How to turn an image into a video with AI

The short answer

To turn an image into a video, choose a video model that takes a start image, upload the image, and write what happens in the clip. In practice it is five steps:

1. Choose a model and switch the studio to "Image to video"
2. Upload the image in the "Start image" field, a file of up to 4 MB. If you also want to set how the clip ends, upload an "End image" in models that support it
3. Write what moves in the prompt: who does what, how the camera moves and what sound there is. There is no need to describe again what the image already shows
4. Pick the length, resolution, aspect ratio and sound, and see the exact price before you send
5. Send it. The clip is saved to your personal library the moment it is ready

For example, animating an image into an 8 second clip at 720p with sound in Veo 3.1 Lite costs ₪0.707.03 crVeo 3.1 Lite, 720p · 8s · with sound right now.

Text to video vs image to video

In text to video the model invents everything: the characters, the place, the colors and the composition. That is great for new ideas, but every generation can look a little different.

In image to video your image is the first frame of the clip. The look is already fixed: the face, the product, the colors and the framing. The prompt only tells the model what happens from that moment on. That makes it the right way when the look matters: a real product, a character you already designed, a logo or a specific place.

There is also a third way, "References", or reference images. There the images do not become the first frame. They show the model how a character, product or style should look in a new clip. It is the way to keep the same character across several clips, and it is explained in the guide how to keep the same character.

Start image and end image

A start image is the image the model starts from, and it is the first frame of the clip. An end image is the image the clip ends on, and the model fills in the motion between the two on its own. In most models an end image works only together with a start image.

An end image is useful when it matters where the clip ends: a product that opens, a before and after, a move from a wide shot to a close up or a move between two places. According to Google, in Veo 3.1 the transition between the two images comes with sound as well, and Google describes Gemini Omni 1.1 Flash as a model that makes a smooth camera move between a start and an end image.

In some models adding an end image changes the price. There is nothing to work out: the price in the studio updates at once when you add or remove an image.

Which model to choose

Almost every video model on the site takes a start image, so the choice depends on what the clip needs:

• The independent ranking. In an independent image to video ranking, which is based on people’s votes and was updated on September 21, 2026, the top four places went to MiniMax H3, Gemini Omni Flash 1.1, Wan 3.0 and Seedance 2.5. All four are available here
• An image that comes alive with sound. The Veo 3.1 family creates sound together with the picture. Start with Veo 3.1 Lite, which Google describes as its most cost-effective video model
• A product on a clean background. Seedance 1.5 Pro has a "Fixed camera" switch: the camera stays still and only what is in the frame moves. ByteDance also describes it as a model that creates voices and effects together with the picture
• Body movement and facial expressions. MiniMax describes Hailuo 2.3 as good at complex movement such as dancing and flips and at subtle facial expressions. Hailuo 2.3 Fast is its faster version, and it works only from an image
• A long clip from one image. Wan 3.0 and Seedance 2.5 reach 30 seconds, and Kling 3.0 reaches 15 seconds
• Quick and simple animation. Gen-4 Turbo and Wan 2.6 Flash work only from an image, and each of their makers describes it as a fast version

If you are not sure, run the same image and the same prompt in two models, at a short length and a low resolution, and continue with what worked.

Which settings to pick

• Aspect ratio. It should match the shape of the image: 9:16 for a vertical image, for Reels and Stories, and 16:9 for a landscape one. In Wan 3.0 the "Adaptive" setting picks a shape that fits the image on its own, and MiniMax writes that in H3 the clip shape follows the image
• Length. For one simple movement a short clip is enough. A longer clip leaves room for more action, and in most models it also costs more
• Resolution. Test the idea at a low resolution, and go up to a high resolution only for the final version. In most models a higher resolution also costs more
• Sound. In most models you can turn it on or off, and in some it changes the price. For Hailuo 2.3 and Gen-4 Turbo the makers list no sound, and Ray 3.2 makes a silent clip according to Luma
• Prompt enhancer. In some models the model expands your prompt before it generates. That helps when the prompt is short, and when you want precise control you turn it off

Which image works

The image must be up to 4 MB. According to Kling, the quality of a clip made from an image depends strongly on the image’s resolution, so a sharp, clean image gives a better clip. It helps when the subject is clear in the frame and there is room around it for the movement you ask for.

What to write in the prompt

The model already sees the image, so the prompt should tell it what happens: who moves and how, where the camera goes, and what you hear. When you describe the image again, the prompt gets longer and the model gets fewer instructions about the movement itself.

It is best to write the prompt in English. Google, for example, writes that in Veo English is fully supported and other languages have not been evaluated. Three examples, with an explanation:

Writing SFX for a sound effect and Ambient noise for background sound comes from Google’s official prompting guide for Veo 3.1. The full structure of a good prompt, with a glossary of camera moves in Hebrew and English, is in the AI video prompt guide.

When there is an end image too

When you upload a start image and an end image, describe the move that connects them. In the example in Google’s guide for this mode, the prompt asks the camera to circle a singer in a smooth 180 degree arc, from a view of her from the front to a view from behind her on stage. That way the model knows not only where to start and end, but also how to get from one to the other.

What it costs to animate an image

The table is computed as the page loads, with the same calculation the studio shows before every generation. Each row animates one image at the settings written in it.

Every model page has a full price table by length and resolution, and every model is on the models and prices page.

When the result is not what you wanted

• The camera moves too much. Write Static shot in the prompt, or pick a model with a "Fixed camera" switch
• The clip ends in the wrong place. Add an end image, so the model knows where to arrive
• The subject moves strangely. Ask for one clear movement, and make the clip shorter
• The framing is off. Pick an aspect ratio that matches the shape of the image, or "Adaptive" in models that offer it
• There is no sound. Check that the model creates sound and that its switch is on

And if a generation fails, all its credits come back to your balance automatically. The details are in the guide failed or stuck generations.

Questions and answers