Image to Video AI: How to Turn a Photo into a Video (2026 Guide)

Quick answer: upload a sharp image to an image-to-video tool, describe how things should move and what the camera does, and generate a short clip. Official tools that do this in October 2026 include Google Flow (Veo 3.1, Gemini Omni Flash), Kling 4.0, Runway and Luma Ray3.2. If the clip has to end on a specific shot, use a first and last frame or keyframe mode. Clip length depends on the model: from 4 seconds to 30 seconds per generation.

How does image-to-video AI work?

Your image becomes the first frame, and the model invents the frames that follow based on your text prompt. Most tools also accept:

• an end frame, so the clip moves from image A to image B
• extra keyframes in the middle of the clip, on some models
• reference images for a character, product or style, so they stay consistent

The image fixes the look; the prompt mainly controls motion. That’s why good image-to-video prompts describe movement rather than the scene.

Which tools can turn a photo into a video?

Checked on each vendor’s official pages on October 18, 2026:

Tool	Image-to-video controls	Clip length	Official page
Google Flow: Veo 3.1 (Lite, Fast, Quality)	Frames to Video with a first frame, or first and last frame; reference images on Lite and Fast	4, 6 or 8 seconds	Flow Help
Google Flow: Gemini Omni Flash 1.1	First frame, first and last frame, ingredients/references, video-to-video editing	4, 6, 8 or 10 seconds	Flow Help
Kling 4.0	Image-to-video, first and last frames, up to 10 keyframes, up to 15 combined references	3–30 seconds	Kling 4.0 guide
Runway	Image-to-video with Gen-4.5 and partner models such as Seedance and Kling	Varies by model	Runway Dev
Luma Ray3.2 (Luma App)	image2video, multi-keyframe control with up to 16 keyframes	5 or 10 seconds native	Luma Ray3.2

Each tool has its own plans and credits. Compare them on the official pricing page linked from each vendor’s site.

Is Luma Dream Machine still a thing?

No. Luma says the current video model is Ray3.2, that “Dream Machine” and Ray2 are deprecated, and that the product is the Luma App (Luma).

Is Sora still an option?

No. OpenAI shut down the Sora app on April 26, 2026 and the Sora API on September 24, 2026. See Is Sora shut down?

How do you turn a photo into a video, step by step?

1. Prepare the image. Use a sharp, well-lit image with the subject clearly separated from the background. Crop it to your final aspect ratio first (16:9 for YouTube, 9:16 for Shorts, Reels and TikTok). If you’re generating the still with an image model first, see Midjourney’s status and official image APIs.
2. Pick the mode. In Google Flow it’s called Frames to Video (Flow Help). Kling calls it Image to Video (Kling).
3. Write a motion prompt. Describe subject motion, camera motion and pace. See the examples below.
4. Add an end frame if needed. For a product turning to show its logo, or a before/after, set the last frame.
5. Generate short first. Use the shortest length to test the motion, then generate the final version.
6. Download and check rights. Read the tool’s terms on commercial use before you publish.

What makes a good image-to-video prompt?

Describe what changes, not what is already in the picture.

Weak prompt	Better prompt
“A woman in a red coat in Paris”	“She turns toward the camera and smiles; light wind moves her hair; slow push-in”
“A sneaker on a table”	“The sneaker rotates 180 degrees clockwise; soft studio light sweeps across; camera stays locked”
“Mountain landscape”	“Mist drifts left to right through the valley; slow aerial dolly forward; golden hour light”

General tips (not test results):

• One main motion per clip. Several big actions in one short clip often warp the subject.
• Name the camera move: push-in, pull-out, pan, orbit or locked-off.
• Use an end frame to protect identity. Pinning the first and last frames keeps a product or face from drifting.
• Keep text out of the source image if you can. Generated motion can distort small text.

When should you use first and last frames or keyframes?

Use them when the ending matters:

• First and last frame: a transition between two exact shots, such as a product closed and then open. Veo 3.1 and Gemini Omni Flash support this in Flow (Flow Help).
• Multiple keyframes: a planned sequence with several beats. Kling 4.0 accepts up to 10 keyframes (Kling 4.0 guide), and Luma Ray3.2 up to 16 (Luma).

For a step-by-step guide to start and end frames in each tool, see first and last frame AI video.

Can you do image-to-video through an API?

Yes. Google offers Veo 3.1 and Gemini Omni Flash in the Gemini API (Veo docs), Runway’s API has an imageToVideo endpoint (Runway Dev), and Kling and Luma have their own APIs. If you’re moving off OpenAI’s video endpoint, see Sora API shutdown: how to migrate.

Is it legal to animate any photo?

Use images you own or have rights to. For what each tool’s terms say about commercial use of the output, see Can you use AI videos commercially? Get consent before animating a real person. Most tools’ policies restrict realistic depictions of real people without permission, so check the policy of the tool you use. This is general information, not legal advice.

More tools by use case are in the AI video tools hub.