ChatGPT Image 2.6: Turn Reference Photos into GIF Stickers, Cards and Product Clips
Image 2.6 generates GIF animations directly. The other change is reference images: upload a photo or a product shot, and the characters and products generated afterwards follow that image instead of being re-described every time.
Put the two together and you get the method in this guide: lock the subject with a reference image first, then make it move.
The update is available on every plan: web, desktop, mobile and Codex all have Image 2.6, and OpenAI says this version is 50% faster (source: OpenAI announcement).
Three uses follow, with prompts you can copy directly. Each one ends with fixes for common failures:
- Make a GIF sticker pack from photos of yourself or your pet
- Reuse the same character image for holiday cards
- Turn real product photos into e-commerce clips
Before you start: reference images and motion descriptions
Whether a GIF works comes down to two things: the quality of the reference image, and whether the prompt states the motion clearly.
How to choose a reference image
- Clear subject: the part you want to keep should fill more than half the frame, tighter is better. Anything hidden by a hand, packaging or another object has to be invented by the model, and the invention rarely matches the real thing.
- Clean background: the more clutter in the background, the more likely it gets drawn into the subject.
- Visible features: the features you want to keep (a cat's markings, a hairstyle, a product logo) must be visible in the image. Seeing them is not enough: name them in the prompt too.
- Multiple images are fine: one front view and one side view.
No reference image? Generate one first
Without a photo, you can generate a static "character sheet" in ChatGPT first, save the one you like, and use it as the reference for every GIF after that.
- Download and re-upload: save the character sheet you like and upload it again for each GIF. That is more reliable than saying "use the one above" in the conversation.
- The sheet is also material: when you publish, post the reference and the result together so readers can see the full "reference → result" process.
Uses one and three each have a "with material" and a "without material" starting point. Use two (holiday cards) reuses the character sheet from use one and needs no extra material.
The general GIF prompt format
A still image only describes what is in the frame; a GIF also has to describe what happens next. This five-part template works as is:
[Subject] Keep the ___ in the reference image unchanged (look / color / outfit)
[Style] ___ (e.g. flat illustration, chibi 3D, watercolor)
[Action] Starts at ___ and ends at ___
[Loop] First and last frames match, seamless loop / plays once and stops on the last frame
[Frame] Square, ___ background, fixed cameraThe parts people skip most are [Loop] and [Frame]. Without "fixed camera", the model may add handheld motion and the image shakes; without a loop instruction, the last frame jumps visibly back to the first.
Use one: a capybara "laid-back team lead" sticker pack
This is the quickest one to try. The subject is a plush capybara: half-closed eyes, a small orange on its head, a "team lead" armband, and an expression that never loses its composure.
Readers with pets can use their own pet's photos; the method is the same.
Step 1: Define the character first
Generating GIFs directly tends to produce a different-looking character in every expression. The steadier approach is to make one static character sheet and use it as the reference for every expression.
With a photo: upload a pet photo and use this prompt:
Using the pet photo I uploaded as reference, turn it into a plush-toy style character.
Keep: coat color, markings and the most recognizable features (be specific, e.g. "a notch in the left ear").
Realistic plush material close-up, with visible fur and stitching detail.
Front view, full body, light beige background, natural light, square frame.Without a photo: generate this guide's subject, the plush capybara "laid-back team lead", directly:
An original plush-toy style capybara character, fluffy short fur, round and stocky,
half-closed eyes, a small orange on its head, an expression of "absolute composure".
Wearing a small armband that says "team lead".
Realistic plush material close-up, with visible fur and stitching detail.
Front view, full body, light beige background, natural light, square frame.The more specific the features you keep, the more consistent the character stays. "A capybara" matches many capybaras; "a half-closed-eye capybara with an orange on its head and a team lead armband" matches only this one. The orange and the armband are the identity markers that hold every later GIF together.
Step 2: Generate the expressions one at a time
Upload the character sheet and generate one expression per run:
Keep the character's look and plush material exactly as in the reference image, and make a GIF sticker:
Action: the capybara keeps its half-closed eyes and nods slowly once, the orange on its head sways slightly.
Text: white bold text "On it." appears at the bottom, with a thin black outline.
Loop: first and last frames match, seamless loop.
Frame: square, light beige background, fixed camera.Change the action and the text to build a whole set. Here are six common stickers:
| Text | Action description |
|---|---|
| On it. | Slow single nod with half-closed eyes; the orange on its head sways |
| Off duty | Eyes snap wide open, it dashes out of frame with the orange, leaving smoke |
| Slacking | Soaking in a hot spring with only its nose and the orange above water |
| Composed | Papers fly and phones ring behind it while it chews grass, expressionless |
| Read | Turns slowly to the camera, blinks once, turns slowly back |
| Thanks boss | Cups the orange in both paws and offers it solemnly toward the camera |
Capybaras barely move, which is actually what makes them good stickers: the less motion, the less that can go wrong.
Common problems
- The character changes in the middle frames: the action is too complex. Keep one or two actions per GIF.
- Garbled or misspelled text: put the text in quotes and state font weight and position separately; if that still fails, generate without text and add it in an editing app. WeChat stickers are small, so less text is actually clearer.
- Too large for WeChat: WeChat limits sticker size, so lower the frame count and dimensions.
Once this set is done, use two is mostly a change of background.
Use two: the same character sheet, a new scene for holiday cards
This extends use one: same character sheet, same prompt, only the scene and greeting change. With the character already made, a card needs three edits.
