Animate a photo with AI

A photo where everything has frozen, and the question that occurs to anyone looking at it again: what would this look like in motion? Today all it takes is to attach the photo and describe in words what should come alive. This page collects everything on the subject: what actually happens to your file, which frames animate cleanly and which do not, how to word the request, and why clip length is the only setting that really moves the price.

The main thing first: what actually happens

The model does not add motion to your file. It takes your photo as the first frame and builds every frame after it by itself: dozens of frames that nobody ever shot.

Everything else follows from that. The less the model has to make up, the closer the clip stays to your photo, which is why the source photo decides the outcome, not a clever turn of phrase. That is a measured property of the method, not a disclaimer added just in case.

The first frame is yours. The model invents all the others: it does not add motion to your file, it shoots the scene again, starting from it.

What animates cleanly and what does not

How hard the job is depends not on the size of the subject or the length of the request but on one thing: how much of the scene the model will have to invent. That single measure orders the whole ladder, from a landscape where the elements move by themselves to readable lettering that will never come out right.

What makes a job hard is not the size of the subject but how much of the scene the model has to invent in its place. The top row almost always works; the bottom row never does.

The top row works almost every time. The bottom row cannot work at all: a video model distorts letters, and text in the frame turns into something that only looks like text. If the words must be readable, animate the frame without them and lay the text over the finished clip as a separate step.

Look at your photo before you pay

Almost everything that can go wrong is visible in advance, right there in the file you are about to upload. The seven questions below cover the great majority of failed attempts.

Look at your own photo

Tick everything that is true of the photo you plan to animate. The source decides the clip, not the wording of the request, and almost everything that can go wrong is visible in advance.

A frame like this animates well

Nothing needs fixing first. Build the brief and start with the shortest clip.

How to word the request

A good animation request is four decisions, not a beautiful description: what is in the frame, one camera move, what must stay unchanged, and what must not appear. Two camera moves in one request come back as jitter, and if you say nothing about sound the engine takes it as “compose some music yourself”.

Build the shot brief

Four decisions, and a finished request appears below that you can copy or open straight in the chat. These four points are what separate a clip that works first time from a series of paid retries.

1. What is in the photo

Both the motion and the list of things to avoid depend on it.

The face is the most noticeable part of the frame: any extra movement reads as a different person.

2. One camera move

Exactly one. Two moves in a row come back as a jittering frame. That is a property of the method, not an unlucky attempt.

The most reliable option. Start here.

3. Sound

Sound is created with the picture in the same pass and costs nothing extra, but if the request says nothing about it, the engine writes music of its own.

Silence has to be requested: if the request says nothing, the engine takes it as “write a soundtrack”.

4. Length

Not decided here but after the first viewing: the request ends by asking for the shortest version. Only extend what you already like. Billing is per second, and a long clip costs exactly as many times more as it is longer.

Finished request

Animate this photo: the person in the photo, breathing, an occasional blink, a barely noticeable turn of the head, a slight movement of the hair. One camera move for the whole clip: the camera is static, only the scene moves.
Keep unchanged: facial features, age, hairstyle, clothing and background.
Must not appear: face distortion, changes in appearance, winks and smiles on cue, readable text or lettering in the frame, any sound or music.
Make the shortest version first; I will watch it and tell you whether to make it longer.

Open the chat and attach the photo

In the chat you only need to attach the photo; the request will already be in the input box.

What it costs, and why

Video is the most expensive thing AI does today, and its price works in the simplest possible way: you pay for seconds of finished clip. That leaves one rule for saving money, and it is not about choosing a model.

What length costs, and what resolution costs

Billing is per second: a clip costs exactly as many times more as it is longer. These are ratios, not prices; ask the chat for the exact figure in your tokens before it generates.

Clip length5 s

Resolution

×1.3

compared with the cheapest option: 4 s at standard resolution

One complete movement with a start and an end. The working length for an animated photo: go longer and the motion starts to repeat.

Which leaves one rule for saving money: start with the shortest clip at standard resolution, and only if you like it, make the same shot longer or sharper. Upscaling a clip you like costs less than half of a fresh 2K generation, and the test plus that upscale add up to exactly what generating in 2K from the start would cost, so the test is free in the end. Trying an idea out on a long clip is the most expensive way to learn that it does not work.

