How to Make Animated GIFs in ChatGPT (Images 2.6)

ChatGPT can now make GIFs. OpenAI released Images 2.6 on 4 November 2026, and the headline feature is motion: describe an animation, upload a photo, or hand it a rough sketch, and it returns a looping GIF instead of a still image. You can then change the playback speed, fix a single frame, and share the loop. This guide shows how to do that, which prompts work best, and what to do when your account does not have the feature yet — because GIF creation is still rolling out, and some accounts genuinely do not have it.

Quick answer: Images 2.6 turns a prompt, photo or sketch into a looping GIF in ChatGPT. How to make one, edit frames and speed, and what to do if your account lacks it.

What Images 2.6 actually adds

Images 2.6 is an image-generation update for ChatGPT, Codex and the API, and GIFs are its most visible new output. OpenAI’s announcement (4 November 2026) lists four things that matter for ordinary users: animated GIFs from a prompt or an image, saved reference images you can reuse across generations, 50% faster generation (OpenAI’s own claim, with no public benchmark behind it), and two new API models — GPT-Image-2.6 Flare and Sunburst. The release also adds SVG exports, image markup and a complexity setting for developers.

The reference-image feature matters most for GIFs. Save a character, product or mascot once and animate it again and again with the same look — the consistency problem that used to be the hardest part of the job. Your mascot can star in a new loop every week without drifting into a different-looking character.

How to make a GIF from a text prompt

The simplest route needs no image at all: type what should move, and say the word GIF. OpenAI’s own demo prompts are almost casually simple — “Make a looping GIF of a paper boat bobbing on the water” is the kind of request that works.

Open a new chat in ChatGPT on web, desktop or mobile. Images 2.6 is available globally across all subscription tiers, according to OpenAI — but it is still rolling out, so your account may not have it yet.
Describe one clear action, not a scene. A GIF is a few frames in a loop, so name a single movement: a boat bobbing, a cat blinking, a chart line falling. OpenAI’s own example prompts ask for a six-frame sprite sheet in a 3-by-2 grid, each frame showing one step of the motion, with the last frame leading smoothly back to the first.
Ask for a looping GIF explicitly. Write “Create a looping GIF of…” rather than “Create an image of…”. If ChatGPT returns a still image instead, your account most likely has not received the feature yet.
Check the loop. Watch the animation through at least twice. The most common failure is a jump between the last frame and the first — the motion should flow back to the start, not snap. If it snaps, ask ChatGPT to smooth the transition.
Adjust the speed. Ask ChatGPT to make the loop faster or slower; reported controls run from 0.5x to 2x. Slow, gentle motion (a boat bobbing, a plant swaying) reads better slow; a reaction or a joke usually lands better fast.
Download and share. OpenAI says GIFs can be shared on repeat — the format is built for messaging apps, social posts and reactions, not for long video. Keep it short: these are loops, not clips.

How to animate a photo or a sketch

You do not need a polished design to start — even a simple sketch can become an animated character. That is OpenAI’s own framing, and its announcement shows the proof: an owl drawn in pen on lined paper, turned into an expressive animated character.

Upload the photo or sketch into the chat, then describe the motion in words. A product shot can become an eye-catching loop; a hand-drawn character can wave, blink or walk.
Name exactly what should move — and what should stay still. “Animate only the owl’s eyes and beak; keep the paper background exactly as it is” works better than “make this move”, because the model needs to know which parts of the image are allowed to change.
Ask for a short loop, one action only. One movement per GIF is the rule that keeps results usable. Two movements in one prompt usually produce a muddled middle frame.
Save the image as a reference if you plan to reuse the character. The same owl, mascot or product can appear in next week’s GIF without being redrawn from scratch.

How to edit a GIF once it exists

A GIF from Images 2.6 is editable in three ways: speed, individual frames, and follow-up changes. You can ask ChatGPT to slow the loop down, redraw one frame that looks wrong (a hand in the wrong place, a face that drifted), or change one element while keeping the rest of the animation intact. Sunburst is the better choice for edits — OpenAI says it brings greater image consistency across edits at medium or high quality — while Flare is the faster option for first drafts.

