01 Core Formula
FLUX 3 reads plain language and expands it before generation — you don't need keyword stuffing, you need a clear sentence. This chapter builds the prompt skeleton first, then walks through two practical dials: the audio switch, and how to pick resolution and duration.
1.1 The Basic Prompt Formula
Everything starts with one sentence: subject plus motion are the two required parts; camera, environment, aesthetics and sound stack on demand. FLUX 3 rewrites your prompt internally, so write what you want to see and hear in full sentences — vague keywords get expanded in directions you didn't ask for.
[Subject] + [Motion] + optional [Camera] [Environment] [Aesthetics] [Sound]
Subject
The protagonist — a person, animal or object. The more specific, the better.
Motion
What the subject does, with pace spelled out. Slow, coherent movements are the most reliable.
Camera
Shot size and movement: close-up, push-in, orbit. FLUX 3 follows explicit camera language well.
Environment
Scene and atmosphere: place, time, light, weather.
Aesthetics
Art direction: footage style, color tone, texture — home video to cinematic.
Sound
What you want to hear: ambience, effects, dialogue. Native audio is the headline feature — chapter 2 goes deep.
The Rooftop Gardener
Prompt
An elderly man in a straw hat waters tomato plants on a city rooftop at sunrise, steam rising gently from the soil. The camera slowly pushes in from a wide shot of the skyline to a close-up of water droplets on the leaves. Warm golden light, documentary texture. Soft birdsong and distant traffic hum in the background.
Subject+motion in the first sentence, camera and environment in the second, aesthetics and one line of sound to close — a solid FLUX 3 baseline.
1.2 The Audio Switch: Finished Cut or Clean Plate
FLUX 3 generates synchronized audio by default at no extra cost: ambience, effects and speech arrive with the picture. Switch it off and you get a silent clip — the clean plate for jobs that go into an editing pipeline with their own music, voiceover and subtitles. On for direct publishing, off for post-production.
Rainy Crosswalk (Silent Plate)
Prompt
A busy city crosswalk in the rain at night, umbrellas streaming past in both directions, neon reflections rippling on the wet asphalt. The camera tracks sideways at knee height, following a pair of red sneakers through the crowd. Cinematic contrast, cool blue tones with warm highlights.
The audio switch is off in this preload — take it as footage and lay your own score over it, with no model-generated ambience to wash out.
1.3 Resolution & Duration: 720p to Explore, 1080p to Deliver
Two resolutions (720p / 1080p) and a continuous 5–20 second dial. The veteran's workflow: explore at 720p — once the composition and pacing check out, rerun the same prompt at 1080p for the final cut. Match the action load to the seconds: one clean beat per 5 seconds is the reliable rhythm; 20 seconds can carry a miniature story.
A Letter in Twenty Seconds
Prompt
A young woman sits by a train window at dusk, unfolds a handwritten letter, reads it, and begins to smile through tears. The camera starts on the landscape rushing past the glass, slowly shifts focus to her face, then settles on the letter trembling slightly in her hands. Soft ambient train sounds, a distant announcement chime. Intimate indie-film tone, warm interior light against the blue evening outside.
Pre-filled with 20s at 1080p — a three-beat micro-story (unfold, read, react) needs the full dial; for iteration, drop to 720p first and come back for the final.
02 Native Audio Prompting
FLUX 3's headline: audio is generated in the same pass as the video — dialogue lip-synced to the speaker, effects pinned to the physical events that cause them. That means sound belongs inside the prompt, not in post. This chapter covers the two writing techniques: dialogue and sound effects.
2.1 Writing Dialogue: Quoted Lines + Tone Direction
Put the spoken line in quotes, then direct the delivery in one clause: tone, pace, emotion. The model lip-syncs automatically and handles multiple languages. Keep lines short — one or two sentences per 10 seconds; a monologue crammed into 5 seconds forces rushed speech.
The Night-Market Vendor
Prompt
A cheerful street-food vendor at a night market flips scallion pancakes on a sizzling griddle, looks up at the camera and says with a warm grin: "Best pancakes on this street — crispy outside, soft inside, try one and you'll be back tomorrow!" Steam rises around him, neon signs glow in the background, the crowd murmurs softly. The camera holds a medium close-up, handheld documentary feel.
Quoted line + "with a warm grin" delivery direction, ambience in one closing phrase — that's the complete dialogue recipe.
2.2 The Sound-Effect Formula: Pin Sound to Events
Sound effects land best when each one is pinned to a visible physical event: write the event, then its sound, in the same breath. Vague requests like "add nice sounds" produce generic mush; "the ladle clangs against the wok" gives the model a target to sync to.