4–15 sLength of one clipWhole seconds. Start with the shortest.

1 moveOf the camera per clipTwo moves in a row come back as a jittering frame.

From the photoAspect ratioA portrait photo gives a portrait clip, a square one gives a square clip.

Attach a photo and describe what in it should come alive. Video costs more than working with images, and your first clip is earnable without paying: tokens for signing up, for five first tasks (a web search and uploading a document, a photo, an audio file and a video) and for coming back on the second and third day of a streak add up to it.

Loading…

Motion and speech are two different jobs

There is one fork worth knowing about separately, and almost nobody guesses it until they hit it. Making a photo move and making it speak are different routes inside the product, and they cannot be combined in one request. That is a limit of the engine itself, not of the interface.

Motion and speech are different routes. A portrait as the “first frame” gives motion; a portrait as a “reference” plus a voice recording gives a person who speaks. They cannot be mixed in one request: that is a limit of the engine itself.

What the second route looks like in practice, what to write in quotes and how to supply a particular voice is covered in the chapter a talking photo.

Where to go next

The chapters of this book take individual cases, each with its own mechanics and limits; the full list is in the table of contents below. If you came here with “free” or “online” in your search, go to animate a photo free and online. If you already tried a voice assistant and found it too tight, see “Alice, animate my photo”. And the old family photograph, with the conversation about respect for memory, lives in the chapter video from a photo of the video generation book.

Frequently asked questions

How do I animate a photo with AI?

Attach the photo to the chat and describe what should come alive: one camera move, what visibly changes and what must stay the same. The model takes your photo as the first frame and builds the rest. Start with the shortest clip, since it shows how the model read the frame. You can try it in the Twelver chat.

Is an animated photo my photo with motion, or a new video?

It is a new video with your photo as its first frame. The model invents every other frame: it does not lay an effect over the file, it reshoots the scene, as it were, starting from your frame. So the more the frame holds that has to be made up, such as people in motion, a busy background or fine detail, the further the result drifts from the original.

How long is the clip?

From 4 to 15 seconds, in whole seconds. For an animated photo the working length is 4–6 seconds: one complete movement with a start and an end. Beyond that the model starts inventing ways to fill the frame, and a second movement nobody asked for appears in the clip.

Which photos animate best?

Sharp photos at their original size, with a distinct foreground and background and no readable text in the frame. Landscapes and views almost always work, because the elements, clouds, water, leaves, move naturally by themselves. Group photos are the hardest: every face in the frame is another chance that the model “fixes” that one.

Will my photo keep its aspect ratio?

Yes. When a photo is animated, the aspect ratio comes from the file itself: a portrait phone photo gives a portrait clip, a square one a square clip. Nothing is cropped to landscape. White borders and old crop marks, however, go into the video as they are, so remove them beforehand.

Will the clip have sound?

Yes. Sound is created together with the picture in the same pass and is not billed separately. The side effect: if the request says nothing about sound, the engine composes a soundtrack by itself. So silence has to be asked for in words: “no music and no extra sounds”.

Can the person in the photo say something?

Yes, but that is a different route: the portrait is sent as a reference rather than as the first frame, the line goes into the request in quotes, and the lip movement is computed together with the picture in a single pass. It cannot be combined with ordinary animation in one request. Details in the chapter a talking photo.

Why did I get a different person?

Almost always because of the source, not the request. From a soft, noisy or messenger-forwarded photo the model invents the facial features too, and invents them afresh every time. Use the original file rather than a forwarded copy, and ask for minimal motion: breathing and micro-movements instead of head turns.

In the Twelver chat you can attach a photo and get a short clip with sound from it, online, in one conversation, with no separate apps. Your first clip is earnable without paying: sign up, complete five first tasks and come back on the second and third day.

Try it yourself

Everything in this guide runs inside Twelver

One chat for text, images, video, music and voice — no separate services or subscriptions.

Open Twelver chat

Related pageVideo Generation

Rate your experience