Image to video AI, briefly.

What “image to video AI” means.

The image becomes the start of the clip. Google’s Veo documentation defines the image input as “An initial image to animate,” and its prompt guide says “Veo uses the input image as the initial frame.”

The Gemini API has two video models, Gemini Omni Flash and Veo. Google says to use Omni Flash as the default for video generation, and describes it as quick at “turning text prompts and images into short videos.” It points to Veo 3.1 for scene extension, last-frame control, or legacy pipelines.

Three ways an image can steer the clip.

The Veo documentation describes three uses for an image.

• First frame. One image is the initial frame, and the model animates from there.
• First and last frames. A second image sets the final frame, and the model interpolates between the two. The last frame must be used together with the first image.
• Reference images. Veo 3.1 and Veo 3.1 Fast accept up to 3 reference images of a person, character or product to preserve its appearance in the video. Veo 3.1 Lite does not take reference images.

What you actually get.

The Veo 3.1 parameter table sets the shape of a clip:

• A clip is 4, 6 or 8 seconds long.
• On Veo 3.1 and Veo 3.1 Fast it must be 8 seconds when you use extension, reference images, 1080p or 4k.
• Resolution is 720p by default. 1080p and 4k are 8 seconds only, and Veo 3.1 Lite has no 4k. Extension is 720p only.
• The aspect ratio is 16:9 by default, or 9:16.
• Audio is always on, and the frame rate is 24fps.
• Each request returns one video.

Choosing the image and writing the prompt.

Because the image is the initial frame, Google’s guide says to pick the image closest to what you picture as the first scene of your video.

The prompt still does work. Google’s prompt guide asks for a subject, an action and a style. You can add the camera, the composition, the lens and the ambiance. Put spoken lines in quotes, and describe the sound effects and ambient noise you want.

For example, with a photo of a cafe table as the first frame: “Steam rises from the cup as a hand slides a croissant into view, warm morning light. Slow push-in, close shot. Someone says ‘Good morning.’ Soft chatter and a coffee grinder in the background.”

What Google lists as the price.

Google’s Gemini API pricing page, last updated 2026-09-24 UTC, lists Veo 3.1 Standard, Fast and Lite as “Not available” on the free tier. On the paid tier it lists these rates per second of video, with audio:

Gemini Omni Flash is also “Not available” on the free tier. Its paid tier is priced per token: $1.50 per 1M input tokens (text, image, video or audio) and $17.50 per 1M video output tokens. Google’s note says billing is calculated at 5,792 tokens per second of 720p video, which “equates to an effective price of approximately $0.10 per second.”

The same page says you are only charged if your video is successfully generated.

What to watch for.

These limits come from the Veo documentation.

• People in your image. For image-to-video, first and last frames, and reference images, the only personGeneration value is allow_adult. In the EU, the UK, Switzerland and the Middle East and North Africa, allow_adult is the only allowed value.
• Watermark. Veo videos are watermarked with SynthID, and the watermark can be verified on the SynthID platform.
• Storage. A generated video stays on the server for 2 days, then it is removed. Download anything you want to keep.
• Language. English is fully supported. Other languages have not been evaluated, so test a prompt in another language before you depend on it.

Why this page.

“image to video ai” was the first suggestion Google Autocomplete returned for the phrase “image to video” when we checked on 2026-09-30. That is the only claim we make about the search: we do not have a search volume for it.

Sources.

• Google Gemini API: video generation ↗
• Google Gemini API: Veo ↗
• Google Gemini API: pricing ↗
• Google Autocomplete, fetched 2026-09-30 ↗