One hard limit: ChatGPT cannot edit an animated GIF you upload. It only edits GIFs it made itself. Bring a GIF from elsewhere and ChatGPT treats it as a still image. The API has the same limit: the image-edits endpoint cannot return GIFs — GIF output only comes from the image-generation endpoint. Always keep the original prompt and reference image, because that is your edit path.

5 copy-paste prompts that work

The prompts that work best describe one action, six frames, and a clean return to the start. OpenAI’s own demo prompts follow that pattern — a six-frame sprite sheet, read left to right then top to bottom, identical framing in every panel, the sixth frame flowing back into the first. Swap the subject and keep the structure.

Reaction GIF: “Create a looping GIF of a paper boat bobbing on the water. Six frames in a 3-by-2 grid, same camera angle in every frame, gentle up-and-down motion, last frame flows smoothly back to the first. No text, no captions.”
Mascot loop: “Create a looping GIF of our brand mascot, a small orange fox, waving hello. Six frames, fixed front camera, same background in every frame, only the arm and ears move, last frame returns to the starting pose. No text, no logos.”
Product loop: “Create a looping GIF of a coffee cup with steam rising and swirling. Six frames in a 3-by-2 grid, identical cup and background in every frame, only the steam moves, seamless loop back to frame one. No text.”
Sketch come alive: “Turn this sketch into a looping GIF. Animate only the character’s eyes blinking and head tilting slightly. Keep the hand-drawn paper style and background exactly as they are. Six frames, smooth loop, no text.”
Before-and-after: “Create a looping GIF of a small plant growing: a seed sprouting, a stem rising, two leaves opening, then gently swaying. Six frames, same pot and background throughout, last frame flows back to the seed. No text, no numbers.”

Two details do real work. “No text” matters because image models still garble words, and a GIF with a misspelled caption is unusable. “Last frame flows back to the first” matters because a loop that snaps is the most common reason a GIF looks amateur.

Flare vs Sunburst: which model to pick

Flare is the fast draft; Sunburst is the careful finish. Both are GPT-Image-2.6 models in the API, and both cost the same — $8 per million image-input tokens and $30 per million image-output tokens, with $5 per million text-input tokens, unchanged from version 2.5.

Model	Best for	What OpenAI says
Flare	First drafts, quick loops, posters and product imagery	As little as 6 seconds per image at low quality; improves layout quality for posters and product visuals
Sunburst	Final GIFs, edits, anything where the character must stay identical	Higher-quality generation and greater consistency across edits at medium or high quality

In the ChatGPT app, the model is chosen for you. The table matters most for developers: the generation endpoint takes an output format of GIF, a frame count (2 to 64 in OpenAI’s image guide, though the API reference says 2 to 16 — stay within 16 to be safe) and a frame duration, defaulting to 9 frames at 200 milliseconds each.

When it won’t work — and what to do instead

The most common problem is not a bad prompt; it is an account that does not have GIF creation yet. OpenAI’s help page says GIF creation “is rolling out”, with no date for every account to have it. Independent checks on 5 November 2026 confirm the gap: one account check could not find dedicated GIF controls at all, and a Pro-account test found ChatGPT generating still images first, only producing a playable GIF after being asked where it was. If that happens to you:

Ask directly for a GIF, not an image. If the reply is a still image, reply “please make this an animated GIF” and see whether a download appears.
Try the sprite-sheet prompt. Asking for a six-frame sprite sheet in a 3-by-2 grid is OpenAI’s own documented pattern, and it is the route several early testers used successfully when a direct GIF request returned a still.
Try ChatGPT, not Codex. Images 2.6 is available in Codex too, but one test on 5 November asked Codex directly for a GIF and received a single-frame PNG after about 21 seconds — the Codex image tool exposes no GIF control.
Do not upload a GIF and ask ChatGPT to edit it. That path is closed by design: uploaded GIFs are treated as stills, and the API’s edits endpoint cannot return GIFs. Start from the original prompt and reference image instead.
Wait, and check again. There is no published per-plan GIF quota — OpenAI has not said how many GIFs any plan can make, or whether a GIF counts differently from a still image — so a missing feature is a rollout gap, not a limit you have hit.

