FLUX 3 Video Prompting Guide

FLUX 3 Video is announced with text, image, and video inputs, clips up to 20 seconds long, native audio, video continuation, keyframe transitions, multilingual dialogue lip sync, typography, and multi-clip support.

Public access is still limited to initial API partners, so this is a planning framework rather than a claim about a tested FLUX 3 prompt syntax. Use it to write a clear creative brief now, then adapt the fields when the public API becomes available.

Start with one shot

Write the prompt as a shot that can be seen and heard, not as a theme.

Weak:

A video about a new coffee brand.

Stronger:

A close product shot of a ceramic coffee cup on a dark counter. The camera moves slowly from left to right while steam crosses the morning window light.

The stronger version identifies the subject, framing, environment, action, and camera movement. It gives each part of the shot a job.

Use a fixed prompt order

A repeatable order makes a prompt easier to revise:

1. Subject — who or what is on screen.
2. Action — the visible change during the shot.
3. Setting — location, time, weather, and relevant surfaces.
4. Framing — close-up, medium shot, wide shot, overhead view.
5. Camera movement — static, pan, track, push in, pull back.
6. Lighting and visual treatment — direction, contrast, color, and texture.
7. Audio — ambience, sound events, dialogue, and silence.
8. Constraints — details that must remain stable.

This order is not an announced FLUX 3 API schema. It is an editorial checklist for making the intent unambiguous.

Match the input to the task

FLUX 3 Video is designed to accept three input types.

Text input

Use text when the shot can begin without a supplied visual reference. Name concrete, visible details and a single primary motion.

Image input

Use an image when the opening composition, subject appearance, product design, or typography already exists. In the text, focus on what changes after the supplied frame: motion, camera, lighting, and sound.

Video input

Use video when the goal is continuation. Describe the next action and what must remain continuous. Avoid rewriting the entire existing clip as though the model had not received it.

Plan duration around visible beats

FLUX 3 Video is announced to generate up to 20 seconds. Treat that as a ceiling, not a requirement.

Divide the requested time into visible beats before writing:

• Opening state.
• Primary action.
• Ending state.

For example:

0–4 seconds: the sealed box sits centered on the table.
4–12 seconds: the lid opens as the camera moves closer.
12–18 seconds: the product rises into the light and holds.

This is duration planning, not evidence that FLUX 3 follows timestamp notation. If the eventual interface does not accept timestamps, the same beat structure can still guide the prose.

Write audio as part of the scene

Because native audio is an announced capability, do not leave sound as an afterthought. Separate four layers:

• Ambience: room tone, traffic, wind, rain, crowd.
• Action sounds: footsteps, a latch, fabric, glass, machinery.
• Dialogue: exact words, speaker, language, and delivery.
• Music: presence or absence, instrumentation, and intensity.

Example:

Audio: a quiet workshop ambience, one metal latch click as the case opens, no music. A calm German voice says, “Bereit für den ersten Test.”

FLUX 3’s announced multilingual lip-sync support makes the language and spoken line relevant prompt inputs. Accuracy still needs testing after public access opens.

Use keyframes to define a transition

For a keyframe transition, write down what must change between the start and end frames:

• Subject position.
• Camera position.
• Lighting.
• Environment.
• Elements that must not change.

Example:

Transition from the supplied daylight storefront to the supplied night storefront. Keep the building geometry and camera position fixed. Pedestrian traffic thins, interior lights turn on, and the sky darkens gradually.

The start and end images carry the visual anchors; the prompt describes the motion between them.

Prompt continuation around continuity

For video continuation, inventory the details that should survive the cut:

• Subject identity and wardrobe.
• Direction and speed of motion.
• Camera path.
• Light direction.
• Audio bed.

Then describe only the next beat. This prevents the continuation brief from becoming a second, conflicting scene description.

Treat typography as a verification task

FLUX 3 lists typography among its announced capabilities. Put exact on-screen copy in quotation marks, state where it appears, and keep it short enough to inspect.

After generation, check every frame where the words are visible. Do not treat one correct frame as proof that the text remains correct throughout the clip.

Build longer ideas from clips

Multi-clip support is part of the announced capability set. Plan a longer idea as a shot list:

1. Establishing shot.
2. Action shot.
3. Detail shot.
4. Closing shot.

Give each clip one visual purpose and carry forward only the continuity details it needs. This keeps revisions local: if the detail shot fails, the whole sequence does not need to be regenerated.

A reusable prompt worksheet

Copy this structure into your brief:

Input: text / image / video
Duration plan: opening beat / action beat / ending beat
Subject:
Action:
Setting:
Framing and camera:
Lighting and treatment:
Audio:
Dialogue and language:
Start/end keyframes, if any:
Continuity constraints:
Exact typography, if any:

The best first prompt is not the longest one. It is the one whose choices can be checked in the output. When FLUX 3 Video becomes publicly accessible, that checkable structure will make real iteration possible.

FLUX is a trademark of Black Forest Labs Inc. AI Flux3 is an independent project and is not affiliated with Black Forest Labs.