Prompting guide

01

The one rule that matters

Direct a scene. Don't list objects.

Say what moves, what the camera does, and what the place feels like — instead of listing what is in the frame. A prompt with clear motion and camera language beats a detailed still description every time.

Three rules specific to this model

The one exception is audio and on-screen text. BFL's own docs recommend these and they do work: no on-screen text or subtitles · no sales voice · no announcer delivery.

02

Pick the right mode first

The three buttons on the left of the composer. Get this wrong and no amount of prompt-writing saves the clip.

03

Four ways to write a prompt

Pick by how much control you want. Longer is not automatically better.

1 · One sentence (the best default)

Shape: [camera] shot of [subject] [doing what] in [where], then the look and the motion.

A low tracking shot of a fox sprinting through wet pine undergrowth at dawn. Mist drifts between the trees as the camera keeps pace beside it. Cool blue morning light, fast but controlled motion, cinematic naturalism.

Good for quick ideas, B-roll, and a single clear subject. The risk: anything you leave out, the model decides for you.

2 · Very short

A red fox leaping through fresh snow, telephoto.

The model fills in the rest, sometimes with something you'd never have written. Start short to explore, then grow the prompt once you know what needs pinning down.

3 · Labelled fields (change one thing at a time)

Camera shot: wide shot, low angle
Subject + action: a lone rider crosses a shallow desert river
Depth of field: shallow (sharp on subject, blurred background)
Lighting + palette: warm backlight with soft rim — amber, cream, walnut
Motion: water splashes around the horse's legs, orange dust hangs in the light
Style: epic western realism

Use this when you want to fix one thing without disturbing the rest.

4 · Timeline (put the action on beats)

0.0–1.5s — locked wide of a still harbor at dawn, boats motionless on glassy water
1.5–3.0s — a slow push-in begins as gulls lift off the water
3.0–5.0s — the sun breaks the horizon, warm light spreads and the camera settles

And the full schema, when several shots must match

Use these six fields when look and story must carry across shots: Core summary (one line for the whole piece) · Scene (place, light, depth of field per shot) · Subject description · Dynamic narrative (what the camera and the character do, with timing) · Audio (per shot) · Style and color.

04

From-image mode has three shapes

Most people only know the first one. All three use the same field — they differ only in how many images you attach and whether you fill in the seconds.

05

Five things keyframes can do

Every one of these uses the same From-image button and the same seconds boxes. Only the number of pictures and the numbers you type change.

The catalogue of shapes below comes from Runware's keyframes guide. The instructions are rewritten for the buttons on this site — Runware pins by frame number, we pin by seconds, which is the same thing said differently.

1 · One picture as the opening frame

Attach one image and leave the seconds empty. Best for a still product shot that should become a clip. Saying “use this image as the first frame” outright anchors it harder.

Use this image as the first frame. An 8-second clip: the camera holds locked on the knife for a beat, then a chef's hand enters from the top of the frame, picks up the knife by its walnut handle, and makes a single clean slice through a green lime.

2 · One picture as the closing frame

Attach one image and type the clip's full length into its seconds box. Use it when the final frame is the actual deliverable — a logo, a packshot, a UI screen that has to land exactly.

An 8-second unboxing shot on a warm walnut wooden counter under soft morning light. The hands set the bag down centred in the frame, fold the tissue neatly to one side, and withdraw. Land on the pinned packshot as the final frame: the bag centred alone on the counter, no hands in view.

3 · Two pictures — before and after

Attach two images and leave both seconds empty. The model invents the motion between the two states. This one demands that the two frames share a composition — same camera position, same light, same objects, only their state changes.

An 8-second locked-off overhead-ish morph between the two frame images: the cluttered desk at the start, the tidy desk at the end. Interpolate a plausible tidying motion between them — an unseen pair of hands stacking the papers, closing the notebook and squaring it, closing the laptop and centring it, lifting the mug out of frame. Camera stays locked in the same framing throughout. Warm morning light unchanged.

4 · Three pictures as a storyboard

Attach three images and fill in every seconds box (remember this site rejects a partial fill). The model runs through the pins in order as one continuous shot.

A 10-second overhead cooking time-lapse on a warm walnut table. Ingredients compose onto the plate in fast-motion between the pinned states: herb oil pooling into the centre by the midpoint, then asparagus and pecorino settling on top by the end.

5 · Pin a picture to a beat

Attach one image and type a fractional second — 3.5 works. Use it when a pose has to land on an exact beat: the peak of a jump, a hit in the music.

