Prompt FLUX for image-to-video

Last updated: October 12, 2026

FLUX.3 animates an approved still into a short clip with sound generated in the same pass, and FLUX.3 Keyframes moves a shot through several stills in order. Pick them when the frame is signed off and the job is motion: a spell charge-up, an idle, a costume change between two approved looks.

Let the agent write it

You don't have to write any of this yourself. The Layer agent knows how to prompt each model it runs, from the same guidance this page is built on. Attach the still in chat, name the model and say what should happen, for example: Animate this with FLUX.3: a slow push-in as the mage charges a fireball on his staff, 5 seconds, with sound and no music. The agent writes the prompt, shows you its plan, and runs it when you approve. Use the rest of this guide when you want to write or adjust the prompt yourself, and see Write a brief for the agent for how to brief it.

When to use it

Use FLUX.3 when you have the frame and need it to move: ability casts, character idles, ambient motion for menus, and short trailer shots, with effects and ambience in the soundtrack. Use FLUX.3 Keyframes when you have two or more approved stills and want one continuous shot that passes through each: a transformation, a costume or faction swap, a storyboard beat by beat.

To make a clip from a prompt alone, see Create a video. For another model on the same job, see Prompt Seedance for image-to-video.

Versions on Layer

FLUX.3

The version this guide's main example was made with. Clips run 5 to 20 seconds at 720p or 1080p, from a First frame and an optional Last frame, with native audio and a Draft switch for a quick preview.

FLUX.3 Keyframes

Takes 1 to 10 images as keyframes and moves through them in order in one take, at the same lengths and resolutions as FLUX.3.

How to write the prompt

One camera move, then the action, then how it moves, then the style, then the sound:

Slow push-in on the mage. He raises the staff in both hands and the ember at its tip swells into a ball of fire, sparks spiralling upward, while his cloak lifts in a rising wind. The motion is slow and deliberate. Stylised painterly game art, colours unchanged. Audio: a low crackle of fire building to a soft roar, wind across the ledge, no music.

The still is the first frame, so it already fixes who is in the shot, where, and how it looks. Every word goes on how the motion unfolds from that frozen moment: what moves, in what order, how fast, where the camera goes, and what it sounds like. That prompt is the one in the clip above.

What works

One camera move per shot

Pick one: slow push-in, handheld follow, pan, orbit, or a static hold. We ran the prompt above against the same action with "the camera pushes in on the mage, then orbits around him, then tilts up to the sky and pulls back to a wide shot", on the same still and seed. The single push-in kept the mage on model and landed every beat. The four-move version jumped framing halfway and ended on a tiny figure under a sky-filling fireball.

Motion in plain, physical words

Say how it moves: slow, deliberate, weightless, abrupt. "Dynamic" and "energetic" leave the motion to chance.

Beats, not adjectives

Name a handful of things that happen in order rather than stacking descriptive words; past a few hundred words, later clauses fight earlier ones. For an action that must land on time, write a short timeline: two or three beats for a 5-second clip.

Draft first

Turn on Draft to find the motion. We ran the prompt above as a draft and as a full render on the same still and seed: the draft landed the same beats in the same order, with softer detail. Change the prompt on drafts, then render the keeper.

Keyframes for a change between two looks

On FLUX.3 Keyframes, the images are read in order as waypoints, and the prompt describes the change between them. We gave it the red-cloaked mage and the same still recoloured blue (made with FLUX.3 Image), with "a ripple of blue magic runs up from the hem of the mage's cloak, turning the red fabric deep blue with silver trim as it passes". The clip opened on the first still, a ring of blue light climbed the cloak, and it landed on the second, with the mage and the scene unchanged.

Sound in the prompt

With Generate audio on, name sound sources tied to what is on screen ("a low crackle of fire building to a soft roar, wind across the ledge") rather than moods. Put a spoken line in quotes with who says it and how ("the mage, low and steady: 'Burn.'"), and keep it well short of the clip's length; if the last word is cut off, shorten the line or lengthen the clip. Say "no music, no subtitles" when you want neither.

Name the style

End with the medium ("stylised painterly game art", "hyper-crisp vector motion design") and "colours unchanged", so the motion keeps the look of the still.

One change at a time

Between runs, change one layer (the camera, the action, the light or the sound). Rewriting the whole prompt shifts what already worked.

What to avoid

Several camera moves at once

Push in, orbit and pull back in one 5-second shot lost our character entirely. Cut a sequence of moves into separate clips.

Keyframes that are too far apart

Start and end frames that show different subjects, scenes or camera set-ups leave the model to invent the path between them. Keep keyframes related; the wider the gap, the less predictable the motion.

Redescribing the still

The first frame already shows the scene. Words spent repeating it are words not spent on the motion.

Settings that matter

First frame and Last frame pin where the clip starts and ends; on FLUX.3 Keyframes the stills go in Reference image, in order. Duration runs from 5 to 20 seconds, or Auto; find the motion at the shortest length, then render the length you need. Aspect ratio sets the shape and resolution; Auto follows the still. Generate audio adds sound in the same pass. Draft renders a quick, low-cost preview you can complete at full quality later. Above the prompt, Camera, Staging, Look and Performance take reference sets for each.

Prompts to start from

Spell charge-up

Slow push-in on the mage. He raises the staff in both hands and the ember at its tip swells into a ball of fire, sparks spiralling upward, while his cloak lifts in a rising wind. The motion is slow and deliberate. Stylised painterly game art, colours unchanged. Audio: a low crackle of fire building to a soft roar, wind across the ledge, no music.

Costume change between two keyframes

Fixed camera. A ripple of blue magic runs up from the hem of the mage's cloak, turning the red fabric deep blue with silver trim as it passes, while the cloak stirs in the wind and the ember on his staff flickers. Stylised painterly game art. Audio: a soft shimmering chime and wind, no music.

Use it in Layer

Add the approved still to a session's board.
Click View on the image, then Turn into video. Create video opens in the side panel with the image as its First frame.
Click the model name at the top of the panel and choose FLUX.3. The first frame stays in place.
In the prompt box, write one camera move, the action, how it moves, the style and the sound.
Turn on Generate audio if the clip needs sound, and leave Duration short while you find the motion.
Click Create video. The cost shows on the button before you click.
When the clip lands on the board, click View to open it in the viewer and play it. Then change one thing in the prompt at a time and run it again.

Learn more

Image-to-video (Black Forest Labs): the three ways to animate stills (a start frame, a start and an end frame, several keyframes) and how to choose between them.
Video generation with FLUX (Black Forest Labs): what a FLUX.3 video prompt needs, why a still's prompt is about motion, and one camera move per shot.
Audio and speech (Black Forest Labs): writing dialogue, ambience, effects and music into the prompt, and keeping spoken lines short enough for the clip.
FLUX 3 video (Black Forest Labs): the modes, durations, audio and draft previews FLUX.3 offers.
Prompting FLUX 3 (Runware): camera moves written the way the model follows them, beats over adjectives, and iterating one layer at a time.