Watch once without pausing
Can a new viewer explain what happened? If not, fix the sequence before chasing surface detail.
Early-access model, live alternative workflow
FLUX 3 Video is still treated here as an early-access Black Forest Labs model, not as a selectable Voor endpoint. The generator above is labelled with the model it actually runs. Use this page to plan the announced longer, native-audio, multimodal workflow without pretending the requested model is already available.
Twenty seconds needs an ending.
A useful FLUX 3 Video brief is not a paragraph of cinematic style words. It is a sequence. The subject enters with known physical facts, something causes the motion, the camera witnesses the change, sound confirms distance and force, and the clip arrives at a frame an editor can actually use.
- Timeline
- Five readable phases across a clip up to 20 seconds.
- Sound
- Dialogue, effects, ambience, music, and deliberate silence.
- Continuity
- Identity, geometry, wardrobe, direction, and light.
A FLUX 3 Video prompt, staged in time
Keep the phase labels while drafting, then combine them into the prompt field above. The labels create an editing plan as well as a generation plan: each beat has a purpose, a review question, and a clean place to split the story if the model is asked to do too much.
- 1.00–3 s
Establish
Name what is already true.Open with the subject, environment, weather, materials, light direction, and camera position. The first frame should be understandable before motion begins. If a person, product, vehicle, or animal must stay recognizable, place those identity facts here rather than hiding them after style language.
- One main subject is readable at thumbnail size
- Travel direction and light direction are explicit
- Reference images have a named role
- 2.03–8 s
Initiate
Give motion a visible cause.Start one central action and let the environment respond. A foot compresses wet ground; a door changes the light; a hand rotates a product; wind pulls fabric. Physical cause and effect gives FLUX 3 Video or the selected live model a clearer temporal problem than a stack of mood adjectives.
- The subject action is concrete and singular
- Camera motion supports rather than competes
- The first sound belongs to an on-screen event
- 3.08–14 s
Develop
Change distance, not the whole world.Use the middle to reveal scale, texture, reaction, or a second consequence. Keep wardrobe, geometry, markings, labels, and spatial relationships stable. If the shot needs a new location, a second character, and a new camera grammar, it probably needs a second generation instead of one overloaded prompt.
- Identity remains stable through the fastest motion
- Dialogue, ambience, and effects occupy distinct roles
- No unrequested logo or readable text appears
- 4.014–18 s
Resolve
Remove energy with intention.Let the action complete. Slow the camera or hold it steady long enough to inspect the result. This is where a product faces the viewer, a character reaches the mark, dust settles, or the environment stops reacting. A resolved beat is easier to edit than a clip that is still accelerating at the cut.
- The main action has a visible completed state
- The soundtrack has room to settle
- Important objects remain inside the delivery crop
- 5.018–20 s
Hold
Design the final frame.Specify the exact last image and ask for a brief stable hold. The ending may need negative space for copy, a centered product, a profile pose, or a composition that loops into the opening. Review this frame at full size because temporal artifacts often collect at the end of a generated clip.
- The final pose is described, not left to chance
- The frame can cut, hold, or loop cleanly
- No continuity detail changes in the last second
Place sound on the same physical map
Native audio is most controllable when the prompt names source, distance, texture, and timing. “Cinematic audio” leaves every decision open. “Close ski-edge scrape, wind rising as the camera accelerates, one distant avalanche rumble, no music” gives the model events that can line up with the image.
Keep dialogue short enough for the visible performance. Name the speaker and quote the exact line. Separate ambience from effects: room tone or wind can continue, while a door latch or hoof strike belongs to a moment. Use silence deliberately before an impact or at the final hold instead of filling every second.
continuous ambiencecamera-relative distancevisible cause
Review four different versions of the same clip
A moving image can feel convincing while hiding broken frames, unsynchronized sound, or a story that only makes sense because you wrote the prompt. Separate the review passes so each problem gets a specific next action.
Watch once muted
Inspect identity, direction, hands, product geometry, edges, reflections, and unwanted camera drift.
Listen without looking
Check whether dialogue is intelligible and every impact, step, motor, breath, or ambience has a plausible place.
Pause at five frames
Review the opening, each major beat, and the final frame at delivery size. Motion can hide single-frame failures.
Production brief
A weathered grey horse crosses a windy highland from left to right. 0–3 seconds: hold a wide, low establishing frame; wet grass bends in gusts. 3–8 seconds: the horse begins to gallop and the camera tracks at shoulder height. 8–14 seconds: move closer without changing direction; hooves compress the ground, mane and coat respond naturally. 14–18 seconds: slow into a steady profile. 18–20 seconds: hold the final frame with open land ahead. Preserve facial structure, grey markings, scale, light direction, and travel direction. Use layered hoof impacts, breath, close wind, and distant thunder; no music and no text.