First, check the video path available to you
Google’s current video help describes a Create video entry in the desktop sidebar, reference inputs, conversational edits, and downloading. Access depends on account eligibility, and some features have regional restrictions. Check that page and your own interface before planning around a particular input or output length.
Open Gemini and choose your next step:
- You can enter a video request: continue with the scene below.
- You only see a text conversation: resolve access first. A written description is useful preparation, but it is not your finished clip.
- A reference input you wanted is unavailable: simplify the plan to the inputs actually offered.
Keep the first attempt small enough to judge. One person, one float, one helping hand, and one continuous view give you a manageable starting point.
Give every reference a separate job
Choose a portrait for appearance. The face should be recognizable at a glance, without sunglasses hiding the eyes or another person competing for attention. A relaxed expression gives you room for a small reaction later. Use an adult’s photo you have permission to upload; the photo guide explains how to assess the face and crop.
A separate scene reference, if you use one, has a different job: showing the pool edge, camera position, and space around the float. An original layout sketch can communicate those relationships without asking the model to copy someone else’s finished video.
Avoid a pile of attractive but contradictory references. A portrait taken from below, a pool photographed from overhead, and a dramatic sideways camera angle leave several incompatible directions to resolve.
Write a short note explaining your priorities: “Use the portrait for the person’s appearance; use the pool sketch for the layout.” Treat that as a request, then inspect the result. References do not guarantee that every feature will remain unchanged.
Describe the contact before describing the reaction
The helping push is the hinge of this scene. If the float reverses before the hand arrives, the joke feels disconnected. If the person reacts before anything happens, the sequence loses its surprise.
Write the action in order: approach, contact, push, reaction, drift. Decide where the helper’s hand enters and what it touches. Keep the far edge of the pool visible enough to establish the setting, while leaving room beside the float for that contact.
For this Flamingo Float AI Gemini guide, the ending stays inside the pool. The float turns away and continues drifting. A fall or a flip would require a different setup; mixing those endings into one request makes the intended outcome harder to judge.
Start with this original scene prompt
Create a vertical video in one continuous shot. One adult from the portrait reference reclines on a pink flamingo inflatable in a rooftop infinity pool above a generic city skyline. The person glances at a phone while the float drifts slowly toward the visible pool edge. Keep the person, float, and waterline in view. A single helping hand enters from the near side, visibly touches the inflatable’s outer rim, and gently pushes it away from the edge. Only after contact, the float changes direction. The person looks up with a brief, amused reaction, lowers the phone, and settles back as the float continues drifting inside the pool. Keep the camera steady and the movement gentle. Preserve the person’s recognizable appearance. Keep the city background secondary. No fall, flip, additional rider, scene cut, or written text.
Read it once before submitting. Remove anything that conflicts with your reference or preferred framing. For more ways to describe the action, visit the Flamingo Float prompt guide.
Fit the story to the length you can actually generate
Use the duration available in your account as a creative constraint. Give the opening only enough space to establish a relaxed person on the float. Reserve the clearest moment for the hand touching the rim, then leave enough of the ending to see the new direction.
If that feels crowded, remove the phone action first. The person can simply be looking away. Next, drop any separate skyline reveal. The city can already be present behind the pool.
Do not compress everything into simultaneous motion. A shorter version still needs a before and after: the hand reaches, the float moves, the person responds. A tiny change of expression is enough; a large performance is unnecessary.
Try this shorter opening instruction: “Begin with the float already approaching the edge and the person looking away.” The storyboard guide offers a fuller breakdown if you want to plan the beats visually.
Revise one thing in the conversation
Once you have a candidate, identify the single change that would make it clearer. Begin with the story’s cause and effect. A beautiful skyline cannot compensate for a push that never visibly happens.
These original follow-up requests each address one problem:
- Contact: “Keep the scene and camera position. Have the helping hand touch the outer rim before the inflatable changes direction.”
- Reaction: “Keep the push and drift. Make the person look up only after the hand has pushed the float.”
- Framing: “Keep the same sequence. Widen the view enough to keep the hand and the entire contact area visible.”
- Ending: “Keep the opening and helping push. End with the float moving away inside the pool, with the person still seated.”
Compare each revision with the previous clip. Asking to preserve something is not proof that it was preserved. Check the face, phone, and float shape again even when your request only concerned the camera.
If several details become worse, return to the simpler brief rather than adding a paragraph of corrections. Save a useful earlier version before experimenting. For a problem that keeps returning, use the visual repair guide to isolate what the viewer is actually noticing.
Review the result in separate passes
Watch once without pausing. Does the sequence make sense immediately? You should understand the approach and the helping push without needing a caption to explain missing action.
Then inspect three moments: just before contact, during the push, and after the reaction. Look for a hand that changes shape, a rim that bends abruptly, or a phone that merges with the person’s fingers. Check whether the waterline remains in a believable place around the float.
Finally, compare the person with the portrait. Pay attention to the whole appearance, including hair, clothing, and proportions. One convincing frame does not establish consistency across the clip.
Keep the version with the clearest action and most consistent subject. Your best candidate may have a quieter expression or a less elaborate skyline than you originally imagined.
Download, then watch the file you will share
Google’s help currently directs desktop users to Share → Download video beneath the result. Check the available controls in your own session.
Play the downloaded file from beginning to end. Listen to any included audio before adding music or dialogue. Match your caption to the ending that is actually visible, and make the AI creation context clear when you post.
Leave final trimming and overlays until you have chosen a candidate. Our editing guide explains how to preserve the contact and reaction during that last pass.
Use this Flamingo Float AI Gemini workflow when you want to direct the scene through prompts and revisions. For a prepared version, open the generator, watch its current sample, and review the displayed cost before starting.