How to do it
Upload the character sheet, apply the prompt below, and change only "scene", "action" and "text":
Keep the capybara's look and plush material exactly as in the reference image, and make an animated holiday card GIF:
Scene: ___
Action: ___
Text: "___" written across the top in a handwritten style, white letters with a slight outline.
Loop: first and last frames match, seamless loop.
Frame: vertical 3:4, fixed camera.Four templates you can reuse
| Holiday | Scene and action | Greeting |
|---|---|---|
| Lidong | Sitting in a red scarf in front of a bowl of hot dumplings, snow gathering on the orange | Happy Lidong, don't forget dumplings |
| Singles' Day | Buried under delivery boxes with only its head and orange showing, blinking slowly | Final payment done, composure intact |
| Christmas | Soaking in a hot spring in a Santa hat while snow falls, the orange replaced by a small bell | Merry Christmas, happy lying flat |
| Birthday | A small cake in front, blowing out the candles slowly, confetti falling | Happy birthday, peace every year |
Swap in your pet's character sheet and the same templates work for every holiday of the year.
Common problems
- Wrong or garbled greetings: keep them under 8 characters; if it still fails, generate a version without text and add it in an editing app.
- Scene props steal the scene: characters changing and props merging into the subject come from the same cause, too many elements in the scene. Keep it to two or three; make prop colors contrast strongly with the subject so the scarf is not drawn as part of the fur.
Use three: turn real product photos into e-commerce clips
The first two uses are personal. This one is for stores and product pages: one phone photo of a product is enough to generate rotation, unboxing and lifestyle clips without a studio, a model or lighting.
However clear the photo, the model only sees it as an image: it does not know what the product is or which parts must not change. So this section explains how to introduce the product in one sentence first, then how to shoot it.
Introduce the product, then generate
The reference image only carries what it looks like, not what it is. Before generating, describe the product in one or two sentences: what it is, what material, and which parts must stay unchanged.
Referring to the uploaded photo, this is a matte haze-blue stainless steel insulated bottle with a dark gray lid.
Keep the body color, proportions and white "MIZU" logo exactly unchanged.The more specific the introduction, the less the model improvises. Once that is done, talk about shooting.
How to shoot the reference photo
- Set it on white paper or a plain desk and use natural light from a window; shoot one front view and one 45-degree view, since a front-only image forces the model to guess the sides.
- Make sure the logo and any packaging text are sharp.
No physical product: when writing a guide, drafting a proposal or trying the effect, generate a fictional product as the reference. This guide uses a "film snapshot" style: grain, flash and slight overexposure, close to an everyday phone photo. Polished AI images are easier to spot; grain and overexposure read more like a real recommendation post:
A casual shot on Kodak Portra 400 film: a matte haze-blue insulated bottle with a white "MIZU" letter logo, sitting on a commuter train seat on a winter morning, with headphones and half a transit card beside it. Disposable camera flash, slight overexposure, visible film grain, slightly off-center composition, like a life photo a friend took. Vertical 4:5.To make the later rotation steadier, generate a second casual shot of the same bottle at 45 degrees and use both as references.
Note: fictional products are for demonstration only. Use real product photos for actual sales.
Three common clips
1. 360-degree rotation (for the main product image)
Referring to the uploaded product photo, keep the product's shape, color, material and logo exactly.
Make a GIF: the product sits on a slowly turning white platform and rotates one full circle at constant speed.
Loop: seamless loop.
Frame: square, light gray gradient background, soft studio lighting, fixed camera.2. Unboxing (for detail pages and recommendation posts)
Referring to the uploaded product and packaging photos.
Make a GIF: top-down view, a pair of hands opens the box lid and takes the product out onto the desk.
Plays once, stopping on the arranged product.
Frame: wood-grain desk, warm natural light, fixed camera.3. Lifestyle (for social posts, film snapshot style)
Referring to the uploaded bottle photo, keep the bottle's color, material and "MIZU" logo exactly.
Make a GIF: on a commuter train on a winter morning, the bottle sits on the seat with a wisp of steam from its mouth, light and shadow sliding across the body frame by frame.
Style: Kodak Portra 400 film texture, visible grain, warm color cast, slight overexposure, like a candid shot.
Loop: seamless loop.
Frame: vertical 4:5, fixed camera.Common problems
- Logo or packaging text distorted: shoot a sharper reference and state "keep the logo unchanged" in the prompt. If it still fails, change the angle so the logo is not facing the camera straight on.
- The back is invented during rotation: add a back-view photo, or rotate only 90 degrees and back.
- Colors differ from the real product: name the specific color in the prompt, e.g. "matte haze blue".
For real sales, the product details in the clip must match the physical item; check them against the product before publishing.
Other uses
The "reference image + motion description" combination has two more uses:
- Animated four-panel comic: generate a static four-panel comic, cut out the last panel, animate it as a GIF and paste it back, so the last panel moves. More steps, worth doing once you know the flow.
- Tutorial GIFs: processes like origami, latte art or workout moves read better as looping GIFs than as text. For anything with form requirements (workouts, for example), verify the motion is correct yourself.
Need a ChatGPT account, or want to upgrade to Plus / Pro for Image 2.6? See the ChatGPT top-up page for supported plans, payment methods and delivery.
Related reading
- ChatGPT Image 2.5 sketch workflow — another use of the same image features
- ChatGPT sticker tutorial — make WeChat / QQ stickers from photos
- How to subscribe to ChatGPT Plus — image generation needs Plus or above
- UPGPTs · ChatGPT top-up service — activation from China