image to video (I2V)

in English: image-to-video

basicspreading

An AI method that creates a few seconds of video from a single still image: it imagines how the scene visible in the image would move.

In plain terms

Imagine giving a photograph to an animator and asking them to draw what happens in the next five seconds. Image-to-video does the same with a machine: the uploaded image becomes the starting point, and the model generates the continuation.

You can mostly specify what you want to see in text, for example, 'the camera slowly zooms in, the girl smiles, the wind blows her hair.' If you write nothing, the model decides itself what should move.

The main difference compared to text-to-video is that here you define the visuals. The characters, product, colors, and style come from your image, making the result more predictable. This is why advertisers and filmmakers like it: they first design the perfect still image and only then animate it.

How it works

Upload an image (photo, drawing, or AI-generated image), and if you wish, describe the desired movement or camera movement in text.
The model converts the image into a compressed internal format and uses this as a starting point, usually as the first frame of the video. In some tools, you can also specify the last frame.
The video frames start as random image noise, and the model cleans this up step by step. This is called a diffusion process. Meanwhile, the frames adjust to each other and to the starting image.
The model learned from countless videos how water, hair, a face, or a car usually moves. From this experience, it estimates the likely continuation; it does not 'understand' the scene.
At the end, the system generates real video frames from its internal format. The result is typically a few-second clip, often with audio on newer tools.

An everyday example

You take your grandmother's old black-and-white wedding photo, upload it to an application, and a few minutes later the image comes to life: your grandmother blinks, smiles, and her veil flutters. Many social media ads are created with the same technique, where a product photo suddenly starts spinning or steaming.

Who is affected

marketerFrom an existing product photo, you can create a short advertising video in minutes without a shoot, in multiple variations. → Marketing

entrepreneur, executiveEven on a small budget you can get video content for your webshop or social media page, but the legal and credibility risks are yours.

parent, student, teacherEven an innocent photo of a child or classmate can be turned into a misleading or humiliating video, which is why it is worth discussing at home and in school. → Education & Family

everyoneA video can be created from any of your public photos showing you doing something you never did. Therefore, a video alone is no longer reliable evidence.

A common misconception

MythThe machine shows what actually happened before or after the photo was taken.

In realityThe model does not know what happened; it only invents a plausible continuation based on learned patterns. The animated great-grandfather's smile is fiction, not a memory.

Limits and risks

Clips are short, and the longer the video, the easier it is to 'drift': the face changes, an object disappears, a hand or caption distorts.
The physics do not always add up: liquids, collisions, and complex movements often appear unnatural, so multiple attempts may be needed.
Moving someone else's photo without their permission may violate personality rights, such as the right to one's image, and can be used for fraud or harassment as a deepfake.
Generated video must be labeled in certain cases under the EU AI Act. Animating a recognizable brand or artwork also raises trademark and copyright issues.

Similar, but not the same

text-to-video – There, you only provide a description and the model generates the visuals; here, you provide the starting image, and the model only adds the motion.
deepfake – A deepfake is a forgery that misleadingly depicts a real person, i.e., a use case, while image-to-video is a technique with which both harmless animation and deepfakes can be created.
traditional animation – The animator designs and directs every movement themselves, but here the model generates the motion and frames, and the final result is only partially controllable.

Related concepts

Broader

generative AI, video generation, synthetic media

Narrower

photo animation, talking avatar, start-and-end-frame control

On the edge

text-to-video, text-to-image, diffusion model, deepfake, watermarking

A brief history

Researchers have long experimented with animating still images, but it reached the general public through diffusion models. In 2023, the first tools available for anyone to try appeared, including Runway Gen-2 and the Stability AI Stable Video Diffusion model.

Where things stand

By the end of 2026, video generation is a basic feature of the major video generators, available in mobile apps too, and an everyday tool in advertising and content production. Longer, consistent and precisely controllable scenes, however, are still not reliable.

Status: 2026 November

Also known as

image-to-video · image to video · I2V · img2vid · image-video generation · image-to-video generation · photo animation · image animation

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.