The safety check OpenAI runs on GIFs

Animation can carry harm that a single frame does not, and OpenAI’s safety testing reflects that. The Images 2.6 system card includes a test with deliberately hard harmful GIFs — cases designed to slip past a checker that looks at frames one by one — and OpenAI reports its safeguards caught 77.7% of them. That number is OpenAI’s own, from its own test, and it is worth stating plainly: nearly a quarter of deliberately tricky harmful GIFs were not caught. Do not use the feature to animate real people without their consent, to fake an event, or to create misleading clips of identifiable people.

Our take: the three-loop test

Before you share any GIF ChatGPT makes, run it through three loops and one question. This is the original check we use for every first GIF, and it catches almost every failure before anyone else sees it:

The seam: watch the jump from the last frame back to the first. If you can see the snap, ask for a smoother transition. A good loop has no visible seam.
The drift: watch one fixed detail — the background, the character’s outfit, the product label — through all six frames. If it changes shape, colour or position, the reference image was not held tightly enough; regenerate with the reference saved, or switch to Sunburst for the edit.
The point: ask what the GIF actually says. If the motion does not add meaning a still image would not carry, the GIF is decoration, and a still would have been better.

Then the one question: would this still be true if the GIF were a still? The best GIFs in OpenAI’s own examples pass it — the chibi office worker watching a stock chart fall only works because the tears move; a static frame of the same scene would not land the joke. That is the test that separates a GIF worth sharing from motion for its own sake. One honest caveat: this guide is desk-researched, not hands-on tested — GIF creation is still rolling out and account-dependent, so the steps above reflect OpenAI’s documentation and verified third-party checks from 5 November 2026, not a test on our own account.

Frequently Asked Questions

Can ChatGPT make GIFs for free?

OpenAI says Images 2.6 is available globally in ChatGPT and Codex across all subscription tiers, including Free. But GIF creation is still rolling out, so a Free account may not have it yet, and OpenAI has published no per-plan GIF quota. If your account lacks it, that is a rollout gap, not a paywall.

How do I turn a photo into a GIF in ChatGPT?

Upload the photo, describe exactly what should move and what should stay still, and ask for a looping GIF — for example, “Animate only the steam rising from this coffee cup; keep everything else exactly as it is.” One action per GIF works best, and saving the image as a reference lets you reuse the same subject later.

Why does ChatGPT give me a still image instead of a GIF?

Your account most likely has not received GIF creation yet. OpenAI’s help page says the feature “is rolling out”, and independent checks on 5 November 2026 found accounts with no dedicated GIF controls. Try asking again explicitly for an animated GIF, try the six-frame sprite-sheet prompt, or try ChatGPT rather than Codex — and check again later.

Can ChatGPT edit a GIF I upload?

No. ChatGPT only edits GIFs it made itself; an uploaded animated GIF is treated as a still image. The API has the same limit: the image-edits endpoint cannot return GIFs. Keep your original prompt and reference image so you can regenerate and edit from those instead.

Sources and methodology

Claims in this article rest on OpenAI’s own announcement “Introducing Images 2.6” (4 November 2026), OpenAI’s Images in ChatGPT help page, and independent checks published 5 November 2026 by Digital Trends, Moneycontrol, Choosely and LaoZhang, plus the OpenAI Developer Community announcement thread. Benchmarks and speed figures are OpenAI’s own claims, labelled as such; the rollout status and account-level gaps come from the independent checks, not from OpenAI. All sources were verified on 6 November 2026. This guide was desk-researched, not hands-on tested, because GIF creation is still rolling out and account-dependent.

Arva Rangwala

Arva Rangwala covers AI news, AI tools, guides and prompts for OpenAIMaster — what is new, what is worth using, and how to put AI to work.