Opening workflow…
Video & film · Intermediate
Dive from orbit into a café that comes alive, in one AI camera move
One unbroken first-person shot from low Earth orbit down to an espresso on a café table. The camera dives through a layer of golden cloud, onto a hillside town, between the buildings into a cobblestone lane, and along it to a restaurant, slowing only in the last second. When it lifts off the cup, the empty lane has come to life with diners, a waiter and people walking past. It is one free still and two Kling 4.0 generations, about 20 credits, joined at a near-still moment so you can't see the join. You get the exact prompts, the three rules that stopped our dive from morphing (one generation per camera move, no end frame, "we are the camera, no drone"), the FFmpeg commands to chain and join the clips, and the sound-effect prompts for the sound bed.
Before you start
Uses paid credits- You’ll need
- Google Flow
Higgsfield- ElevenLabs
- FFmpeg
- What it uses
- Paid generationsThe amount depends on your Higgsfield plan and how many takes you try.Higgsfield pricing
- The tutorial
- Free with an account · 4 steps
Tutorial outline
4 steps, in order. Open the tutorial for the instructions, prompts, settings and checks.
Make the orbit start image
Generate the dive in ONE generation
Pin the café reveal with an end frame, then generate it
Add sound and cut the Reel
Exactly what the creator used
- Google Flow
- Any recent version
- Higgsfield
- Any recent version
- ElevenLabs
- Any recent version
- FFmpeg
- Any recent version
Google Flow — Nano Banana Pro — the orbit start image (step 1) — free with a Google AI Pro/Ultra plan (ChatGPT's image tool works too) Higgsfield — Kling 4.0 (kling4_0), image-to-video — the dive (step 2) and the reveal (step 3) — about 1 credit per second at 1080p: 12 + 8 = 20 credits ElevenLabs — Sound Effects — the sound bed (step 4) — your ElevenLabs plan (optional: Kling's own audio works alone) FFmpeg — any recent build — last frame, join, mix — free Provenance: October 23, 2026. Orbit still from Nano Banana Pro in Google Flow. Dive: Kling 4.0 on Higgsfield, 12 s, 1080p, start frame only (12 credits). Reveal: Kling 4.0, 8 s, from the dive's last frame (8 credits; one earlier 8-credit take rejected). Sound: Kling native audio plus ElevenLabs Sound Effects, mixed in FFmpeg to -15 LUFS. Edit and encode in FFmpeg.