FLUX 3 Max Prompt Guide: The Four-Part Paragraph

A good FLUX 3 Max prompt is one plain-language paragraph per shot, written in a fixed order: subject and action, camera, light and look, then sound. This FLUX 3 Max prompt guide shows how to write each part, how much dialogue fits a clip, how the model expands short prompts, and how prompting changes for image to video.

FLUX 3 Max is a post-trained variant of FLUX 3 Video, the video model from Black Forest Labs, the team behind the FLUX image models. Its developers describe it as tuned for stronger prompt adherence and better visual quality. If adherence is the headline, the prompt is your main lever. And because every clip comes back with sound generated in the same pass, the prompt directs the audio as much as the picture.

What is the best FLUX 3 Max prompt structure?

Write one paragraph per shot, in this order:

Here is the full pattern 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.

The spoken line is ten words, which leaves the 10 seconds room for the sip, the smile and the clink.

The order is deliberate. The subject and action come first because they are what the shot is about, and framing, movement and style wrap around them. Sound goes last because it hangs off the action.

Duration, aspect ratio and resolution stay out of the prompt. You set them in the VIDEO AI ME editor: 5, 8, 10, 12, 15 or 20 seconds, 720p or 1080p, and auto, 9:16, 16:9, 1:1, 4:3, 3:4 or 21:9. Match the ratio to the placement: Google's video ad specs for YouTube recommend 16:9, 9:16 and 1:1, with 1080p as the recommended standard.

How do you write the subject and action?

Name one subject and give it one focused beat: she lifts the jar and twists the lid, a sneaker drops onto wet tarmac. Twenty seconds is room for one beat done well, not a montage.

• Use verbs, not adjectives. "A glowing, premium serum bottle" describes a still image. "A hand tips the serum bottle and a single drop gathers at the dropper tip" describes motion, which is what a video model renders over time.
• Write the details that matter to you. Age range, clothing, the product's colour and material, the room. Anything you leave open, the model decides.
• One scene per clip. If the idea needs a kitchen, a car and a gym, that is three clips, each with its own prompt, joined in the editor.

A weak subject line reads "a man using a protein shaker, energetic". A usable one reads "a man in a grey hoodie shakes a black protein shaker hard three times, then pops the cap and drinks".

How should you describe the camera?

Give the camera one framing and one movement, in plain words: "medium close-up, slow push-in", "handheld, eye level", "locked-off wide". Too many camera instructions in one sentence make the shot harder to read: "slow push-in" is clear, "handheld crane orbit with a whip pan" is not.

As a starting rule, handheld at eye level suits creator-style UGC, a slow push-in makes one detail grow in importance, locked-off keeps the camera out of the way of a demo, and a slow orbit suits product hero shots. Every move, with phrasing for each, is in the FLUX 3 Max camera movement guide.

What goes in light and look?

Name the light the way a gaffer would: time of day, where the light comes from, how hard it is. "Warm morning light from a window on the left" gives the model a source and a direction. "Hard overhead studio light, black background" gives it a set. Then add one phrase on the grade: "soft natural grade", "high contrast, deep blacks", "muted film tones".

Cut "cinematic", "4k", "masterpiece" and "high quality". They take space and describe nothing the model can act on.

How do you prompt sound in FLUX 3 Max?

Audio is always on in FLUX 3 Max, so every clip has a soundtrack whether you direct it or not. Write sound in this order:

1. Spoken lines, in double quotes, with who says them: She says, "..." or A calm male voiceover says, "...".
2. Ambience and effects: the room, the street, the specific sounds of the action.
3. Music, or "no music".

Attribution matters more than it looks. Black Forest Labs' prompting documentation for FLUX 3 notes that a quoted line without a visible speaker or a voiceover cue may be treated as text that belongs in the frame, and that an off-screen voice needs a clear voiceover or narration cue. Never leave a quote floating on its own.

• Name the sound on the action. "The pump clicks as her thumb presses down" binds the effect to a moment the picture already contains.
• Write the ambience even when it feels obvious. "Quiet kitchen ambience, a fridge hum" makes the clip feel like a place. Name the sound source you want rather than asking for silence.
• Use placement cues when space matters: "traffic on the left, a kettle on the right".

Sound is not optional on placements that play it by default: Meta's Reels ads page notes that Reels play with sound on by default. If you plan to lay your own voiceover or music over the clip, still prompt the ambience, then replace the track with your own audio afterwards. The full playbook is in FLUX 3 Max sound and dialogue prompts.

