How to Dive Into One Eye and Come Out of Another With Google Flow (2-Clip Prompt Recipe)

November 14, 2026 14 visits

Camera moves · CM-002

Dive into one eye and come out of another world

Picture this: a woman has a golden desert reflected in her eye. A camera dives into her pupil, falls into glowing darkness, and comes out of the eye of a second person standing in that very desert.

  • DifficultyExtreme
  • Format9:16 Shorts and Reels
  • ToolGoogle Flow
  • Method2 clips, 3 images

The shot is coming soon

This video is AI-generated with Google Flow.

That is the shot we are making in Google Flow, two clips joined into one move. It is exactly the kind of “wait, how did they do that?” video we love to make. Below we go through it step by step, so you can see why every part is there. Every prompt is at the end, ready to copy.

The recipe

The idea in one line

Two clips + three matching images + one shared dark frame + a face reveal. That is the whole recipe. The steps below show how to put it into words.

01Two clipsThe dive and the reveal.
02Three imagesEvery clip starts on one.
03A shared dark frameBoth clips end on it.
04A face revealBack from warm light to a face.

Before you start

What you will learn

You will learn
  • How to link two clips with start images and a shared dark frame
  • How to keep a face under your control with your own pictures
  • How to hand over from one clip to the next on a shared frame
  • How to plant a small story question that the ending answers
What you need
  • Google Flow, with a Veo model
  • An image tool that makes 9:16 images
  • A few credits, so you can try more than one version
  • A video editor to join the clips

Flow’s features and settings change, so check your own screen as you follow along.

Step by step

How we build the shot

Step 01

Pick a shot with a hidden story

A camera dive into an eye is a great doorway, but a doorway alone is only a trick. We add a question: the first eye reflects a golden desert. The answer arrives at the end, when the camera comes out of the second eye and that person is standing in the very same desert. A small mystery like this gives people a reason to watch to the end.

Step 02

Plan two clips and one shared frame

Google Flow makes clips of at most 10 seconds, so we plan two clips. Clip 1 dives into the first eye. Clip 2 starts on that same dark and comes back out of warm light to the second person’s face. Both clips share the dark picture, so they join cleanly.

Step 03

Set up Flow for Shorts and Reels

Set the shape first: we make videos for Reels and Shorts, so we generate at 9:16. Use Frames to video, choose the highest-quality model you have, and set the clip length in Flow’s settings to the longest it offers. Never write seconds inside a prompt.

ModeFrames to video
Aspect ratio9:16
ModelHighest quality
Clip lengthLongest offered

Step 04

Make three images that start every move

Every clip starts and ends on pictures you made, so Flow has less to invent. That is three images: the first eye, a soft dark frame that both clips share, and a portrait of the second person. Make them all at 9:16.

Our thinkingUsing your own pictures at both ends keeps the faces under your control.

Step 05

Image 1: an eye with something wrong in it

Describe the first eye like a portrait, and plant the secret: a tiny desert reflected on its surface. Nothing else is in the frame, only the eye.

What we wrotea hazel-green iris with fine detail like autumn leaves, long dark lashes, soft natural skin texture, a tiny golden desert dune under a low sun reflected on the surface of the eye

We also say “the pupil exactly in the centre of the frame” so the dive has a clear target.

Step 06

Image 2: the shared dark frame

Image 2 is plain soft darkness with tiny warm lights and a few grains of sand. Clip 1 ends on it and Clip 2 starts on it, so it is the picture where the two worlds meet.

What we wroteSoft darkness filling the whole tall vertical 9:16 frame, tiny drifting points of warm golden light and a few drifting grains of golden sand scattered through it

Our thinkingA dark frame is easy to draw, so Flow does not have to draw two eyes and a world change at once.

Step 07

Image 3: the second person

Make the portrait of the second person with the eye in the exact centre. If you want your own face in the shot, upload your photo to your image tool as a reference first. We ask for the same face, hair and skin tone, a normal natural eye with no reflection on it, and the desert and low sun behind them. This is the answer to the question from the first eye.

What we wroteKeep the person's face, hair and skin tone exactly as in the photo and do not change their identity.

The desert behind them is the same desert that was in the first eye, so the story closes the loop.

Step 08

Clip 1: the hook

Start on the first eye and give the viewer one calm moment to notice the desert in the eye. Then dive.

What we wroteShe blinks slowly once. The camera then pushes smoothly and steadily straight into the pupil, the hazel iris sweeping past the edges of the frame, and the whole frame falls into soft darkness with tiny drifting points of warm light.

Start image 1, end image 2. We open with a slow blink because the first second decides whether people keep watching.

Step 09

Clip 2: back from the warm dark to the face

Start on the shared dark frame. A new pupil appears and widens, and the camera pulls back out of it to reveal the second person. Because Flow can invent a new face here, we repeat the identity words in the prompt.

What we wrotethe exact person in the last frame of this clip, the same face, hair, skin and scarf, not a new person

Start image 2, end image 3. If Flow still draws a new person, see the fix below.

Step 10

Join the clips and keep the sound running

Put Clip 1 first and Clip 2 second in your editor. They meet on the dark of Image 2. Keep the sound running across the join, and trim a frame or two if the join looks jumpy.