[Visual event] with [its sound] + optional [ambient bed]
Visual Event
The on-screen action that makes noise: a door slams, glass clinks, footsteps splash.
Sound Cue
The effect pinned to that event — named concretely, in the same sentence as the action.
Ambience
The background bed: wind, crowd murmur, room tone. One phrase is enough.
The Glassblower's Workshop
Prompt
In a dim workshop, a glassblower pulls a glowing orange blob from the furnace with a soft whoosh of heat, rolls it on the steel table with a low metallic scrape, then blows steadily into the pipe — the glass swells with a faint crackle. Sparks drift in the background, the furnace hums constantly. Slow push-in on the molten glass, warm amber glow against deep shadow.
Every sound is welded to its event — whoosh, scrape, crackle each have a visible cause, and the furnace hum is the ambient bed.
03 Image to Video
One first-frame image plus a motion description, and FLUX 3 brings the picture to life — the scene already exists, so the prompt's center of gravity shifts to motion and sound. A first frame sets the opening; first-last frame owns both ends of the shot.
3.1 Bringing a Still to Life
An image-to-video prompt covers three things: what moves, how it moves, and what it sounds like. Don't restate what's already in the picture — wake it up instead: drifting mist, rippling water, flickering light are high-hit-rate small motions.
Morning on the Lake
Prompt
Bring this photo to life: thin mist drifts slowly across the still lake, the wooden rowboat rocks gently, and a heron takes off from the pier with a slow flap of wings and a soft splash. The camera holds almost still, then drifts forward a touch. Quiet morning ambience, distant water lapping.
Upload 1 first-frame image; the prompt writes only motion and sound — the scene itself is already in the picture.
3.2 First-Last Frame: Own the Start and the End
Upload a first frame and a last frame, and the model generates everything between them. Prompting changes accordingly: skip the start and the end (they're in the images) and write only the journey — the pace of change and how the camera plays along. The bigger the visual gap between the two frames, the more process description you should provide, and the longer the duration you should allow.
The Paper Plane's Flight
Prompt
The paper plane leaves the child's hand, wobbles, catches an updraft, and glides smoothly across the courtyard toward the open window, climbing gently as it goes. The camera tracks alongside in one continuous follow shot. Soft afternoon breeze, leaves rustling, a child's delighted laugh off-screen.
In image-to-video mode, upload a first frame (the throw) + a last frame (the window); 8 seconds gives the flight room to breathe.
04 Storyboards & Clip Extension
This chapter covers FLUX 3's two timeline-control tools, which live in two different places. Storyboard keyframes run in reference mode: upload 3–10 images and they become keyframes — the first opens the clip, the last closes it, the rest land evenly in between. Clip extension is its own model, FLUX 3 Extend: hand it one video (up to 15 seconds) and it continues the shot from its final frames.
4.1 The Storyboard: Direct With Keyframes
A storyboard prompt narrates the beats in order — the images fix the look of each beat, your words direct how one beat flows into the next. Three to ten images; an explicit duration is required so the model can place the keyframes on the timeline. Keep beats proportional to the seconds: two beats in 5 seconds, three to four in 10–15.
[Beat 1 → Beat 2 → Beat 3] + optional [Pacing] [Continuity]
Beats
One clause per keyframe, in upload order: what this moment is and how it moves.
Pacing
How beats transition: slow dissolve, match cut, continuous motion — just name it.
Continuity
A closing line that locks light, palette and mood across the beats.
Espresso in Three Beats
Prompt
Three beats of one espresso: whole beans tumble into the grinder, fine grounds cascade into the portafilter and are tamped level, then dark espresso streams into the white cup with rising crema. Each beat flows into the next with a gentle match on motion. Warm café light, rich brown and cream tones, the grinder's buzz giving way to a quiet pour, consistent mood throughout.
Upload 3 reference images (beans / tamping / pour) in storyboard order — the first opens the clip, the last closes it, the middle lands evenly between.
4.2 Extending a Clip: Keep the Shot Rolling
Switch to the FLUX 3 Extend model (video-extend mode), upload one video (up to 15 seconds, MP4) and FLUX 3 continues from its final frames — same scene, same motion, new content. The prompt describes only what happens next, not what already happened. Chain extensions to build sequences far beyond the 20-second single-generation limit. Note this is the premium lane: continuation costs more per second than fresh generation, so extend the takes that deserve it.
The Drone Shot Continues
Prompt
The camera keeps gliding forward over the coastline, rounds the cliff, and reveals a hidden cove with turquoise water, a lone sailboat anchored inside. The motion stays smooth and continuous, golden-hour light consistent with the incoming shot. Gentle waves and wind, a distant gull.
Upload your source clip (the cliff approach) to the source-video slot; write only what happens next — the model owns continuity with the footage you gave it.