How to Make a GIF in ChatGPT: Step by Step, With Prompts That Work and Text That Holds

ChatGPT can now make a GIF from words and from a picture. How to do it, what to write so the loop looks smooth, how to keep text correct in every frame, and three prompts to copy.

Since GPT Image 2.6, the image model OpenAI released on November 4, 2026, ChatGPT can make an animated GIF. You need no software and no editing skills: write what happens, optionally attach a picture, and get a short loop for a post, a story or a message. We ran it ourselves, both inside ChatGPT and through the model directly, including Hebrew text. Here is how to do it, and these are the sentences that separate a GIF that looks smooth from one that jumps.

How to Make a GIF in ChatGPT, Step by Step

• Open a conversation in ChatGPT. OpenAI says the feature is rolling out gradually, so it reaches accounts at different times.
• Want a GIF of your own character or product? Upload a still of it. Want something new? Words are enough.
• Ask for a seamless looping animated GIF, describe one movement, and pick an aspect ratio if you need one.
• Send and wait. OpenAI says it may take a few minutes, and our run took 1 minute 21 seconds.
• Open the GIF you got, and from there change its speed, save it or share it.

Worth knowing: The starting point is words or a still picture. According to the OpenAI help page, an existing GIF you upload to the conversation cannot be edited. Want to change an old GIF? Save one frame of it as a picture and start from that.

What We Got in a Real Run

We uploaded one character sheet to ChatGPT and asked for a dance on a beach with a Hebrew caption, from a Plus account. We got 9 frames, 1.3 seconds long, at 414 by 414 pixels. The character kept the same face, hat and shirt, and the Hebrew came out right in all nine frames. That is one run, so it is an example and not a promise: check every GIF before you publish it.

Four Rules for a GIF That Looks Smooth

• A locked camera: write The camera stays fixed. The background stays put and only the character moves.
• The end returns to the start: The last frame returns exactly to the first frame. This is the sentence that removes the jump in the loop.
• One small movement: a wave, a wink or steam rising from a cup. Over a few frames, a simple movement is what looks good.
• Text stays put: write The text stays in the same position in every frame, so the letters do not dance.

Three Prompts to Copy

The prompts are in English, because that is how the model takes the most precise instructions, and the text that should appear on screen is written in Hebrew inside quotes. All three ran for us. The first is a neon sign for a business, with Hebrew. Swap the quoted words and the type of business:

A seamless looping animated GIF of a glowing neon sign on a dark brick wall of a small bakery at night, photorealistic. The camera stays fixed. Only the neon moves: it flickers twice and then glows steady, and a soft pink and blue light pulses on the bricks. The last frame returns exactly to the first frame. The neon reads in big Hebrew letters, exactly: "יוצא מהתנור". Every Hebrew letter must be correct and read right to left. The text stays in the same position in every frame. No other text anywhere.

The second is a product loop with no text. Only the steam moves, and that is what makes it look like an expensive shoot. Swap the cup for your product and the steam for whatever moves around it:

A seamless looping animated GIF, a cinemagraph of one espresso cup on a marble cafe table in warm morning light, photorealistic. The camera stays fixed and the cup and the table never move. Only the steam rises and curls slowly above the cup, and the last frame returns exactly to the first frame. No text anywhere.

The third moves a picture you already have, such as a business mascot. Attach the picture and send. For us it was a clay falafel mascot:

The falafel character from the attached picture jumps happily on the pita with a small bounce and waves its raised hand at the camera, in the same real clay stop-motion style. No text anywhere.

Text Inside a GIF: What Keeps the Letters Right

In a GIF every frame is its own picture, so the letters get several chances to change. Three things helped us: write the text in quotes after the word exactly, add the sentence Every Hebrew letter must be correct and read right to left, and keep the words few. When you run the model directly, choose Sunburst at medium: in every GIF we ran that way the Hebrew came out right. Before publishing, go through the GIF frame by frame. A letter that moves for a moment stands out a lot in a loop that repeats.

When You Want to Choose: Frames, Shape and Price

Inside ChatGPT you talk to the model, and it decides most settings. In the Cadabra AI studio you choose up front: the model, the shape (512 by 512, 768 by 432 or 432 by 768), the number of frames up to 16, and the speed. A Sunburst GIF took us about half a minute and came out at 512. Again that is one run, so it is an example and not an average. The price appears before you approve:

• A GIF from words, no text (Flare, low): 1 credit.
• A GIF from words, with Hebrew (Sunburst, medium): 2 credits.
• A GIF from your picture, 6 frames (Flare, medium): 3 credits.
• A GIF from a picture, with Hebrew (Sunburst, medium): 4 credits.

Prefer working inside the conversation? Connect the Cadabra AI connector to ChatGPT or Claude, and ask for the same GIF in the same conversation, with your choice of model and the price up front.

Where to go next

• The full GPT Image 2.6 guide: more GIFs and prompts to copy
• 105 image codes, and here they run
• 61 codes that turn a picture into a clip
• More guides on the blog