FLUX 3 Max Text to Video

FLUX 3 Max Text to Video

FLUX 3 Max Text to Video is a frontier text-to-video model, a post-trained version of Black Forest Labs' FLUX 3 multimodal foundation model. Describe a scene and it returns a finished clip with synchronized native audio — no starting image required.

What is FLUX 3 Max Text to Video?

FLUX 3 Max turns a written description of subject, action, camera movement and sound into a video of up to 20 seconds at 720p or 1080p. It is the post-trained, latency-optimized version of FLUX 3 Video: the base model was trained further for motion coherence, prompt adherence and visual quality, and its inference path was rebuilt so a 10-second clip comes back in about the time it takes to read this sentence. On Artificial Analysis's preliminary image-to-video arena it is tied for first in blind human-preference voting.

Key Features

• •Native synchronized audio in the same pass, so an impact lands with its sound on the right frame.
• •Motion that stays true to life — fast-moving subjects keep their shape from first frame to last.
• •Strong prompt adherence across busy, multi-subject and even anachronistic scenes.
• •Durations from 5 to 20 seconds, 720p or 1080p, aspect ratios from 21:9 to 9:16.
• •Reasoning-level control and a seed for reproducible generations.

Best Use Cases

Because a draft and a final come from the same fast model, FLUX 3 Max fits rapid previsualization, storyboards, social and ad clips, and concept exploration where you iterate many takes of one shot. Its motion fidelity suits fast action — sport, performance, machinery, crowds — and native audio suits scenes built around a sound event. In testing, a prompt for three taiko drummers on a night festival stage returned clean rigid-body motion with blurred sticks holding their shape and drum hits locked to the picture.

Prompt Tips and Output Quality

Name concrete subjects, one clear action, a camera move motivated by that action, and one line describing the sound. Keep reasoning level high for complex, multi-element prompts and drop to low for simple scenes. Leave audio on for anything with a sound event.

How to Use FLUX 3 Max Text to Video

FLUX 3 Max Text to Video generates a complete clip with synchronized native audio from a single prompt. The quality of the output tracks the quality of the description, so write the prompt like a shot list.

Writing the Prompt

Lead with the concrete subject, then one clear action, then a camera move that the action motivates, and finish with one line about the sound. Specific nouns and a single, legible motion read far better than stacked adjectives. For multi-subject or unusual scenes, spelling out who does what and in what order keeps the model on track. Because audio is generated in the same pass, naming the sound — a drum hit, a line of dialogue, an engine — ties it to the on-screen event.

Choosing Parameters

• •duration: 5 to 20 seconds. Start at 5 while you iterate on a shot, then extend once the motion and framing are right.
• •resolution: 720p for quick drafts; 1080p for final delivery. Both are billed per second of output.
• •aspect_ratio: auto maps to 16:9. Use 9:16 for social, 21:9 or 2:1 for a wide cinematic frame, 1:1 for feeds.
• •generate_audio: on by default. Keep it on for any scene with a sound event; switch off for a silent clip.
• •reasoning_level: high plans more about a complex prompt before rendering and follows it more closely; low is faster for simple scenes.
• •seed: -1 gives a fresh result each run; set a fixed number to reproduce an exact clip.

Recommended Settings

The default example was generated at duration 5, resolution 720p, aspect ratio 16:9, generate_audio on, reasoning_level high, seed 94355960. That combination returned a clean, reproducible result: three taiko drummers in motion with rigid-body fidelity on the sticks and drum hits locked to the picture. For fast iteration, keep 720p and reasoning high, vary the prompt and seed, then switch to 1080p and lengthen the duration for the final render. Reach for reasoning high whenever a prompt asks for several subjects, a specific camera move, or an unusual concept.

Examples to try

Each example ran end to end at the settings shown — its motion, framing and audio came from FLUX 3 Max in one pass.

• •Performance with synchronized audio (default). Prompt: "Three taiko drummers in indigo happi coats drive their wooden bachi into a huge nagado-daiko and two shime-daiko in unison, lunging low on every downbeat, sweat flicking off their arms; a hanging red paper lantern sways behind them on a wooden matsuri stage. The camera dollies laterally past them at the height of the drum skins. Audio: thunderous synchronized drum booms locked to each strike, sharp rim clacks, a shouted 'Sore!', a crowd murmuring in the dark." Three drummers strike in unison and every boom lands on the frame it is struck. Tip: name each sound in an "Audio:" line with generate_audio on.
• •Fast motion that keeps its shape. Prompt: "Four thoroughbred racehorses thunder around the final dirt bend of a racetrack, jockeys crouched low in bright silks, clods of earth flung up behind the hooves; the camera tracks alongside at rail height, level with the leader's shoulder. Audio: pounding hooves in a quickening rhythm, a jockey's sharp shout, a rising crowd roar." Legs, dust and silks stay clean at full gallop with no smearing. Tip: keep reasoning_level high for busy multi-subject motion and track the camera with the action.
• •Faces and materials in close-up. Prompt: "Extreme close-up of an elderly watchmaker's face and hands, a brass jeweler's loupe screwed into one eye, as she lowers hair-thin tweezers to seat a tiny gold gear into an open pocket-watch movement; daylight from a workshop window rakes across her lined skin and the polished brass. Audio: a faint mechanical ticking, the tiny click of the tweezers, a quiet room tone." Lined skin and the jeweled gold movement hold up at close range while features stay stable. Tip: give a close-up one quiet action, then use 1080p for finals.

Also built for: "a baker sliding a scored loaf into a stone oven, the peel scraping," or "a 9:16 busker fingerpicking a guitar as a train rumbles past."

Does FLUX 3 Max generate audio?

Yes — native synchronized audio is on by default, rendered with the video.

How long can a clip be?

Up to 20 seconds in a single generation.

What resolutions are available?

720p for fast drafts and 1080p for final delivery.

Do I need an input image?

No — the text-to-video endpoint builds the whole scene from your prompt.

How is it different from FLUX 3?

It is a post-trained, latency-optimized variant of FLUX 3, scoring higher on the preliminary image-to-video arena than base FLUX 3.