Step 11

Generate a few versions and pick the best

AI video is rarely one and done. Generate a few versions of each clip, keep the best of each, and join them. If the second face does not look like your photo, remake Image 3 from a clearer front-facing photo first.

If it goes wrong

Ideas to try, one change at a time

The second face does not look like your photo
Remake Image 3 from a clearer, front-facing photo with even light and no sunglasses, and check that it is in the end-frame slot.
Flow draws a new person in Clip 2
The start frame is usually followed more closely than the end frame, so use the fallback: generate Clip 2 as a dive that starts on your portrait (Image 3 as the start frame, Image 2 as the end frame) with the fallback prompt below, then reverse the clip in your editor, mute its sound and add desert wind.
Clip 2 never reaches the face
Shorten the prompt: drop the sand and wind words and keep only “pulls back out of the pupil and reveals the face”.
Clip 1 never reaches the darkness
Remove the blink and keep only the push-in.
The eye changes shape
Remake the image with a simpler, sharper eye.
The join between the clips jumps
Use exactly the same Image 2 file for both clips, and trim a frame at the join.
The move is too fast or shaky
Add “very slow, steady, smooth” before “pushes”.
Black bars, or a wide frame inside the vertical video
Check that Flow’s aspect ratio is 9:16 and that all three images are 9:16.
Music sneaks in anyway
Keep the sound line, add the negative prompt, and remove the music in your editor.

Copy and paste

The full prompts

Make the three images at 9:16 first. Then in Flow choose Frames to video for each clip, set the start and end images shown in each box title, choose the longest clip length, and paste the clip prompt.

Image 1 · 9:16
Photorealistic cinematic extreme close-up of an adult woman's open eye, seen from the front, the eye filling the tall vertical 9:16 frame, the pupil exactly in the centre of the frame, a hazel-green iris with fine detail like autumn leaves, long dark lashes, soft natural skin texture, a tiny golden desert dune under a low sun reflected on the surface of the eye, shallow depth of field, fine film grain, no text.
Image 2 · 9:16 · shared dark frame
Soft darkness filling the whole tall vertical 9:16 frame, tiny drifting points of warm golden light and a few drifting grains of golden sand scattered through it, a faint amber glow at the very edges of the frame, photorealistic, cinematic, no text, no people.
Image 3 · 9:16 · upload your photo first
Use the uploaded photo as the reference. Keep the person's face, hair and skin tone exactly as in the photo and do not change their identity. Create a photorealistic cinematic close portrait of this person looking straight into the camera in a tall vertical 9:16 frame, one open eye exactly in the centre of the frame, wearing a sand-coloured scarf, a normal natural eye with no reflection on it, a vast golden desert dune and a low sun behind them, windblown sand drifting through the air, warm golden-hour light, shallow depth of field, fine film grain, no text.
Clip 1 · Image 1 to Image 2
A single continuous camera move in a tall vertical 9:16 frame. An extreme close-up of an adult woman's open eye, a tiny golden desert reflected in the hazel iris. She blinks slowly once. The camera then pushes smoothly and steadily straight into the pupil, the hazel iris sweeping past the edges of the frame, and the whole frame falls into soft darkness with tiny drifting points of warm light. One unbroken move, no cuts, steady, photorealistic, cinematic. Sound: no speech and no background music. Soft rain on glass and a slow heartbeat; a deep rising whoosh as the camera enters the pupil; then near silence.
Clip 2 · Image 2 to Image 3
A single continuous camera move in a tall vertical 9:16 frame. Out of soft darkness with tiny drifting points of warm golden light, a small round pupil appears in the centre and slowly widens. The camera pulls smoothly and steadily back out of it, the iris sweeping outward, and the skin and a sand-coloured scarf come into view as it reveals the face of the exact person in the last frame of this clip, the same face, hair, skin and scarf, not a new person, looking straight into the camera in warm golden-hour light, a vast desert dune and a low sun behind them with windblown sand drifting through the air. The camera settles on the face. Steady, one unbroken move, no cuts, photorealistic, cinematic. Sound: no speech and no background music. A reverse whoosh; warm open desert wind and drifting sand.
Clip 2 fallback · only if the person changes · Image 3 to Image 2, then reverse it
A single continuous camera move in a tall vertical 9:16 frame. The camera pushes smoothly and steadily straight toward the eye of the person in the first frame, keeping their face exactly as in the first frame. The skin and the scarf sweep out of view as the eye grows, the iris fills the frame, and the pupil opens until the whole frame falls into soft darkness with tiny drifting points of warm light. The person keeps looking straight into the camera. Steady, one unbroken move, no cuts, photorealistic, cinematic. Sound: no speech and no background music, only a faint soft wind.

Only if your Flow screen has a negative prompt box, you can add this to both clips:

Optional negative prompt
a different person, a new face, a different woman, cuts, scene changes, text, letters, logos, extra eyes, mismatched eyes, a distorted pupil, eyelid glitch, camera shake, fast motion, cartoon, low resolution, black bars, a wide letterbox frame, background music

Your turn

Which impossible shot should we make next?

Try it, change the two worlds and the hidden detail, and tell us what you get. Tell us in the comments, and follow us on Reels and Shorts for more impossible camera moves.

This video was made with AI (Google Flow).