Connect with us

NEWS

ChatGPT Adds GIFs After OpenAI Killed Sora Video

OpenAI put looping GIFs in ChatGPT after shutting Sora, using Images 2.6 stills rather than video, as weekly image volume hit 4 billion.

Published

on

OpenAI added a native GIF generator to ChatGPT on November 4, 2026, through its new Images 2.6 model. A text prompt or an uploaded photo can become a looping animation, with speed and frame edits in the same chat window.

People now create 4 billion images per week across ChatGPT, Codex, and the API, OpenAI said, months after the company shut Sora, its standalone video app and API, in March 2026. Images 2.6 sells that motion as GIFs, and both new API models still list video as unsupported.

ChatGPT Ships GIFs on a Model That Bars Video

Images 2.6 is live in ChatGPT and Codex on web, desktop, and mobile, across every subscription tier, OpenAI said on November 4. The consumer pitch is simple: describe a loop, or upload a still, and get an animation you can share without leaving the thread.

The same announcement is careful about what that motion is for. Marketers can animate a mascot. People can make reactions and memes. Teams can “use animated storyboards to explore character movement before video production.” That last use is the tell. OpenAI is selling a loop as a sketch for a film tool it no longer offers in the product.

GIFs are the native language of social media, and GPT-Image-2.6 lets every OpenArt creator speak it fluently from a single prompt.

Coco Mao, CEO and co-founder, OpenArt

OpenAI posted the launch the same day, with GIFs and reusable reference images as the two consumer hooks and a pair of API models behind them.

GPT-Image-2.6 Flare is the fast lane. GPT-Image-2.6 Sunburst is the detail lane. Flare, the speed model, lists video as not supported. Sunburst uses the same line. The motion ChatGPT just gained is still an image job.

The Sora Shutdown Left a Hole for Loops

On March 24, 2026, OpenAI said it was discontinuing Sora in the consumer app and the API. A spokesperson said the Sora research team would keep working on world simulation to help robots with physical tasks, as compute demand grew and the company narrowed what it ships.

Sora had been the splashy bet: a standalone app, a developer API, and a public fight over likeness and copyright. Shutting it removed the only OpenAI product that treated moving pictures as the point. ChatGPT kept stills. The weekly stills kept climbing.

THE SORA-TO-GIF CALENDAR

  1. March 24, 2026: OpenAI discontinues the Sora consumer app and the Sora API, and points the research team at robotics simulation.
  2. September 8, 2026: ChatGPT Images 2.5 ships with the Flare and Sunburst split, and OpenAI says people already create more than 3 billion images a week.
  3. November 4, 2026: Images 2.6 adds native GIFs, reusable reference images, SVG export in the API, and a weekly count of 4 billion images.

The gap between those dates is the product decision. Full video was expensive enough to kill as a public service. A looping GIF, assembled from related stills, fits the machine that already prints billions of pictures. OpenAI even demoed the idea as a sprite sheet: six frames in a 3-by-2 grid, same character, same camera, then sliced into a loop.

How a ChatGPT GIF Is Built

It is not a clip with a soundtrack and a timeline. In ChatGPT you ask for a loop, or you hand the model a photo or a rough drawing and ask it to move. After it renders, you can change playback speed, step through frames, and patch a single cell without opening another app.

Developers get the same idea through GIF output on the generate endpoint. Set output_format to gif, describe the subject and the loop, and decode the b64_json payload as a .gif file. The image edit endpoint does not return GIFs at all. If frame 6 is wrong in an API job, the whole loop has to be generated again.

GIF CONTROLS IN THE API

  • Frame count: Integers from 2 to 16, default 9, only when the output format is gif.
  • Frame duration: 20 to 2,000 milliseconds per frame, default 200, rounded to the nearest 10 milliseconds.
  • Default loop: Nine frames at 200 milliseconds, which is a 1.8-second cycle, the length of OpenAI’s lantern demo.
  • Other formats: png, jpeg, webp, gif, and svg, with GIF and SVG limited to models that support them.
  • Hard limits: GIF output does not support streaming, partial images, masks, or custom compression.

That default 1.8-second loop is long enough for a reaction and short enough to post. It is also the length you get when an image model draws a handful of related stills and staples them together. Complex physics, crowded scenes, and two characters interacting are where the seams show, because there is no continuous video model underneath.

A 4 Billion Image Week and Two API Models

The volume is the reason a modest loop matters more than a dead video app. OpenAI said people now create 4 billion images a week across ChatGPT, Codex, and the API, more than half a billion a day. On September 8 it had put the figure at more than 3 billion images a week.

THE VOLUME SPIKE

  • Weekly count: 4 billion images across ChatGPT, Codex, and the API, as of the November 4 announcement.
  • Daily average: More than half a billion images a day, OpenAI’s own phrasing on the same page.
  • September baseline: More than 3 billion a week when Images 2.5 launched on September 8, 2026.
  • Speed claim: Images 2.6 generates images 50% faster than before, with Flare as low as six seconds at low quality.

Those stills were already leaking into other apps whenever someone wanted a loop. Native GIFs keep the next step inside ChatGPT. Reference images tighten that hold: save a face, a pet, or a mascot once, then call it back instead of re-uploading the file for every prompt.