A 7-second locked-off tightly-framed wide shot of the athlete on the same gym floor. She runs into frame from the left over the first two seconds, plants her feet, and jumps straight up. At the peak of the jump she holds the exact pose from the pinned frame image, then falls into a clean landing over the next two seconds.

06

Camera vocabulary

Drop these straight into the prompt — they work as-is.

Shot size

extreme close-up · close-up · medium shot · cowboy shot · full shot · wide shot · establishing shot · two shot · macro

Angle

eye level · low angle · high angle · bird's eye · worm's eye · ground level · POV · over-the-shoulder · profile shot · dutch angle · fourth wall

Composition

center framing · rule of thirds · symmetry · negative space · leading lines · frame within frame · foreground occlusion · silhouette · reflection framing

Camera movement

pan · tilt · dolly in · tracking shot · orbit · crane · handheld · steadicam follow · whip pan · push through · dolly zoom · camera roll · locked-on

Focus and lens

shallow depth of field · deep focus · rack focus · wide angle (24mm) · telephoto compression · fisheye · macro lens · anamorphic flares · halation · vignette

Time and speed

slow motion · speed ramp · freeze frame · bullet time · timelapse · long exposure look · fast motion · boomerang · cinemagraph

Light

golden hour · rim light · chiaroscuro · volumetric light · haze · hard light · neon practicals · spotlight · underwater light

Transitions

match cut · whip transition · foreground wipe · jump cut · quick cuts · pass-through

07

Audio and dialogue

Audio is generated alongside the picture and lip-synced for you. Scenes whose sound you can guess from the image come out better than abstract ones.

The shape the docs recommend. Use only the lines your scene needs.

A weathered fisherman leans on the rail of a small boat at dusk, talking to someone off-camera.
Dialogue: the fisherman says, "We turn back before the light goes."
Ambience: open water, a low steady wind, hull creaking.
Effects: rope tapping against metal, one gull passing overhead.
Music: none.
No on-screen text or subtitles. No announcer delivery.

Direct the voice with concrete words

Give an age range, an accent, a register (low · mid · bright · soft · rough), a recording (close and dry · across a room · phone) and a delivery (lightly amused · hesitant · practical) instead of vague adjectives like “nice voice”.

British man in his thirties with a warm low-mid voice, recorded close and dry.
Conversational and lightly amused, one relaxed breath, imperfect human timing.
No sales voice, no over-enunciation.

Thai and other languages work. Tag each line with its language, and if one person switches languages, say it is the same voice. Native script or romanised both work — but listen back and check.

08

Continue mode

The source clip sets the motion, the cast and the look. The prompt only says what happens next.

A herd of African elephants walks steadily toward the camera across a dry savanna beneath a huge hazy orange sunset, the animals growing larger in frame as the sun sinks lower behind them.

09

Why a prompt sometimes gets blocked

This section comes from our own logs, not from anyone's documentation.

There are two checks, not one: one reads the prompt (fails at about 30 seconds) and one watches the finished video (about 130 seconds). That is the answer to “why did the same prompt work yesterday” — each generation is different, so the second check can rule differently.

• Copyrighted character or brand names — by far the most common, and no tolerance setting helps
• Children plus horror — each is fine alone, together it is a red flag
• Bedroom or waking-up context together with a female character

The fix is to describe instead of name: swap the cartoon's name for a blue round-headed robotic cat with a red collar and bell.

A test that tells you which half is the problem

Keep the same image, but use the shortest prompt you can write. If it passes, the prompt was the problem. If it still fails, the image is — and no wording will save it.

Three friends stand together and talk. The camera stays still.

10

The three quality levels

The cheapest way to work: explore on Standard because it is fast and cheap, then re-run the prompt you like on HD.

11

Before you press generate

Did you write all five?

• Subject and action — who moves, and how
• Camera — still, pushing in, handheld follow, top-down, or a whip
• Place and mood — where, what light, what time, what feeling
• Quality of motion — slow, jerky, floating, chaotic, precise
• What must not change — especially with an image or a source clip

Trap check

• No “no …” instructions about the picture (audio and on-screen text are the exception)
• No character, brand or real-person names
• With an image attached, you are not re-describing what it already shows
• If you typed seconds, you filled every box, in ascending order
• If you're adding your own music, audio is off
• For Continue, the source clip is 15 seconds or shorter
• Could it be longer? Longer means steadier

Then iterate, one change at a time

The first prompt is almost never the last. BFL's docs walk it up like this — note that each step adds one thing rather than rewriting everything, otherwise you never learn what helped.

a video of a eagle

a closeup video of a eagle, the eagle sits on a tree in a forrest

a cinematic closeup of an eagle perched on a pine branch in a misty forest, feathers ruffling in the wind, slow push-in, golden-hour light