Google Flow Image to Video: Start and End Frames

Starting from an image is the most controllable way to make video in Google Flow. The frame fixes the composition, the character, the product and the light before any motion is generated, so the model only has to decide how things move. Flow calls this Frames: a start frame on its own animates an image, and a start frame plus an end frame makes a clip that begins and ends exactly where you want. This guide covers the steps from Google's Flow help, which models support what, how to make good frames, and how to prompt the motion between them.

Adding frames, step by step

1. Open a project in Flow and go to the prompt box.
2. Click the model name, then Video, then Frames. (Flow's prompt box opens with the image model Nano Banana Pro selected, so this switch matters.)
3. Drag an image to + Add start frame. For a set ending, drag a second image to + Add end frame.
4. In the text prompt, describe the action or transition that happens between the frames. You can also simply ask Flow to animate the image.
5. Choose the aspect ratio, number of outputs, model and length, then click Generate.

Google notes that feature availability differs by country and region and that certain image types may not be supported.

Which model supports what

The table follows Google's "Learn about Flow models and supported features" page at the time of writing; Flow tells you if you pick a feature the active model does not support. Google lists start and end frames with Omni Flash as available on Flow on the web.

Making frames that animate well

Flow includes image models for exactly this job: Nano Banana Pro (the default for Ultra subscribers), Nano Banana 2.1 and Nano Banana 2 Lite, which Google describes as the default at no charge. You can also upload images made elsewhere, or save a frame from an existing clip: pause on it, hover, click Save frame, and use it as a start frame, end frame or ingredient.

• Same aspect ratio as the clip. Make the frame at the shape you will deliver.
• Room to move. Leave space in the direction of motion.
• Clean detail. Fix hands, faces and edges in the still; motion exaggerates flaws.
• Matched pairs. A start and end frame should share the camera position, light and subject. The best way to get a matched end frame is to edit the start frame rather than generate a new image.

Prompting the motion

Google's best-practice advice for Flow is that your text prompt should complement, not contradict, your visual inputs, and that it should refer to the frames you provided. In practice:

1. Describe what changes, not what is there. The frame shows the woman at the window; the prompt says she turns towards the camera and smiles.
2. Name the camera. "Static camera", "slow push in", "gentle pan left". Without it the model chooses.
3. Add the sound. Veo generates audio with the picture, so close with a sound sentence: ambience, effects and "no music" if you will score it yourself.
4. One action per clip. Eight seconds hold one clear action well; split anything longer into several clips.

With an end frame, describe the path between the two images: "the camera slowly orbits to the left until the bottle faces us; the light stays soft and even". Large differences between frames force an invented transition, which is where objects distort.

From one clip to a sequence

A frames clip is often the first shot of a longer piece. If it is an 8-second Veo 3.1 clip, you can extend it (the extension is generated with Veo 3.1 Lite). Or save its last frame and use it as the start frame of the next clip, then arrange the clips in Scenebuilder. Both are covered in making longer Veo videos. When a character must appear in several shots, combine frames with ingredients as described in Flow ingredients and Scenebuilder.

Fixing common problems

• The subject morphs. Ask for smaller movement, keep the camera still, and check the frame for flaws.
• A cross-fade instead of motion. The start and end frames are too different. Make the end frame by editing the start frame.
• Unwanted music or captions. Add "no music" and "no on-screen text" to the sound sentence.
• The wrong model ran. Flow may switch to a compatible model if the one you chose does not support a feature; check the model name before generating.

Try it

The ceramicist lifts the bowl from the wheel and turns it slowly towards the window light. Static camera, shallow depth of field. Sound: a soft studio hum, clay squeaking under her fingers, no music.

Use a start frame of a ceramicist at the wheel made with Nano Banana. Generate 8 seconds on Veo 3.1 Fast, then add an end frame of the bowl held up to the light and compare how much tighter the motion becomes.