How to Make GIFs in ChatGPT With Images 2.6

What if the doodle on your desk could wave back at you? OpenAI launched ChatGPT Images 2.6 on November 4, 2026, with faster image creation, better consistency across edits, and the ability to turn a prompt, photo, or sketch into a looping GIF. GIF creation is still rolling out, and you can now save a reference image and reuse it in another chat by typing a name like @Pepper.

What Changed in ChatGPT Images 2.6

Say you ask ChatGPT to move a character from a city street to a beach. You want the background to change, and you want the character to show up in the same outfit. Images 2.6 is better at keeping details like that intact, and there are a few more changes you’ll notice when you create or edit an image.

• Faster image creation. OpenAI’s six-second figure applies to still images made with Flare. It gives you an idea of how quickly the model can work.
• More consistent edits. You can ask for a new background while keeping details like your character’s clothing consistent across versions.
• Looping GIFs. You can start with a written prompt or an image and ask ChatGPT to add motion. You can then change the playback speed or edit individual frames.
• Saved reference images. Give a photo a name once and call it up in another chat with an @mention.
• Photo markup. Draw over the area you want changed so ChatGPT knows where to apply your edit.

Images 2.0 arrived in April 2026, followed by Images 2.5 on September 8. Fast forward to November, and OpenAI puts weekly image creation across ChatGPT and the API at more than 4 billion, compared with 3 billion in September.

There are two models in the release. Flare is built for speed, while Sunburst is designed for higher image quality and better consistency when you edit the same image several times.

How to Make a GIF in ChatGPT

You can start with a photo you already have or describe an animation from scratch. Either way, think about what you want to move before you send the prompt. A cat sitting on a porch and a cat jumping off the porch give ChatGPT two very different jobs, so spell out the action as you go through these steps.

1. Open ChatGPT. Start a chat on the web or in the desktop or mobile app.
2. Describe the scene or upload an image. Use a photo or sketch if you already have a subject. If you’re starting with text, describe what you want to see and how you want it to move.
3. Ask for a looping GIF. Name the format and the action in the same request. If you uploaded a sketch, for example, ask ChatGPT to turn it into a GIF of the character waving.
4. Refine the movement. If your first GIF barely moves, ask for a bigger gesture or more action. If the whole scene moves when you only wanted the character to wave, ask ChatGPT to keep the camera still.
5. Adjust playback or edit a frame. Change the playback speed if the loop feels rushed. You can also edit a single frame if you spot something you want to fix at one point in the animation.
6. Download or share the GIF. Save a copy, use the shareable link, or send it directly to X, LinkedIn, or Reddit.

You may get still frames while GIF creation is rolling out to your account. If you already have an animated GIF and want to change it, ChatGPT currently supports frame editing on GIFs it creates. Uploaded animated GIFs can’t be edited.

Prompt Ideas That Work Well

Give your subject one clear action and tell ChatGPT how you want the loop to look. You can try these prompts as written or swap in a subject of your own.

• Make a chibi character hop twice and throw both hands up in a victory dance. Use a transparent background.
• Make a looping ’90s anime reaction GIF where the character gasps, blinks, and returns to their starting expression.
• Animate this pet as a clay character that tips its head to one side before returning to an upright sitting pose.
• Make this doodle into a felt puppet that waves at the camera.
• Create a paper cutout birthday GIF with candles flickering and confetti falling around the cake.

How Saved Reference Images Work

You see, the same photo can come in handy for more than one chat. You might animate your dog today and ask for a watercolor portrait next week. If you want to reuse the same photo for both, save it under a name you’ll remember.

• Save the photo. Upload the image and ask ChatGPT to save it under a name like Pepper. If ChatGPT offers to save the photo, you can accept and give it a name.
• Call it up in another chat. Type @Pepper in your prompt, as in “Paint @Pepper as a watercolor portrait.” ChatGPT can use the saved photo without another upload.
• Check which version you saved. In one early test, ChatGPT saved the original cat photo under @Chester. The edits made in the chat were separate from the saved reference.

OpenAI has yet to explain where you can manage or delete saved references. Use Data Controls to choose whether your uploads can be used for training, and save photos you have permission to use.

Who Gets Images 2.6 and What It Costs

You can get the Images 2.6 model on a Free account, but GIFs and saved references have their own rollout caveats. OpenAI has yet to confirm when Free accounts will get those two features or publish GIF limits by plan. Here’s how image access compares across the US plans.

The rollout also covers Codex, though one early test found the new features were still unavailable there. Access details for free Codex accounts remain unclear. Not sure which plan fits you? Read our breakdown of ChatGPT Go vs Plus and what the shared ChatGPT and Codex usage limit means.

What Developers Should Know About the API

If you build with the image API, choose Flare when speed matters and Sunburst when you want higher image quality and more consistent edits. OpenAI says the two API models are 50% faster. The price per token stays the same, but the estimated cost of a finished image depends on the quality level you request.

To make an animation, set output_format to gif in a generation or edit request. You can ask for 2 to 16 frames and set each frame to last 20 to 2,000 milliseconds. Leave those values alone and you’ll get nine frames at 200 milliseconds each, which gives you a 1.8-second loop.

You can export a new image as a static SVG or use a transparent PNG overlay to show the model where to make an edit. The complexity option accepts low, medium, or high, with high selected by default and no separate fee. For more on OpenAI’s recent pricing changes, see what GPT-6.1 Terra means for your API bill.

Safety Rules and Watermarks

What if each frame looks harmless on its own, but the motion changes what the GIF shows? OpenAI added checks for sexualized movement and harmful content that only becomes apparent across frames, though some violations can slip through. There are two things you can look at here, how often unsafe outputs got through in testing and how you can verify a generated file.

Safety Tests and Image Verification

OpenAI tested 3,077 prompts designed to get past its safety checks. The share of unsafe outputs that got through fell from 1.92% to 1.72% for Sunburst and rose from 1.49% to 2.14% for Flare. OpenAI described both differences as statistically insignificant.

Some users said they saw stricter refusals after the release. On November 8, OpenAI Support said an automated refusal by itself was insufficient to establish that the rules had tightened and said it had passed along the feedback.

Images generated in ChatGPT, Codex, and the API carry C2PA metadata and a SynthID watermark. OpenAI provides a public verification tool you can use to check a file.

How It Compares With Google’s Nano Banana 2.1

If you want to make a GIF, the choice comes down to which tool can give you an animated file. If you mostly make still images, you might care more about reference inputs or what each generation costs. Here’s how the two releases compare on those details.

On November 7, Artificial Analysis gave Sunburst an Elo score of 1259, compared with 1176 for Nano Banana 2.1. That early comparison measured still images generated from text prompts, so it tells you about a different task than making a loop. If Gemini’s image tool gives you trouble, try our fixes for Gemini image generation.

Is It Worth Trying Now

If GIF creation has reached your account, try it with a photo or doodle you already like. A small gesture is enough to see how ChatGPT handles movement, and you’ll have something specific to adjust after the first pass. Start with a wave, a head tilt, or a quick reaction.

Watch the loop once before you change the prompt. If the motion looks right but goes by too fast, slow the playback and watch it again. And if you’re still getting still frames, you can come back to that same idea as GIF access reaches more accounts.

FAQs

You might already have a GIF idea and still be wondering whether your account can make it. Here’s what to know about access, saved photos, and the speed claims before you try your first loop.