A photo of the dog on the sofa. One sentence with it: "Make him wag his tail." A few seconds later, he wags. On a loop. As a GIF you can drop straight into the family chat.
ChatGPT can do that now. At least once the feature has reached your account.
Last week OpenAI rolled out Images 2.6, the new version of the image generator in ChatGPT. Two changes matter for regular users: animated GIFs and reference images that ChatGPT remembers. Both are available on the free account too. With restrictions, of course.
GIFs: motion on demand
A GIF is a short, silent image format that plays in an endless loop. The internet has lived on them for decades: reaction pictures, cats, people falling off chairs.
Images 2.6 gives you two ways to make your own GIF. Either you describe what should be visible and how it moves. Or you upload a photo and ask ChatGPT to animate it. Afterwards you can adjust the speed and touch up individual frames of the animation.
Sounds like a toy. It is. But it's one a surprising number of people actually need: the waving snowman for holiday greetings, the reaction in the group chat, the little animation for the club website.
A sample prompt to get started:
Animate this photo as a GIF: the candle on the table flickers gently, snow is falling outside the window. Calm, slow movement, no camera motion.
There are two catches. First, OpenAI is rolling out the GIF feature gradually. If you don't have it yet, you're not doing anything wrong. Second, according to one report ChatGPT can't yet edit GIFs you upload yourself. It only animates what it generates itself or turns a still image into. OpenAI hasn't published a maximum GIF length.
Reference images: ChatGPT remembers faces
The second change is less flashy but matters more day to day. According to OpenAI, you can save a photo as a reference and reuse it in later chats. It's meant for people, pets and characters.
Until now it went like this: if you wanted to see your dog in five different scenes, you uploaded the photo five times. And every time the dog looked a little different. Now the template is ready and ChatGPT draws on it. For a comic character, a homemade children's book or the club mascot, that's real progress.
Sound familiar? Google launched something similar over the summer: Gemini turns your own photos into AI images. The big providers clearly all want the same thing: your pictures stored with them and nowhere else.
What's free, what costs money?
According to OpenAI, Images 2.6 is available on every plan, including the free one. On desktop, on your phone and in the browser. That doesn't mean unlimited, though.
For the free account, OpenAI only says image creation is "limited". There's no fixed number per day, and none for GIFs or reference images either. In practice: it's plenty for trying things out. If you want to build a whole series in one afternoon, you'll hit the wall at some point and get a friendly nudge towards a subscription.
That's no scandal. Generating images takes real computing power, and an animation is simply a lot of images. According to OpenAI, more than four billion images are now created per week across all channels. Somebody pays that bill. Google, Meta and the rest do it the same way, by the way: free to try, but with a limit.
For developers, two new models are coming to the programming interface (API, meaning direct machine access without an app around it): one tuned for speed, one for quality. OpenAI says they generate images 50% faster. The API bills by usage; nothing is free there. For beginners it's the wrong starting point anyway. The app is enough.
Before you save your face
One thing is worth a moment's thought. A saved reference photo doesn't disappear after the chat. It stays, so ChatGPT can find it again later. That's the whole point.
Still, no need to panic. If your photos live in your phone's cloud or on Instagram, you've long been handing pictures to big corporations. A reference image at OpenAI isn't a dam breaking, just one more storage spot. For the sober comparison, see our reality check Privacy and AI: ask your phone first.
Two things you can actually do:
- Only save your own photos as references. Yourself, your dog, your houseplant. For friends, colleagues or other people's children: ask first. They have a right to their own image, even if the AI doesn't check.
- Clean up now and then. Generated images sit in ChatGPT's library and can be deleted there. Where saved references can be removed again, OpenAI currently doesn't explain clearly. It's worth a look in the settings under "Personalization".
Who is this for?
Almost anyone who already uses ChatGPT. You need no prior knowledge, no API, no subscription. Open the app, upload a photo, write a sentence. Done.
If you need animations professionally, short GIF loops won't serve you well. There are dedicated video generators for that. They play in a different league, and a different price range.
And an honest meta note: the header images on this blog come from a self-hosted Stable Diffusion; ChatGPT only steps in when something breaks. Animated header images won't be happening here for now. Promise.
Just try it
Pick a photo that belongs to you. The dog, the coffee mug, the houseplant that has been begging for water for weeks. Write down what should happen. See what comes out.
If it doesn't work yet, wait a few days. OpenAI is rolling the feature out step by step.
Unpaid recommendation, my own opinion, as always here.