene.

Media Benchmarks and Hands-On Reviews

Independent tech publications ran structured tests on both the browser interface and the API, uncovering specific strengths and weaknesses:

• Analytics Vidhya (Browser Test): In its hands-on review, Analytics Vidhya generated a true 12-frame, ~2.16-second looping GIF of a teapot releasing heart-shaped steam. While the loop worked as advertised, the review noted that text rendering inside images can still drop or miss characters.
• Kingy.ai (API Test): Testing 2.6 against 2.5 in a hands-on comparison, Kingy.ai generated a 512×512, ~2-second loop (10 frames at 200ms). They reported that while text remained visible, the loop seam between the final and first frame was noticeable, and motion pacing across steps was uneven.
• Zhidongxi / 智东西 (Multi-Scenario Test): Chinese tech outlet Zhidongxi reported generation times around 10 to 15 seconds for animated outputs. Their testing found that claymation-style prompts maintained strong subject consistency, whereas high-fluidity photorealistic human scenes suffered from noticeable stutter (appearing closer to 1–2 fps), background shifting, and watermark drift. Animating old photographs performed particularly poorly, though increasing reasoning/thinking effort improved the physical plausibility of movements.

Negative Takes and Criticisms

Not every day-one user was won over. Because GPT Image 2.6 generates discrete frames rather than continuous high-frame-rate video, several testers pushed back against the hype:

• Blunt rejection of the animation quality: X user @BlackthorneAI tested the GIF animation feature, attached a sample output, and delivered a blunt verdict: “I fucking hate it.”
• Architectural skepticism: @Ian_Perf stated they were unimpressed by the update, arguing that 2.6 shares the same underlying architecture as 2.5 and retains the same fundamental limitations.
• Inconsistent GIF triggering in ChatGPT: Researcher Varun Chopra (@VarunChopr17597) reported that out of 17 prompts intended to produce animations, only 12 returned actual GIFs while 5 still output static images.
• Blur and stutter complaints: Even software vendor @HitPawofficial noted in a promotional post that ChatGPT’s native GIFs can sometimes look blurry or choppy (though the post served as a marketing pitch for their own enhancement software).

Current Technical Limitations to Know

Before wiring gpt-image-2.6-flare or gpt-image-2.6-sunburst into an automated pipeline—or relying on ChatGPT for client deliverables—teams should account for several documented restrictions in OpenAI’s Image Generation Guide:

1. No GIF output on /v1/images/edits: The API image editing endpoint does not support "output_format": "gif". If you want to animate an existing static image, you must either route the workflow through the /v1/images/generations endpoint or use the conversational ChatGPT interface.
2. Unsupported API features in GIF mode: When "output_format" is set to "gif", the API does not support streaming, partial images, custom output_compression, or prompt_rewriter.
3. Documentation discrepancy on frame_count: As noted earlier, OpenAI’s official guide states frame_count accepts values from 2 to 64, while some third-party and API reference pages cite 2 to 16. Developers should validate against the live API schema rather than hardcoding assumptions at the upper bound.
4. Motion and text artifacts: Based on third-party reviews from Analytics Vidhya, Kingy.ai, and Zhidongxi, users should expect visible loop seams on non-cyclical actions, occasional dropped letters in text prompts, and jerky motion if attempting photorealistic human movement rather than stylized or cartoon loops.

Frequently Asked Questions (FAQ)

Can ChatGPT make animated GIFs natively now?

Yes. With the November 4, 2026 release of ChatGPT Images 2.6, all ChatGPT and Codex subscription tiers globally (across web, desktop, and mobile) can generate animated GIFs directly from text prompts or uploaded photos, with controls to adjust playback speed and view or edit individual frames.

What is the difference between GPT-Image-2.6 Flare and Sunburst?

According to OpenAI, gpt-image-2.6-flare is optimized for speed and layout (such as posters and product visuals), capable of generating low-quality outputs in as little as 6 seconds. gpt-image-2.6-sunburst is designed for medium and high quality settings where fine detail and multi-turn editing consistency are the priority.

Can I export a GIF using the /v1/images/edits API endpoint?

No. According to OpenAI’s API documentation, the /v1/images/edits endpoint does not support GIF output. To generate a GIF via the API, you must use POST /v1/images/generations with "output_format": "gif".

How many frames can a GPT Image 2.6 GIF have in the API?

By default, the API generates 9 frames at 200 ms per frame. For the maximum cap, OpenAI’s official image generation guide lists a frame_count range of 2 to 64, while some API reference pages and third-party sources list 2 to 16. Developers should rely on live API validation for the exact enforced cap.

Why did my ChatGPT prompt return a static image instead of a GIF?

Early user testing—such as Varun Chopra’s day-one test where 5 out of 17 prompts returned static images—shows that ChatGPT does not always trigger GIF mode automatically. Explicitly requesting an “animated GIF” or “seamless looping GIF” in your prompt helps ensure the model routes the request to GIF output.

Sources

• OpenAI Official Announcement (Nov 4, 2026): Introducing ChatGPT Images 2.6
• OpenAI on X (Nov 4, 2026): @OpenAI launch post
• OpenAI Developer Documentation: Image Generation Guide
• OpenAI Deployment Safety: ChatGPT Images 2.6 System Card — GIF Risks, Safeguards, and Evaluations
• The Verge (Emma Roth, Nov 4, 2026): OpenAI brings animated GIFs to ChatGPT with Images 2.6
• Analytics Vidhya (Nov 2026): GPT Image 2.6 Review
• Kingy.ai (Nov 2026): ChatGPT GIFs: Images 2.6 vs 2.5 Hands-On Test
• Zhidongxi / 智东西 (Nov 2026): GPT Image 2.6 Hands-On Test
• Community Demos & Day-One Reactions on X: @SSHCodes, @plasm0, @Krouxzz, @VarunChopr17597, @xinzhang322, @Awe_automation, @BlackthorneAI, @Ian_Perf, @VIBEQUIRKLABS, @cellcogai, @HitPawofficial, @higgsfield_ai, @fal