Skip to main content
MindStudio
Pricing
BlogAbout
My Workspace
ChatGPT Images 2.6animated GIF AIOpenAI image model

ChatGPT Images 2.6: How to Make Animated GIFs (And Where It Breaks)

OpenAI's Images 2.6 model brings faster generation and animated GIF creation to ChatGPT and the API. Here's how it works and its face problem.

Edited by Luis Chavez-Mattos, Director of Product RSS
ChatGPT Images 2.6: How to Make Animated GIFs (And Where It Breaks)

What is ChatGPT Images 2.6?

Images 2.6 is OpenAI’s latest image generation model, available in both ChatGPT and the API. It generates static images noticeably faster than its predecessor, and its headline new trick is the ability to create animated GIFs directly from a text prompt. You can also edit those GIFs after the fact with follow-up instructions, and the model supports SVG output and markup-guided editing, where you draw or annotate directly on an image to steer a change.

TL;DR

  • Animated GIF generation is the main new capability in Images 2.6, letting you type a prompt like “make an animated GIF of a monkey roller skating” and get a finished loop in well under a minute.
  • Editing works on GIFs, not just stills, so you can ask for changes like adding a hat or moving the scene to a different setting and the model will regenerate the animation with those tweaks applied.
  • Face consistency is the model’s weak spot. Animating a photo that includes a real human face often produces a visibly different-looking person, because the model is generating new frames rather than literally moving pixels.
  • The workaround is selective animation. Telling the model to animate only a background object or element while leaving a person’s face static preserves likeness much better than animating the whole scene.
  • Generation speed has improved meaningfully. A full GIF, which likely involves multiple generated frames stitched together, can come back in roughly 30 seconds, and single static images have been clocked around 20 seconds.
  • The model also adds SVG support and markup-guided editing, both accessible through the API as well as the ChatGPT interface, which opens the door to more structured design and illustration workflows.
  • Volume matters here. OpenAI’s image tools reportedly process billions of image generations a week, which is part of why incremental model upgrades like this get rolled out broadly rather than as a niche feature.
MindStudioRemy
Live with CEO Dmitry Shapiro

Big news

for MindStudio builders.
Tue, Nov 1711:30am PTZoom · 1 hourLive Q&A
Hear it
first
Nov 17

How do you make an animated GIF in ChatGPT?

The process is a single prompt. You describe what you want animated, submit it, and ChatGPT returns a looping GIF, generally within 20 to 60 seconds depending on complexity. In testing shown by creators covering the release, a prompt asking for a monkey roller skating produced a finished animated GIF in about 20 seconds, and a more elaborate request (an email to an insurance company plus a matching urgent-looking GIF) took around 52 seconds to generate both outputs together.

Under the hood, the model is almost certainly generating a sequence of individual frames and stitching them into a loop, similar to how traditional GIF animation works. What’s notable is that this doesn’t come with an obvious speed penalty compared to generating a single static image. A single skyline image was generated in roughly 20 seconds in one test, and a full animated GIF came back in a similar window, suggesting the frame generation and compositing pipeline is reasonably efficient rather than doing many expensive full-resolution passes.

Can you edit a GIF after it’s created?

Yes. Once a GIF exists, clicking into it surfaces an edit option, and you can send follow-up instructions the same way you’d edit a static image. Examples demonstrated include adding an accessory (a baseball hat) and changing the setting (moving a roller-skating scene from a sidewalk to a street). Both edits were accepted and the model returned an updated GIF reflecting both changes in roughly the same generation time as the original.

This matters for practical use. It means you’re not stuck regenerating from scratch with a longer, more detailed prompt every time you want to adjust something. You can iterate the same way people already iterate on static ChatGPT images, just applied to a moving file.

Why does Images 2.6 distort faces in GIFs?

The limitation comes from how the model animates a scene. Rather than taking a source image and literally shifting pixels frame by frame (the way traditional animation or video interpolation tools work), the model generates new images for each frame to create the illusion of motion. For objects, backgrounds, and non-human elements, this is usually fine. For human faces, it’s a problem: each regenerated frame is a new AI interpretation of that face, and the model doesn’t reliably preserve identity across frames.

