AI Music Video Loops

Overview

Build prompts for AI video tools that produce visually hypnotic short loops for music — atmospheric, immersive, seamlessly repeatable. The core principle: don't write a scene where things happen. Write a scene that exists and breathes.

Output target: 3–5 seconds. Motion is ambient, environmental, emotionally stable, and loop-friendly.

When to use this skill

• The user wants a video prompt for music — looped backdrop, music video B-roll, visualiser, social-media background.
• The user mentions Sora, Runway, Veo, Kling, Pika, Luma, or "AI video".
• The user asks for hypnotic / atmospheric / ambient short visuals.
• The user wants the same vibe rendered across multiple style variations.

Do not use when:

• The user wants a full narrative video script with scenes and dialogue — use a screenwriting approach instead.
• The user wants a single still image — use an image-prompt skill.
• The user wants a long-form video (>10 sec) with story progression.

Instructions

Core philosophy

The loop has to feel:

• ambient
• environmental
• continuous
• emotionally stable
• loop-friendly

It must not feel like a cinematic sequence, a short movie, or an action scene.

Motion to use

rain · snowfall · fog drifting · smoke · floating particles · dust · cloth movement · light flicker · reflections · water ripples · fire embers · slow moving shadows · subtle camera drift · steam · clouds · neon glow pulses · CRT flicker.

Motion to avoid

walking · dancing · talking · dramatic gestures · fast camera movement · turning around · scene transitions · object transformations · sudden lighting changes · story progression.

Camera

Use: locked camera · slow dolly · subtle push-in · tiny cinematic drift.

Avoid: handheld · aggressive movement · spinning · fast zooms.

Layering principle

Every good loop contains three depth layers. Layered motion hides loop seams naturally.

1. Foreground motion — rain, particles, fog, smoke, embers.
2. Midground focus — the main subject or environment.
3. Background depth — distant lights, buildings, haze, stars, mountains.

Prompt structure

Order matters. Build every prompt in this order:

[environment + mood],
[subtle environmental motion],
[lighting style],
[layered depth],
[camera behavior],
[loop-friendly atmosphere],
[minimal motion],
[cinematic quality]

Universal loop phrases (mix 3–5 per prompt)

subtle atmospheric motion · seamless loop feeling · ambient movement · hypnotic repetition · environmental motion only · emotionally stable scene · cinematic depth · layered atmosphere · minimal movement · slow breathing environment · static composition · dreamy continuous motion · immersive ambience · loop-friendly motion · no dramatic action.

Output format

Return a single prompt per requested style as a comma-separated descriptor block, ~30–50 descriptors. If the user asks for multiple styles, return one labelled block per style.

End every prompt with a loop-friendliness phrase (e.g. seamless loop feeling or loop-friendly motion) so the AI generator biases toward repeatable output.

Examples

Dark Ambient

abandoned cathedral interior, drifting fog, floating dust particles,
candle flicker, layered shadows, subtle atmospheric movement, locked
camera, cinematic depth, hypnotic ambience, seamless loop feeling,
minimal motion, emotionally stable scene

Lofi

cozy bedroom during rain, steam from coffee cup, soft monitor glow,
subtle curtain movement, rainy window reflections, warm cinematic
lighting, ambient atmosphere, slow environmental motion, static
composition, seamless loop feeling

Synthwave

neon highway at night, rain reflections on asphalt, drifting fog,
distant car lights, subtle glow pulses, cinematic depth, minimal
camera drift, hypnotic atmosphere, environmental motion only,
seamless loop feeling

Tribal

ancient ritual fire in desert night, floating embers, smoke drifting
in wind, cloth movement, atmospheric dust particles, cinematic
shadows, slow ambient movement, layered depth, immersive loop feeling

Notes & limitations

• 3–5 seconds is the sweet spot. Shorter feels too cut; longer drifts out of the layered-motion zone and the AI introduces unwanted change.
• One subject, no character action. A figure in the scene is fine if they're nearly still (looking out a window, sitting by a fire). Walking or gesturing breaks the loop.
• Three depth layers is non-negotiable. Single-layer scenes look flat and the loop seam shows. Always foreground + midground + background.
• Tool differences exist. Sora and Veo handle longer loops cleanly; Runway and Pika favour 3–4 sec; Kling is strong on environmental motion. If the user names the tool, lean into its strengths.
• No transitions, no cuts. A loop is one continuous take. Anything that implies a cut (different angle, different time) will break it.

Changelog

• 0.1.0 — initial version.