FLUX 3 Max Audio: Sound and Dialogue Prompt Guide

FLUX 3 Max audio is always on: every clip comes back with ambience, effects, music and any spoken lines you write, generated in the same pass as the picture. Direct it from the prompt: dialogue in double quotes with who says it, two or three words per second, the ambience named, and "no music" when you mean it.

For ad work, that changes the order of production. There is no silent clip waiting for a sound pass and no hunt for stock ambience. The sound arrives with the picture, so the prompt is where your audio decisions get made. This guide covers where sound goes in a prompt, how to write lines the model will speak, how to direct ambience and effects, when to ask for music, and what to do when you need a different track.

How does FLUX 3 Max audio work?

Yes, on every clip. FLUX 3 Max is a video model from Black Forest Labs, and on VIDEO AI ME it generates sound in both modes, text to video and image to video. The audio covers ambience, effects, music and any spoken lines you write, all produced in the same pass as the picture.

If your job depends on your own voiceover driving the picture, compare the options in FLUX 3 Max vs Boreal. For everything else, the prompt is your mixing desk.

Where does sound go in a FLUX 3 Max prompt?

Last. The method is one paragraph per shot, in four parts:

1. Subject and action. Who or what, doing what. One focused beat.
2. Camera. Framing and movement.
3. Light and look. Time of day, light source, grade.
4. Sound. Spoken lines in double quotes with who says them, then ambience and effects, then music or "no music".

Here is the full example for a 9:16, 10 second text to video clip:

A woman in her thirties holds a glass bottle of cold brew up to the camera in
a sunlit kitchen, then takes a sip and smiles. Handheld, eye level, slight
push-in. Warm morning light from a window on the left, soft natural grade.
She says, "This is the only coffee I make at home now." Quiet kitchen
ambience, the clink of the bottle on the counter, no music.

Sound goes last for a practical reason: by then the picture has established the people and the objects, so the sound can refer to them. "The clink of the bottle on the counter" only works because the bottle and the counter are already in the shot. Inside the sound part, keep the order fixed, line first, then ambience and effects, then music, so nothing gets left undecided.

The rest of the method, camera and light included, is in the FLUX 3 Max prompt guide.

How do you write dialogue FLUX 3 Max will speak?

Quote it and say who speaks. Write She says, "..." or The man at the counter says, "...". A quoted line with a named speaker leaves no doubt about what is spoken and by whom.

Budget the words. Roughly two to three spoken words per second of clip, so a 10 second clip holds about 20 to 25 words. Write fewer than you think, because the action needs time too.

Write for the ear. Contractions, short sentences, one idea per line. People in ads do not speak in brochure copy, and a line that reads well on a page can sound stiff out loud. Say it with a timer before you generate.

Add a delivery cue when tone matters. "She says quietly", "he laughs and says", "she says to camera". The model expands whatever you leave open, so if the tone is part of the ad, write it.

Keep two speakers short. A two-person exchange in 10 seconds works best as two short lines, each with its own named speaker: The man asks, "Is that the new one?" The woman answers, "Since Monday."

Check names first. Brand names and unusual words are the first thing to listen for in a draft. A mispronounced product name is an error you cannot caption around.

For creator-style scripts written to these budgets, start from the FLUX 3 Max UGC prompt templates.

How do you prompt ambience and sound effects?

Write the ambience even when it feels obvious. A kitchen sounds like a kitchen to you, but if you leave the ambience open, the model decides for you. "Quiet kitchen ambience", "a busy café, low chatter", "wind across an open field".

Name the sound on the action. Tie each effect to something the picture shows, so it lands on the right moment: "the pump clicks as her thumb presses down", "the zipper rasps as he opens the bag".

Use placement cues. Position sounds in the scene: "traffic on the left, a kettle on the right". It helps the sound feel like it belongs to the space on screen.

The same rules apply in image to video. The image already carries the look, so the prompt is about motion and sound:

The bottle slowly rotates on the marble counter as condensation runs down the
glass. The camera makes a slow half-orbit. A crisp cap twist, a soft fizz,
light lo-fi music.

Should you ask for music?

Always decide, one way or the other. Write the music you want, or write "no music". Leaving it out means the model decides.

Ask for music when the clip has no dialogue and needs energy, such as a product reveal or a cinematic opener. Describe it briefly and give it a level: "light lo-fi music", "soft piano under the voice", "a slow, warm synth pad".

