ChatGPT Images 2.6: what it is and how to make GIFs

ChatGPT Images 2.6 is OpenAI's new image model, and its headline trick is animated GIFs you make from a prompt or a still image. To try it, open a chat in ChatGPT, describe the motion you want, and ask for a looping GIF. It also lets you save reference images, and developers get two new API models, Flare and Sunburst.

The posts on X

@OpenAI announced it on November 4 with a short pitch: "Now with GIFs and reference images." A follow-up post from @OpenAI lists the GIF controls: "Adjust the speed, edit individual frames, and share it on repeat."

The @ChatGPT account explained the second feature. You can keep reference images of yourself, pets and characters ready for new creations, with "no need to upload them again."

This builds on Images 2.5, the model behind ChatGPT Try On.

How to make a GIF in ChatGPT

OpenAI hasn't published frame or length limits for ChatGPT GIFs, according to Fello AI, so keep your first attempts simple.

1. Start a new chat and, if you want, upload the still image to animate.
2. Describe one motion, and ask for a seamless loop.
3. Adjust the speed and fix any bad frames.

Here are three starter prompts of our own. We haven't run them, so treat them as starting points:

• "Animate this logo with a slow shine moving left to right. Make it a seamless loop."
• "A coffee cup with steam rising, flat illustration style, looping GIF."
• "Animate this product photo with a gentle 360 degree turn on a white background."

Kingy AI reports that ChatGPT can't edit an animated GIF you upload, and that saved reference photos get an @name you mention in later chats. The API doesn't have that saved-photo feature. Fello says OpenAI hasn't documented how many references you can keep or how to delete them.

As for who gets it, eesel says Images 2.6 is on every ChatGPT plan, including Free, though no per-plan GIF limits were published. MLQ says GIF creation is still rolling out, so you may not see it yet.

Flare vs Sunburst pricing

In the API, gpt-image-2.6-flare is built for fast iteration and gpt-image-2.6-sunburst for detailed visual work. The announcement post in OpenAI's developer forum says Flare can return an image in as little as six seconds and that Sunburst stays more consistent across edits.

Both cost the same on OpenAI's pricing page: $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens. Those rates match 2.5, but the bill may not. Here's what eesel reports from OpenAI's cost calculator for output at 1024x1024:

So low and medium quality cost roughly three to four times as much per image. One poster in that forum thread reported 344 output tokens on 2.5 and 1,136 on 2.6 at medium for the same prompt. If you run a high-volume, low-quality pipeline, re-estimate before you switch. The model pages also list no Batch support for 2.6.

GIF and SVG in the API

OpenAI's image guide has you set output_format to gif, with frame_count defaulting to 9 and frame_duration_ms to 200, which gives a 1.8-second loop. When we read the guide on November 6 it listed 2 to 16 frames and support on both generations and edits. Eesel and Kingy report a cap of 64 frames, and eesel says edits aren't supported, so test this before you build on it. Streaming, partial images and custom compression don't work with GIFs.

SVG output returns one static file per request, on the generations endpoint only. There's no fixed price, because you pay for a reference image plus the tokens used to write the SVG.

Our read

The GIF feature looks like the fun part, and the cost change is what developers should check first. The 50% speed claim has no published baseline, as Kingy points out. Quality reports are mixed too: Fello quotes Hacker News commenters who found results inconsistent, with one calling a chess animation too splotchy to use. We'd try simple subjects first.

OpenAI's system card also admits that motion-related violations are hard to detect and its safeguards may miss some. If you're publishing GIFs under a brand, a person should watch each one first, since a problem may only show up across frames. For a competing option, see Google's Nano Banana 2.1, or Canva Craft if your team already works in Canva.

What it means for businesses

Short animated assets just got cheaper to attempt. A loop for an email header or a product post no longer needs a designer for the first draft, though you'll still want one for anything that has to match your brand exactly. Saved references should also help keep a mascot or product consistent from one piece to the next.

You'll also need to decide who approves what goes out. New Face Design's free process audit can help you map where tools like this fit in your marketing work.