The API keeps the split OpenAI introduced with Images 2.5. Flare is for fast iteration and cleaner layouts on posters and product shots. Sunburst is for tighter sameness across edits when quality is medium or high. List prices did not change. Text input is $5.00 per 1 million tokens, cached text $1.25, image input $8.00, cached image $2.00, and image output $30.00. Flare and Sunburst bill the same rates. You pick them on time and control, not on the price tag.

A new complexity setting sits beside quality. Quality is how finished the pixels look. Complexity is how much planning the model does before it draws. OpenAI’s advice for most edits is low complexity plus whatever quality the job needs, with medium or high complexity reserved for crowded scenes.

Frame Controls Stay Inside the Chat

The consumer win is not a new file type. It is that the file no longer has to leave. You can slow a loop, open it frame by frame, and rewrite one cell the way you already rewrite a paragraph in ChatGPT. Share sits on the same screen.

That is also why the API gap stings. ChatGPT can repair frame 6. The generate call can mint a fresh GIF. The edit call cannot output gif, so the fussiest part of the new workflow is locked to the chat product. Anyone building a stamp factory or a mascot pipeline still has to treat a bad frame as a full reroll, or bounce the file into ChatGPT by hand.

On November 5, some ChatGPT sessions still returned a wait message when asked for a GIF, even though OpenAI had said Images 2.6 was available on all tiers the day before. The still-image speed-up can land before the animation toggle does. Early loops also slip on tiny features. Brows fuse with eyelids. Extra hairs appear. The character holds, the micro-drawing does not.

The work this will actually eat is the work people already do with stills: reaction faces, brand mascots, chat stickers, a product shot that wiggles. Town, a startup building a work assistant, used the API to test an animated character, Monty, as a typing indicator. Picsart, named as an early access customer on the OpenAI page, said on November 5 that GPT Images 2.6 was live across its tools. Figma Weave’s creative director, Itay Schiff, said the company is adding the model for production work inside a reusable workflow.

Flare, Sunburst, and a GIF the Editor Cannot Touch

For anyone wiring this into software, the consumer GIF and the API GIF are cousins, not copies. ChatGPT hides the frame count. The API makes you set it. ChatGPT will edit a frame. The API will not give you a GIF back from the edit route. Both Flare and Sunburst can emit gif and svg. Neither accepts video as a modality.

FLARE VERSUS SUNBURST

Model Job Speed claim GIF Video Image output price
GPT-Image-2.6 Flare Fast iteration, posters, product shots As little as six seconds at low quality Generate endpoint Not supported $30 per 1 million tokens
GPT-Image-2.6 Sunburst Detail and sameness across edits Slower; aimed at medium and high quality Generate endpoint Not supported $30 per 1 million tokens
ChatGPT Images 2.6 Chat, Codex, all tiers 50% faster than the prior image model Create, speed, per-frame edit Not offered Included in ChatGPT plans

Markup, which already existed in ChatGPT, now comes to the API, so a team can circle a region instead of describing it in prose. SVG export is there for icons, print, and anything that has to stay sharp when it scales. Batch pricing for 2.6 was not on the public table at launch, which matters if you still run bulk stills through 2.5 at a discount.

Testers on OpenAI’s developer forum also said low and medium quality jobs now return more output tokens than they did on the 2.5 models. The sticker price per million tokens is unchanged. The bill can still rise at the settings people use for rough drafts, which is exactly where Flare’s six-second pitch will get used.

Saved Faces and the Meme Loop

Reference images are the quiet half of the launch. Keep a photo of yourself, a pet, or a character on file, then reuse it without uploading again. Combined with a 2-to-16-frame GIF, that is a stamp kit: one saved mascot, endless loops, no sprite sheet drawn by hand. The ChatGPT brand account made the intended use obvious on November 5.

We have a gif for you, the account wrote, then told people to make it weird and send more GIFs of their dog dancing. That is the product. Not a short film. A dog, looping, in the app that already makes 4 billion stills a week.

OpenAI still talks about storyboards that come before video production. The model cards still refuse video. Sora is still gone. What shipped on November 4 is the motion that fits the still-image factory, with frame tools in the chat window and a generate-only GIF in the API. For a lot of the people already in that factory, a 1.8-second loop is the whole job.

Harry is the editor of IAQABA, an independent publication he owns and runs. A decade in journalism, beginning as a reporter and now as the editor of his own titles, has left him with a clear test for what deserves a story: it has to change what a reader knows or decides, and it has to rest on something he can point to. That rules out recycled press releases, forecasts with no data behind them and rumours that no document supports. It leaves room for a great deal, and the site covers news, business, science and technology alongside sports, entertainment and lifestyle, with travel, auto and gaming given the same standard rather than lighter treatment. Sources are primary wherever possible: the regulator's filing, the company's own statement, the transcript, the dataset, or the product on Harry's desk. Figures are checked before they are published and rechecked if a reader questions them. Mistakes are corrected under a published policy. Readers across the world can reach him directly at support@iaqaba.com.

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Trending