Realistic video out of Vidu — the multimodal generator from Shengshu Technology — comes from restraint, not from stuffing the prompt with adjectives. Give one clear subject, one continuous action, one named camera move, and real-world lighting, then let the Vidu Q2 engine (Q2 and Q2 Pro for the highest detail, Q2 Turbo for fast drafts, all 1080p at 24fps with synced 48kHz audio) do the physics. This guide covers the formula first, then the four modes, then 10 full prompts you can paste and adapt.
Every example ends with a Settings line, because resolution, aspect ratio, clip length and movement amplitude are app controls in Vidu, not inline parameters — you set them in the panel, never with --flags. Once the formula clicks, browse the full best Vidu prompts roundup for 38 more, and keep the Vidu prompt cheat sheet open for the camera-move and amplitude tables.
The Vidu prompt formula
Write one flowing sentence in this order: subject + appearance → action/motion → camera move → setting → lighting & mood → settings/amplitude. Each part does one job, and skipping any of them is where realism breaks down.
1. Subject + appearance. Name exactly one main subject and describe it concretely — "a barista in a linen apron," not "a person." A vague subject forces Vidu to guess, which is where warping and identity drift start.
2. Action / motion. Give the subject a single continuous action. One thing happening — pouring, walking, turning — reads as real; two actions or a cut inside one shot confuses the model and produces smearing.
3. Camera move. State one explicit move Vidu knows: slow push-in, pull-back, dolly, tracking, orbit, aerial/drone, tilt, pan or static/locked-off. Vidu follows camera language closely, so this is what gives the motion direction instead of drift.
4. Setting. Place the subject in a clear environment with a couple of anchoring details. This grounds the physics — reflections, shadows and scale all key off the scene you describe.
5. Lighting & mood. Describe the light like a cinematographer would — "soft window light with a warm bounce," "low golden backlight." Realistic lighting is the single biggest lever on whether a clip looks shot or synthetic.
6. Settings / amplitude. These live in the app: pick resolution, aspect ratio, length, and movement amplitude (auto / small / medium / large). Keep the whole prompt to roughly 25–60 words and one action per clip.
Pick the right mode
Vidu has four generation modes, and choosing the right one matters more than any wording. Match the mode to what you already have locked in hand.
Text-to-video (T2V) invents the whole scene from your description — use it when nothing is fixed yet and you want Vidu to build the shot. Follow the six-part formula in full.
Image-to-video (I2V) animates a still you upload, and it is the most reliable route to a realistic result because the look is already locked. Describe only the motion to add, then say "keep the face and background identical." Do not re-describe the whole scene.
Reference-to-video is Vidu's signature feature: upload up to seven reference images — characters, props, scenes — and Vidu fuses them into one consistent clip. Name which image each element comes from and add "keep each character consistent with its reference image." Save recurring subjects in the My References library to reuse them across shots.
Start-and-end-frame takes a first and last frame and fills the transition. Keep the two frames' aspect ratios within 0.8–1.25 of each other, set amplitude to small for a smooth morph, and use only a brief prompt to steer mood, motion and camera — the frames carry identity. For the reference workflow in depth, see the Vidu reference-to-video prompts.
10 example prompts
These ten cover the main use-cases and follow the formula exactly. Copy one, swap the bracketed parts, and set the app controls from the Settings line under each.
1. Realistic b-roll — kitchen morning
Steam rising from a fresh mug of black coffee on a worn oak kitchen table, tiny wisps curling and bending in the still air, slow push-in toward the rim of the mug, soft morning window light with a warm bounce off pale walls, quiet homely mood, shallow focus, gentle ambient kitchen sound.Settings: Vidu Q2 · 16:9 · 1080p · 6s · amplitude: small. Why it works: one tiny motion (rising steam) plus a slow push-in gives Vidu a single direction, so the clip reads as real b-roll instead of a moving photo.
2. A person — woodworker at the bench
A bearded woodworker in a canvas apron running a hand plane along a length of oak, thin curled shavings peeling off and drifting to the floor, slow tracking shot moving with the plane, warm workshop light from a high side window, sawdust in the air, focused craft mood, soft shaving sound.Settings: Vidu Q2 · 16:9 · 1080p · 6s · amplitude: small. Why it works: a single contained hand action plus a track that follows the tool keeps the person stable and the motion legible.
3. A product — headphones on a stand
A pair of matte-black over-ear headphones resting on a brushed-metal stand on a dark reflective surface, a soft light sweep travelling across the ear cups, slow 360-degree rotation of the stand, clean studio lighting with a crisp rim highlight on the edges, premium minimalist commercial look, deep shadows.Settings: Vidu Q2 Pro · 1:1 · 1080p · 5s · amplitude: small. Why it works: a slow single-axis rotation and small amplitude keep hard product edges from warping, which is what sells a hero shot.