Write "no music" when a spoken line carries the ad and must be clear, when the clip is meant to feel like a creator filmed it on a phone, or when the clip is one of several you will join. Music generated separately for each clip will not carry across the cuts, so "no music" in every prompt keeps the joins clean.

How do you replace FLUX 3 Max audio?

Audio is always generated, so there is no setting that produces a silent clip. To replace it, swap in your own audio afterwards.

Plan for this in the prompt:

• If you will replace the audio with a voiceover, do not write dialogue for an on-screen person. Their mouth would move to words you then remove.
• Keep the sound part short when you know it will go. "Quiet room tone, no music" is enough. Nothing in the picture should be performing to sound you plan to delete.
• If the person on screen must speak your exact recorded voice, that is a job FLUX 3 Max does not do on VIDEO AI ME. Boreal takes an uploaded audio track.

What do sound prompts look like for common ad jobs?

UGC testimonial, 9:16, 10 seconds, text to video:

A man in his twenties sits in his car in a parking lot and holds up a protein
bar to the camera, then takes a bite. Handheld selfie angle, eye level. Soft
overcast daylight through the windshield, natural grade. He says, "Twenty
grams of protein and it actually tastes like a cookie." Muffled traffic
outside, the wrapper crinkles as he opens it, no music.

SaaS desk scene, 16:9, 12 seconds, text to video:

A woman in her thirties at a standing desk turns from her monitor to the
camera and smiles. Locked-off medium shot, eye level. Cool daylight from a
window behind her, clean neutral grade. She says, "I used to spend Mondays on
this report. Now it is done before my coffee." Low office hum, a few keyboard
taps, no music.

Cinematic opener, 21:9, 20 seconds, text to video:

A lone cyclist rides along an empty coastal road at dawn as waves break below
the cliffs. Slow drone-style tracking shot from the side, wide. Pale blue light
turning gold on the horizon, filmic grade. No one speaks. Soft gusts of wind,
waves on the left, the steady tick of the bike's freewheel, a slow
ambient score building gently.

Each one follows the same order, and each one decides the line, the ambience, the effects and the music.

Why does sound matter on paid placements?

Because many placements play it. Meta's Reels ads guidance states that Reels default to sound on, and that ads play with sound on unless the user has turned the sound setting off. Meta also lists quality audio among its Reels creative recommendations. A clip that arrives with designed sound is ready for that placement without an extra pass.

Two cautions. Add captions in the editor anyway, for viewers who watch muted. And if a generated person speaks a testimonial-style line, read the FTC's guidance on endorsements, which says an endorsement must reflect the honest opinion of the endorser, along with our guide to disclosure for AI UGC ads.

What should you check before a clip goes live?

• The spoken line matches what you wrote, word for word.
• The lip sync holds up. Watch the generated speech and lip sync closely before any clip goes live.
• Brand and product names are pronounced correctly.
• The music is what you asked for, or absent if you wrote "no music".
• Effects land on the actions they belong to.
• Captions match the words actually spoken.

Draft at 720p, where a 10 second clip costs 170 credits, and deliver at 1080p for 290 once the sound is right. Plan allowances are on the pricing page.

Frequently Asked Questions

Does FLUX 3 Max generate audio?

Yes. On VIDEO AI ME, every FLUX 3 Max clip comes back with sound generated in the same pass as the picture: ambience, effects, music and any spoken lines you write into the prompt. This applies to both text to video and image to video.

Can I turn off the audio in FLUX 3 Max?

Not at generation. Audio is always generated. If you need a different soundtrack, replace the track with your own audio afterwards (your track replaces the generated sound rather than mixing with it), and avoid writing dialogue for an on-screen person if you plan to replace the sound.

How do I make a FLUX 3 Max character say a specific line?

Write the line in double quotes and name the speaker, for example: She says, "This is the only coffee I make at home now." Put it in the sound part of the prompt, after the subject, camera and light, then check the speech and lip sync in the draft.

How many words of dialogue fit in a FLUX 3 Max clip?

Roughly two to three spoken words per second, so a 10 second clip holds about 20 to 25 words and a 5 second clip about 10 to 12. Write fewer than you think, because the action needs time as well.

Can I upload my own voiceover to FLUX 3 Max?

No. FLUX 3 Max on VIDEO AI ME generates its own audio and does not take an uploaded track. Boreal takes an uploaded audio track. You can also replace the FLUX 3 Max track with your own audio afterwards.

Will FLUX 3 Max add music to my clip?

It can, so decide in the prompt. Describe the music you want, such as light lo-fi music, or write "no music". If you leave it out, the model decides for you.