Google Flow: How to Make a Smooth Video from Two Photos

Do you want a video that starts on one photo and ends on another photo? In Google Flow, you can try to make such a video by giving it a start frame and an end frame. A frame means one still picture in a video. The start frame is the first photo; the end frame is the last one.

AI cover illustration — this is not a Google Flow screenshot, input pack or actual video result.

In this guide we will start with a simple vase shot: first a slightly wide view, then a close view of the same vase. The prompt below is a suggestion for practice; it does not guarantee a perfect result or the same details.

Watch the result first, then learn the steps in Hindi

Below is a 10-second actual Google Flow result. After that, a 1 minute 27 second Hindi guide shows the same input photos and real Flow screenshots. The guide is based on screenshots; buttons and the interface may look different depending on your account.

Actual AI-generated Flow demo: from a wide view to close to the vase.

Hindi tutorial: two photos, a simple prompt, settings, checking the result and download.

Before you start

Keep these ready: Google Flow in a computer browser, an account that has access to the feature, and two matching photos of your own. Check the current credit cost before you generate. According to Google’s model guide, Gemini Omni Flash 1.1 supports start/end frames; use the web for this feature. Availability may be different in different regions.

1. Choose two photos that match each other

Place a small vase on a table. In the first photo, show a little space around the vase. Take the second photo in the same place, with the same light, a little closer to the same vase. Do not change the vase, leaf or background between the two.

Hold the phone straight and choose the same shape for both photos. For a Reel, vertical 9:16 is useful: that is, a tall image like a phone screen. If you start with one horizontal and one vertical photo, matching the framing can be hard.

For practice, do not choose a very big change, such as an empty room in one photo and a crowded wedding in the other. It is easier to understand one small camera movement first.

2. Add the photos in Flow

Following Google’s documented desktop steps: open a project, click the model name in the prompt box, then choose Video → Frames. Drag the wide photo into + Add start frame and the close photo into + Add end frame.

Look at both previews first. If the order is reversed, the video may also move from close to far. Go to the next step only after the photos have uploaded.

3. Write what should happen in between

A prompt is the written instruction you give to the AI. Paste this text into the prompt box in Google Flow. You do not need to send it to a separate image-making chat.

Begin with the supplied start image and move gently toward the supplied end image. Use one slow, steady camera push-in toward the terracotta vase. Keep the vase, leaf, table and background unchanged. Preserve the same daylight and natural shadows throughout. No cuts, no new objects and no sudden camera movement.

How to use a prompt →

If you are using a cup instead of a vase, replace “terracotta vase” with “ceramic cup”. Also remove “leaf” if there is no leaf in your photo. Keep only the objects that are in your photos in the prompt.

4. Make one output and watch all of it

Check the video model, ratio and duration in the available settings. First select one output; then press Generate. When the AI result comes, watch all three parts: the start, the middle and the end.

It is not enough for only the first frame to look good. Is the vase changing shape in the middle? Is the leaf disappearing? Is the background moving? Does the ending get close to the photo you supplied? Do not treat the video as final without checking these things.

What to change if the video goes wrong

Objects are changing: First compare the two input photos. If the leaf or light is already different in them, start again with matching photos.
The movement is too fast: Keep just one slow push-in. Do not also ask for spin, zoom and a flying camera.
The face is changing: Giving start/end photos does not guarantee that identity stays locked. If you are making a shot of a person, check the face through the whole clip.

Before a paid retry, decide on one specific problem and change only the instruction related to it. You do not need to press Generate again and again without checking.

5. Download the video

If you like the result, hover over that video asset in the project and choose More → Download. Select an available file type. This route is given in Google’s download guide.

Open the saved file on your computer and play all of it. Check the actual quality and shape of the video. Do not take this guide as a promise of fixed 4K or free export; check the output options available for your model and account.

Cannot see the Frames option?

Check that you are using Flow in the browser and are in Video mode. Look at the current feature list for the selected model. If the option is not available, do not try to unlock it by making extra paid generations; check your account and region access.

Once the motion clip is ready, if you want to plan its opening shot, get ideas from the simple Reel opening guide. First make the image-to-video clip, then do the editing for the final Reel.

Want a different look?Build your own prompt — pick the outfit, background and lighting, then copy it in one tap.

Open Prompt Builder