4. Anime — commuter on a rainy platform
An anime schoolgirl with a navy blazer waiting under an umbrella on a rain-soaked train platform at dusk, rain streaking past and reflections shimmering on the wet ground, she glances up as a train light approaches, slow push-in on her face, moody teal-and-amber palette, crisp Makoto Shinkai-inspired line art, soft rain sound.Settings: Vidu Q2 · 16:9 · 1080p · 6s · amplitude: auto. Why it works: naming the exact style plus one small head motion lets Vidu lean on its strongest look — clean, stable anime — without overreaching.
5. Reference-to-video — character in a market
Using the reference images, the traveler from image 1 walks through the spice market from image 2, pausing to look at a stall as vendors move behind her, slow tracking shot easing beside her, warm afternoon light with dust in the air, documentary mood, keep the traveler's face and outfit consistent with her reference image.Settings: Vidu Q2 Reference-to-Video · 2 reference images · 16:9 · 1080p · 6s · amplitude: medium. Why it works: naming which image supplies the person and which supplies the place, plus a keep-consistent line, lets Vidu fuse two references into one grounded shot.
6. Image-to-video — animate a portrait
Bring this portrait to life: the man blinks naturally, gives a small warm smile and turns his head slightly toward the camera, his hair moving a little in a gentle breeze, keep the lighting, background and face exactly identical to the photo, add a slow subtle push-in, natural and realistic.Settings: Vidu Q2 Image-to-Video · 1080p · 5s · amplitude: small. Why it works: describing only the small added motions and locking the face with "keep identical" produces a believable live portrait without distorting the likeness.
7. An action beat — sprinter off the blocks
A sprinter in a red kit exploding out of the starting blocks on a wet outdoor track, spray kicking up from the surface, muscles taut, fast tracking shot chasing just behind and to the side, hard low morning backlight rimming the body, high-contrast sports grade, motion blur on the limbs.Settings: Vidu Q2 · 16:9 · 1080p · 5s · amplitude: large. Why it works: large amplitude gives the explosive energy, and the chase-behind track gives that energy a clear direction so it does not turn into wobble.
8. A vertical social clip — cafe counter
A hand sliding a slice of chocolate cake with a fork onto a marble cafe counter, a dusting of cocoa drifting off the top, slow push-in framed vertically on the plate, bright soft daylight from a window on the left, fresh appetizing food-content look, shallow focus, quiet cafe ambience.Settings: Vidu Q2 · 9:16 · 1080p · 5s · amplitude: small. Why it works: a tight vertical frame with one small food motion and a near-static camera keeps a macro shot crisp for TikTok and Reels.
9. A dialogue line — synced audio
A weary detective in a dim office leans toward the camera and says clearly, "We're running out of time," his brow furrowing as he speaks, static camera with a subtle slow push-in, cool blue light from a window blind striping his face, tense noir mood, quiet room tone under the line.Settings: Vidu Q2 · 16:9 · 1080p · 5s · amplitude: small · Background Music off. Why it works: writing the exact spoken line and holding the shot to that one delivery lets Vidu's 48kHz synced audio keep the lip-sync clean.
10. A start-end morph — day to night
Smoothly transition from the start frame to the end frame: the same city street shifts from bright afternoon to glowing evening, streetlights easing on and shadows lengthening, a gentle cinematic push-in throughout, matched lighting and composition between the two frames, seamless natural morph, no sudden jumps.Settings: Vidu Q2 Start-End Frame · 16:9 · 1080p · 5s · amplitude: small (keep both frames within a 0.8–1.25 aspect ratio). Why it works: the two frames carry the scene, so a short "transition, keep lighting matched" instruction plus small amplitude is all Vidu needs for a smooth morph.
Movement amplitude & camera moves
Amplitude sets how much motion Vidu adds; the camera move gives that motion a direction. Get the pairing wrong and even a good prompt warps.
Amplitude, matched to the subject. Use auto for portraits and calm faces, small for products, macro shots and smooth transitions, and medium or large for full-body action and fast camera work. When you go to large, always pair it with a clear camera instruction — tracking, dolly, orbit — so the extra energy has somewhere to go. Large amplitude with no camera move is the number-one cause of wobble.