In practical testing, a photo of a specific person turned into a GIF came out looking like a noticeably different person once animated. The model is described as better than many other tools at holding likeness, but still not reliable enough to trust for anything where accuracy matters, like a recognizable headshot or branded photo of a real individual.

Is there a workaround for the face problem?

The enterprise platform for building software with AI

Build an onboarding app for 4,000 new hires.

Build ↵
Self-hosted
SOC 2
Role-based access
Audit log

Yes, and it’s straightforward: animate only part of the image and explicitly instruct the model to leave faces or people static. In one demonstrated example, a video thumbnail was turned into a GIF by asking the model to animate just a dinosaur graphic in the scene while keeping the person in the shot unchanged. The result kept the person looking like themselves, at the cost of lower resolution, while still achieving a sense of movement through the animated element.

This suggests a general rule for anyone using the feature: full-scene animation is fine for illustrations, objects, landscapes, and abstract content, but any image containing a real person’s face should be animated selectively if likeness needs to be preserved. For marketing images, thumbnails, or anything with a recognizable person, that constraint matters a lot more than it does for a cartoon monkey on roller skates.

What else does Images 2.6 support?

Beyond GIFs, the model adds SVG output, meaning it can produce vector-based images rather than only raster/pixel images. Vector output is useful for logos, icons, and anything that needs to scale cleanly without losing quality, since SVGs are defined mathematically rather than as a fixed grid of pixels.

The model also supports markup-guided editing, where a user can annotate or draw directly on an image and have the model interpret those marks as editing instructions. This isn’t entirely new as a concept in ChatGPT’s image tools, but Images 2.6 extends it through the API as well, which means developers can build this kind of annotation-driven editing into their own applications rather than relying only on the ChatGPT interface.

Is Images 2.6 worth using for GIFs specifically?

For non-face content, this looks genuinely useful rather than gimmicky. Static documents, cold outreach emails, educational slides, and presentations are often visually flat, and a short looping GIF is a low-effort way to add motion without producing a full video. Since the same prompt-and-edit workflow already familiar from ChatGPT’s image tools carries over to GIFs, there’s effectively no new learning curve.

For anything involving real people’s faces, the current limitation is real enough to plan around. If you need a GIF of a product, a mascot, an illustrated character, or a scene without a specific recognizable person, Images 2.6 handles it well and quickly. If you need to animate a photo of yourself or someone else and keep them recognizable, expect to either accept some likeness drift or restrict the animation to a background element and keep the face still.

Frequently Asked Questions

What is Images 2.6 in ChatGPT?

It’s OpenAI’s updated image generation model, available in ChatGPT and via the API, that adds faster generation, animated GIF creation, SVG output, and markup-guided editing on top of the previous image model’s capabilities.

How long does it take to generate a GIF in ChatGPT?

In demonstrated examples, GIFs were generated in roughly 20 to 60 seconds depending on complexity and whether multiple edits were requested at once.

Can I edit an animated GIF after ChatGPT creates it?

Yes. You can click into the generated GIF and send follow-up text instructions, such as adding an accessory or changing the setting, and the model will regenerate the GIF with those changes applied.

Why do faces look different in animated GIFs made with this model?

Because the model generates new frames to simulate motion rather than directly manipulating the original image’s pixels, and it doesn’t reliably preserve a specific person’s likeness across those newly generated frames.

Automation tools are the old way.Enterprises build with AI on Remy.
Zapier
n8n
Make
Airtable
RemyEnterprise-ready
  • Self-hosted
  • SOC 2
  • HIPAA
  • Audit log
  • Role-based access
Built into every app you make

How do you keep a face consistent in a ChatGPT GIF?

Instruct the model to animate only a specific element or part of the scene, such as a background object, and explicitly tell it not to animate the person or face, which keeps that portion of the image static and visually consistent.

Editorial standards

Presented by MindStudio

No spam. Unsubscribe anytime.