Three Meters Above the Sky AI Prompt: Copy and Use

Use this prompt as a starting point for the Te He Perdido Entre La Gente scene: one person watches from an interior balcony while another dances on the floor below. It is written for a video tool that accepts a reference clip and two image references. If your tool uses different reference slots, follow its current interface and map each image to the matching role.

Try the two-photo AI video generator

Copyable prompt

Create an original, photorealistic vertical video. Use the reference video only for camera movement, pacing, and general atmosphere; do not reproduce an exact movie frame or imitate identifiable actors.

Use Image 1 for the adult person standing on an interior balcony, looking down. Use Image 2 for the adult person dancing on the crowded floor several meters below. Keep both people in their assigned positions. Preserve their distinct identities, faces, hair, clothing, and accessories throughout the video. Never swap or merge their faces.

Set the scene in a lively nighttime venue with warm lights, a visible upper balcony, and a dance floor clearly below it. Show a natural, nostalgic moment as the person above watches the person dancing below. Use a smooth cinematic camera move that connects both levels while keeping the height difference easy to understand.

Create a new scene rather than copying a film still. Use natural expressions and realistic movement. Do not add a commercial song, copyrighted dialogue, logos, or text. Output one polished 9:16 MP4 video.

How to adapt the prompt

1. Replace the scene details only when your video tool supports them; keep the balcony, dance floor, and vertical relationship explicit.
2. Assign one clear photo to each person. For this prompt, Image 1 is the balcony person and Image 2 is the dancer.
3. Keep identity instructions short and direct: preserve each face and do not swap the two people.
4. Add duration or resolution only if the tool has those controls. Do not repeat settings in the prompt when the generator already applies them.
5. Generate a short preview first when the tool allows it, then adjust the reference images or framing if needed.

Two-person example

For a couple, use one clear photo of each adult and keep the same role mapping: the first person watches from the balcony, and the second dances below. For a friends version, leave the scene unchanged and replace only the relationship or mood description. Clear reference photos usually matter more than adding a long list of style words.

How image-to-video workflows differ

Video tools do not all handle references in the same way. Some provide separate character or image-reference slots, while others rely on a starting frame or a general prompt. Follow the labels in the tool you are using instead of assuming every uploaded image has the same role. This site's generator uses a reference video for motion and atmosphere, plus two photos for the people being replaced.

The prompt describes the scene; the generator's settings control supported options such as aspect ratio, duration, and resolution. Check the displayed credit estimate before creating a video.

Common fixes

• The faces are swapped: make the image order explicit and use two distinct, unobstructed photos.
• One person appears in both roles: repeat that each reference image belongs to a different person and keep the roles separate.
• The balcony is hard to see: describe the balcony as an upper interior level and the dance floor several meters below.
• The video looks too much like a film still: ask for an original scene, new framing, and generated background details.
• The result has no trend song: add audio later from the social platform's own library and check its availability for your account.

FAQ

Can I paste this into any AI video tool?

You can adapt the wording, but reference-image support and prompt controls vary by tool. Map each photo to the correct role using the controls currently shown by that tool.

Does the prompt guarantee the same result every time?

No. Video generation is variable. The prompt and photos guide the result, but movement, framing, and small visual details can change between generations.

Can I use photos of someone else?

Only with permission. Avoid sensitive images, review the generated video before sharing, and do not imply that someone took part in a real event when they did not.