Describe the scene, motion, mood, and camera direction, then generate video with a model and parameter set that fits the job.
Describe subject, motion, mood, and camera, and generate cinematic clips with native audio. Powered by Seedance 2.0 — the model currently winning blind preference votes — alongside Veo 3.1 Fast.
A storm at sea, a chase through neon streets, a quiet diner at night — write it like you would tell a friend, and the model handles cinematography. Veo output is repeatedly described as "directed, not generated": intentional light, believable physics, real camera grammar.
Slow push-in, timelapse bloom, handheld energy — motion direction in the prompt is followed, and duration is yours to set (4–15 seconds depending on model). The clip matches the brief instead of approximating it.
Rotating hero shots, pour shots, unboxings — commercial-style footage that used to need a studio session generates from a paragraph. Per-second pricing with the exact cost shown upfront keeps campaigns budgetable.
Native dialogue, ambience, and effects arrive with the visuals; 9:16 vertical and 16:9 widescreen are both first-class. The output is a postable clip, not a silent draft.
Each of these started as written text and nothing else. The prompt that produced it is printed underneath.
A ceramicist in a sunlit workshop looks up from the wheel and speaks warmly to camera: "This one finally came out right." Clay-dusted hands, natural skin texture, shallow depth of field, soft window light, ambient studio room tone.
A golden retriever puppy sprinting across a sunlit meadow toward the camera, ears flapping, slow-motion feel, backlit grass glowing, joyful energy.
A parkour athlete vaulting across rooftop gaps at golden hour, dynamic tracking camera, accurate body mechanics and momentum, dust kicked up on landings, cinematic sports energy.
Handheld vertical walk through a neon night market, steam rising from food stalls, lanterns swaying, people passing in soft motion blur, lively street ambience.
Close-up of an elderly fisherman laughing warmly at dusk on a boat deck, wrinkles and windblown hair rendered naturally, genuine expressive performance, handheld documentary feel.
Two tiny paper boats racing down a rain-gutter stream after a storm, water rushing over pebbles, playful close follow shot.
Vertical fashion film: a model in a flowing emerald dress walking toward camera down a rain-slick runway, fabric rippling naturally with each step, confident stride, editorial lighting.
Close-up of a chef\'s hands rapidly julienning a carrot on a wooden board, knife tapping fast, a handwritten chalkboard menu reading "SOUPE DU JOUR" visible behind, busy kitchen sounds.
One written scene, several engines — pick the one whose motion you like.
| Input | Text prompt only — no image required. |
|---|---|
| Prompt languages | All 27 interface languages; non-English prompts are enriched before generation. |
| Duration | 4 to 30 seconds depending on the model. Longer pieces are built by extending a clip you like. |
| Resolution | 480p and 720p across the range; 1080p and 4K on Seedance 2.0 standard. |
| Audio | Some engines generate natively synced audio in the same pass; others return silent clips. Marked in the panel. |
| Cost | Per second of output, from 15 credits/second at 480p, shown in full before you generate. |
| Failed generations | Not charged — credits stay in your balance. |
Write the scene, choose the engine, generate.
Describe subject, action, setting, and camera movement. Text to video models respond to shot language — "slow dolly in", "handheld", "wide" — far more than image models do.
Each text to video engine has its own strengths in motion and audio. Cost is per second and shown before you generate.
Download the clip, extend it, or take it into further editing. Prompt and settings stay attached in history.
Text to video models read shot descriptions, not scene summaries. Four components, in this order.
Who or what is in frame, with one distinguishing detail.
a chef in a stained white jacket
One action, described as it unfolds.
plates a dish, tweezers placing the last herb
Position and movement. State it even when the answer is “no movement”.
locked-off overhead shot
Light, colour, and finish.
hard kitchen light, high contrast, documentary
a woman walking in the rain
a woman in a long coat walks toward camera through evening rain, umbrella tilted back, shop lights blurred behind her, slow dolly back at eye level, shallow focus, cool blue with warm window light
The short version specifies a subject and a weather condition and nothing else — camera, direction, light and lens are all left to the model. The long version fixes each one, which is why it produces the same shot twice.
product video for a skincare bottle
a frosted-glass serum bottle rotates slowly on a wet stone surface, camera locked off at product height, single soft key light from the right, water droplets catching highlights, macro, clean commercial look
Product video fails on framing and light far more often than on the object. Fixing the camera, naming one light, and choosing macro gives the model the three decisions it would otherwise guess.
Prompt-driven video is one route. Reference images, avatars, and editing are the others.
How the current text to video engines differ on motion, audio, and cost.
The real price of a usable clip, engine by engine.
Parameters and per-second pricing for the workhorse text to video engine.
A fast engine and the prompt style it rewards.
When the Pro engine still wins on a text-driven shot.
Turning a plain sentence into a shot description a model can follow.
Text to video is billed per second of output. Free accounts get 80 credits a day; paid plans open the premium engines and longer durations.
Perfect for getting started
Billed annually · $240/year
Billed annually · $300/year
You describe the scene — subject, motion, mood, camera — and an AI video model generates a clip from scratch, including native audio. No source footage is needed.
Seedance 2.0 Fast for affordable iteration, Seedance 2.0 for multi-shot stories, Veo 3.1 Fast for single-shot cinematic polish, Seedance 1.5 Pro when you need exact 1080p/duration control.
Roughly 4 to 15 seconds per generation depending on the model, with scene extension available on Seedance 2.0 to continue sequences.
Yes — Seedance and Veo generate dialogue, ambience, and effects natively. Audio can be disabled on Seedance for a lower per-second rate.
Per second, varying with model, resolution, and audio — always displayed before you generate, with automatic refunds on failures.
Name the subject, the action, the setting, the mood, and the camera move. The built-in prompt enhancer expands short ideas into well-structured video briefs automatically.
Yes — choose 9:16 per generation for vertical platforms, or 16:9 for YouTube and web.
Separate the subject, the action, and the camera. "A chef plating a dish" is a subject; "slow push-in on a chef plating a dish, shallow depth of field, warm kitchen light" is a shot. Text to video models follow the second far more reliably.
On the engines that support native audio, yes — including synced speech. Others produce silent clips. Which is which is marked in the model panel before you spend credits.
Longer durations drift more, because the model has more frames to fill and less anchoring. Shorter clips hold the prompt more tightly; if you need length, generate a short take you like and extend it rather than asking for 30 seconds up front.
One-click tools for everything around generation — polish, restyle, and share.