Camera language Vidu follows. Stick to moves it recognises: slow push-in and pull-back for reveals, dolly and tracking to move with a subject, orbit for products and pairs, aerial / drone for landscapes, tilt and pan for scale, time-lapse for changing light, and static / locked-off when the subject's own motion should be the only movement. One move per clip — a stacked "push-in then orbit then pan" gives Vidu nothing to commit to.
Mistakes to avoid
Most bad Vidu clips trace back to a short list of fixable errors. Run through these before you burn credits.
Cramming multiple actions or cuts into one shot. One continuous action per clip. If you need a sequence, use a short Shot 1 / Shot 2 / Shot 3 script and keep each shot to a single beat, or generate separate clips and stitch them.
No camera move. Leaving out the move lets Vidu drift. Name one every time, even if it is just a subtle slow push-in.
Amplitude too high for the subject. Large amplitude on a product or a face causes warping. Drop to small for anything with hard edges or a recognisable likeness.
A vague subject. "A person in a city" gives Vidu too much to invent. Describe the subject concretely so it stays stable.
Asking for readable text on screen. Vidu does not render clean lettering reliably — add titles and captions in your editor afterward.
Expecting more than eight seconds. A single generation runs 2–8 seconds. For longer stories, stitch multiple clips rather than pushing one prompt past its limit.
Mismatched frames in start-end mode. If the first and last frame's aspect ratios fall outside 0.8–1.25 of each other, the transition breaks — match them first.
Frequently Asked Questions
What makes a Vidu prompt look realistic?
Realism in Vidu comes from restraint, not detail-stuffing. Give one clear subject with concrete appearance, exactly one continuous action, and a single named camera move, then let natural lighting do the work. Keep movement amplitude at small for product and close-up shots so surfaces do not warp, and only raise it for full-body action. Vidu Q2 at 1080p and 24fps handles physically plausible motion best when the prompt describes real-world light and one thing happening at a time.
What is the Vidu prompt formula?
Subject and appearance, then action or motion, then camera move, then setting, then lighting and mood, then settings and amplitude. Write it as one flowing sentence of roughly 25 to 60 words rather than a bullet list. The subject anchors the frame, the single action gives Vidu one thing to solve, the camera move gives the motion direction, and lighting sets the realism. Resolution, aspect ratio and length are app settings, so keep them out of the prompt text.
Which Vidu mode should I use for realistic video?
Use text-to-video when Vidu should invent the whole scene. Use image-to-video when you already have a photoreal still and only want it to move — this gives the most realistic result because the look is locked and you describe only the added motion plus keep the face and background identical. Use reference-to-video when specific people, products or places must stay consistent across clips. Use start-and-end-frame when you have a first and last frame and want Vidu to fill a smooth transition between them.
How long should a realistic Vidu prompt be?
Aim for 25 to 60 words for a single shot. That is enough to name the subject, one action, one camera move, the setting and the lighting without burying Vidu in competing instructions. Longer prompts tend to add contradictions that produce warping. If you need more than one beat, use a short Shot 1, Shot 2, Shot 3 script inside one prompt and keep each shot to a single action.
How do camera moves and movement amplitude work together?
Movement amplitude controls how much motion Vidu adds — auto, small, medium or large — while the camera move gives that motion a direction. Use auto for portraits, small for product and smooth transitions, and medium or large for full-body action. Whenever you raise amplitude to large, pair it with a clear camera instruction such as tracking, dolly or orbit, otherwise the extra energy turns into wobble and distortion instead of directed movement.
Why does my Vidu video look warped or blurry?
The usual causes are movement amplitude set too high for the subject, cramming several actions or cuts into one shot, or a vague subject Vidu has to guess at. Drop amplitude to small for close-ups and products, keep to one continuous action per clip, and describe the subject concretely. Also avoid asking for readable text on screen and do not request more than eight seconds in a single generation — stitch multiple short clips instead.
Can Vidu add realistic sound and dialogue?
Yes. Vidu Q2 generates 48kHz sound effects synced to the on-screen action and has a Background Music toggle. Describe the sound in plain language inside the prompt — rain on a window, ambient market chatter, or a specific spoken line — and Vidu triggers matching audio. For a talking character, write the exact line you want spoken and keep the shot to that one delivery so the lip and audio sync stays clean.
Does start-and-end-frame mode need a detailed prompt?
No. In start-and-end-frame mode the two uploaded frames carry the identity and composition, so a brief prompt is enough to steer mood, motion and camera. Keep movement amplitude at small for a smooth morph, and make sure the two frames' aspect ratios are within 0.8 to 1.25 of each other or the transition breaks. A line like a smooth cinematic push-in with matched lighting between the frames is usually all it needs.