11Labs AI

Flux 3 video: turn a scene idea into a testable shot

A video brief needs to explain change, not just appearance. Start with one visible event and decide what would make the sequence believable. This 11Labs AI guide separates the FLUX 3 video family from image generation and walks through a practical prompt review.

Identify the workflow

A video model is a different task from an image model

Black Forest Labs describes FLUX 3 as a video model. Start your own shot plan by deciding what happens, what the viewer sees first and how the event ends.

The FLUX.3 Video workspace uses a written prompt and an optional photo reference. Enter your scene description, add a photo when it helps establish the composition and check the selected model before creating. Projects involving an existing video or a separate soundtrack need a workflow with those input controls.

FLUX.2 image workflows address a still image. Do not use an image-model result as proof of how a FLUX 3 clip will move, and do not assume that an image prompt alone specifies the event you want.

Original brief: a lantern settles after a gust

A single paper lantern hangs above an empty courtyard at dusk. A light gust pushes it to the right, then it swings back and gradually settles. The camera stays at eye level without moving. Warm light from the lantern falls onto the nearby wall, while the courtyard remains visible in the background.

The scene has one cause and one response. The gust explains the movement; the settling gives the clip an end state. This is easier to inspect than a prompt that simultaneously asks for a crowd, a camera orbit and several lighting changes.

If you start from a photo, choose a composition that leaves room on either side of the lantern. Describe the motion after the starting frame. The reference can help establish appearance, but it cannot show the whole timing of a swing.

Check motion and continuity before style

  1. Read the event

    Watch once without stopping. Can you identify the gust, the return and the settling? If the clip never reaches a readable end state, simplify the instruction.

  2. Track the attachment

    Check where the lantern hangs. A convincing surface texture cannot hide a changing cord or a lantern that detaches from its support.

  3. Inspect the surroundings

    Look for a stable wall and consistent light. Note whether a change seems motivated by the action or appears without a cause.

  4. Revise one issue

    Reduce the swing or clarify the fixed camera before changing the entire visual style. Keep the prompt and selected model name with the result.

Use a reference to define contact and light

The hand-to-lid contact, hive edges and low sunlight offer specific visual relationships to inspect. Turn them into review questions: where does the hand meet the lid, which edges stay readable and what explains the light?

A beekeeper in protective clothing holds a hive lid beside wooden beehives in a sunlit meadow near sunset
Reference from the Black Forest Labs FLUX 3 Video page. It is not an output generated by 11Labs AI.

Flux 3 workflow questions

When is a photo reference useful?

Use one when the starting composition or appearance matters. Leave space for the intended action and describe what should happen after that first moment.

How should I check the final clip for an edit?

Review its actual duration, dimensions and sound, then place it in the intended sequence. Check the opening and ending as carefully as the strongest frame.

How do I compare another video model?

Keep the same starting brief and record the selected model name. Compare a specific event, such as the lantern settling, before changing the setting or style.