How much dialogue fits in a FLUX 3 Max clip?

Budget roughly two to three spoken words per second of clip, then write fewer than you think:

Treat those as ceilings. The action needs time too: a sip, a pause, a product turning in the hand. If the last word gets cut off, shorten the line or lengthen the clip. For two speakers, name each one and keep the turns short.

Read every line out loud first, and watch the generated speech and lip sync on every clip before it goes live.

How does prompt expansion change what you write?

FLUX 3 Max reasons about your prompt and expands it before rendering, so a short prompt still produces a full shot. It is also why specific prompts still win: the expansion fills in whatever you leave open.

A runner ties her shoes on a park bench at dawn.

That prompt hands the model the camera, the light, the shoe colour, the sound and the music. If any of those matter to the ad, write them:

A woman in her twenties in a navy running jacket ties the laces of white
running shoes on a wooden park bench, then stands and bounces twice on her
toes. Low medium shot from the side, slow push-in. Cool blue dawn light with
a warm glow on the horizon, crisp natural grade. Birdsong, distant traffic on
the left, the scuff of shoes on gravel, no music.

If a detail is part of the brief (the product colour, the line, the framing), it belongs in the prompt.

How do you write FLUX 3 Max prompts for image to video?

Attach one reference image and FLUX 3 Max switches to image to video: your image becomes the first frame, the prompt drives the motion, and the audio is still generated. The image already carries the look, so write what moves and what is heard.

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.

Do not re-describe the picture. If the photo shows a matte black bottle on linen, write what happens next: "a hand enters from the right and lifts the bottle as the camera pushes in".

With the aspect ratio on auto, the clip follows your image, and a named ratio overrides that. FLUX 3 Max takes one image only, as the start frame; if a brief needs several reference images, Seedance 2.5 and MiniMax H3 Max take several. The walkthrough is in how to use FLUX 3 Max image to video.

What should you leave out of a FLUX 3 Max prompt?

Small on-screen text and exact logos inside generated frames are unreliable on every video model. Captions added in the editor, and text added after export, stay sharp, and changing them costs no regeneration.

How do you test FLUX 3 Max prompts without wasting credits?

Test at 720p, deliver at 1080p. FLUX 3 Max costs 17 credits per second at 720p and 29 credits per second at 1080p, the same for text to video and image to video, and 1 credit is 1 cent of generation cost.

Change one part of the prompt per draft: if the line was cut, trim the line and nothing else; if the framing is wrong, edit only the camera sentence. Change three things at once and you will not know which one fixed it. Once a draft holds, generate the delivery version at 1080p. That is a fresh generation, so the tighter the prompt, the closer it lands to the draft you approved.

The Starter plan's 1,400 monthly credits cover about eight 10 second clips at 720p, so prompt discipline is also budget discipline. Allowances for every plan are on the pricing page, and each clip length is costed in FLUX 3 Max pricing. To start from something other than a blank box, take a prompt from the FLUX 3 Max prompt examples and change one part at a time.

Frequently Asked Questions

How do you write a good prompt for FLUX 3 Max?

Write one paragraph per shot in four parts: subject and action, camera, light and look, then sound. Put spoken lines in double quotes with the speaker named, add the ambience and effects, and say whether you want music. Set duration, aspect ratio and resolution in the editor, not in the prompt.

How long should a FLUX 3 Max prompt be?

Long enough to settle every detail that matters to the brief, usually three to five sentences for one shot. The model expands short prompts, so a one-liner still renders, but anything you leave open is decided for you.

Can FLUX 3 Max generate dialogue?

Yes. Audio is always on and generated in the same pass as the picture, including spoken lines you write into the prompt. Put the words in double quotes, name who says them, and budget about two to three words per second of clip. Check the speech and lip sync before the clip goes live.

Should I put on-screen text in a FLUX 3 Max prompt?

No. Small text and exact logos inside generated frames are unreliable on every video model. Keep the prompt about picture and sound, then add captions in the editor and prices and logos after export, where they stay sharp and editable.

How is an image to video prompt different from a text to video prompt?

In image to video your image is the first frame and already carries the look, so the prompt describes motion and sound only. In text to video the prompt carries everything: the subject, the camera, the light and the sound.

Do short prompts work in FLUX 3 Max?

Yes. FLUX 3 Max expands your prompt before rendering, so a short prompt still produces a full shot. Use short prompts to explore, then write out the details that matter.