Verified against Kling 3.0 · 2026-08-08
Generate a UGC-style testimonial clip that doesn't read as AI-made
A Kling 3.0 brief for a phone-shot-looking testimonial clip — deliberately imperfect framing and natural delivery instead of the polished symmetry that immediately flags a video as generated.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Subject and action: a woman in her late 20s in a hoodie, sitting on a couch holding a a small ceramic travel mug at chest height, speaking directly to camera, reacting genuinely surprised at how long the coffee stays hot, pausing naturally between phrases as if thinking of the next sentence rather than reading a script. Camera: static handheld phone-style shot, held slightly off-center and tilted 2-3 degrees, a slight autofocus hunt in the first second present throughout — do not correct or stabilize it. Environment and lighting: a cluttered home office with laundry visible in the background, single practical light source (window or lamp), visible minor exposure flicker, no studio evenness. Style and mood: casual first-person UGC, unpolished and sincere, not commercial-grade. Audio: room tone with slight ambient noise from the setting, voice slightly close-mic'd and a little uneven in volume, no music, no professional voiceover clarity. Do not: center the subject perfectly, smooth out the camera wobble, or add cinematic color grading — any of these will make the clip read as an ad rather than UGC.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Kling 3.0's default tendency, like most video models, is to resolve ambiguous framing toward symmetry and smooth, studio-even motion — which is exactly the visual signature that makes a generated clip read as an ad rather than user-generated content. This prompt fights that default by naming a specific imperfection (a tilt in degrees, a described autofocus hunt) and explicitly instructing the model not to correct it, because a vague instruction like "make it look like UGC" gives the model nothing concrete to hold onto and it will default back to its cleanest interpretation. Describing the delivery as a paused, thinking-through-it reaction rather than a scripted line also matters mechanically: Kling's audio-visual sync is built around natural speech cadence, and feeding it a beat-based description ("reacting genuinely surprised") produces more organic mouth movement and timing than a quoted script line would, which tends to generate over-articulated, presenter-style delivery instead. The explicit do-not list at the end functions as a guardrail against the model's own smoothing bias re-asserting itself over the course of generation.
What you get back
A 7-second handheld clip: a woman on a couch holds up a ceramic mug, camera slightly tilted and hunting focus for the first second, she pauses mid-sentence then says the coffee's still hot with a small surprised laugh, uneven window light, room tone audible, no music.
Verified against
Kling 3.0 3.0 · 2026-08-08
Changelog
- 2026-08-08 — Initial publish, verified against Kling 3.0.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
