The basic formula
Every model on NovaMotions AI reads plain language, so there is no special syntax to learn. What makes the difference is order and focus. Describe who or what is on screen, what happens, where it happens, and how it is filmed. Then add sound if the model supports it.
Think of it as briefing a camera crew. A crew needs to know the subject and the action first. The setting, the look and the camera move come next. The mood and the sound finish the picture.
Template
Subject + action, then setting and look, then camera move, then sound and mood.
Subject and action
Be specific about who is on screen and what they do. A young chef flips a pancake reads better than a person cooking. One main action per clip is clearer than three.
Setting and look
Time of day and light change everything. Add the place, the light and a style, like golden hour on a rooftop, shot on 35mm film, warm and soft.
Camera and sound
Name the move in plain words: slow push-in, low tracking shot, handheld follow, drone pull-back. Then the sound, like rain on glass or a crowd cheering, and any spoken line in double quotes.
Night market cook
A street cook in a white apron tosses noodles in a flaming wok at a busy night market, sparks rising into the steam. Warm string lights and neon signs behind her. Slow push-in from a medium shot to a close-up of the wok. Sizzling, clanging metal and lively chatter.
Lighthouse keeper
An old lighthouse keeper climbs the last steps of a spiral staircase and lights the lamp as a storm rolls in over the sea. Cold blue dusk with the warm beam cutting through the rain. Low angle tracking shot up the stairs, then a slow tilt to the beam. Howling wind and thunder.
Multi-modal control with @ mentions
Words describe a scene. Reference files pin it down. In Media mode you can upload images, short video clips and audio files, then point to each one in your prompt by its tag. Files are numbered in the order you add them: @Image1, @Image2, @Video1, @Audio1 and so on. Click a file's tag in the media bar to drop it into the prompt.
How many files you can add depends on the model. Seedance 2.0 Pro, for example, takes up to 9 images, 3 video clips and 3 audio files. The limits for every model are listed on its workspace page.
Template
Take [what] from @File, [what] from @File, then describe the new scene.
Give each file one job
A face from @Image1, an outfit from @Image2, the camera move from @Video1, the beat from @Audio1. A file with a clear job is copied faithfully.
Say what to take
Reference the character in @Image1 is vague. Keep the face and hairstyle from @Image1 is precise. Name the part you care about.
Keep clips and audio short
Reference clips and audio work best between 2 and 15 seconds. Audio needs at least one image or video in the same generation.
Your host, your track
The woman from @Image1, wearing the leather jacket from @Image2, walks through a rainy neon street and turns to the camera with a smile. Copy the slow handheld follow from @Video1 and cut the steps to the beat of @Audio1.
What to upload
- @Image1A clear portrait of your character
- @Image2The jacket on a plain background
- @Video1A short walking shot with the camera move you like
- @Audio1A music loop of a few seconds
Text in video: slogans, subtitles and speech bubbles
Models can draw short text into the frame, and the ones with native audio can also speak it. Text works best when it is short, quoted exactly and given a time and a place on screen. For anything longer than a few words, add it in your editor afterwards.
When you write the line in another language, the voice and any subtitles usually follow that language.
Template
"The exact words" + when they appear + where on screen + how they look.
Slogans
Put the words in double quotes, say when they appear and describe the type, like bold white letters fade in at the center at the end.
Subtitles
Ask for subtitles at the bottom of the frame that match the spoken line. Keep the line to one or two short sentences.
Speech bubbles
Name the speaker and the line, then ask for a comic-style speech bubble next to them. It suits illustrated and playful styles best.
End-card slogan
Macro shot of a cold glass of lemonade on a sunny wooden table, condensation running down the glass, a slice of lemon dropping in with a splash. The picture softly blurs and bold white letters reading "Squeeze the day" fade in at the center. Bright summer light, fizzing sound.
Voice with subtitles
A calm narrator speaks over a slow time-lapse of stars fading into dawn above a mountain lake: "Every morning, the world starts again." Matching subtitles appear at the bottom of the frame in clean white text. Soft ambient music and a light breeze.
Comic speech bubble
Hand-drawn comic style. The kid from @Image1 holds up a lopsided sandcastle on a sunny beach and shouts, "Best one yet!" A white speech bubble with his line pops up beside his head. Waves and seagulls in the background.
What to upload
- @Image1A drawing or photo of your character
Image references: multi-view and multi-element
Image references are the most reliable way to keep a character, a product or a place exactly as it looks. There are two ways to use several images. Multi-view means several angles of the same subject, so the model understands its shape from every side. Multi-element means different subjects in different images, combined into one scene.
Clean, well lit images on simple backgrounds work best. Avoid heavy filters and text overlays unless the text is part of the product.
Template
Keep [subject] from @Image1 to @Image3 consistent, then describe the new scene and the move.
Multi-view
Upload the front, side and back of the same product or character. Ask the camera to orbit and the model can show every side without inventing details.
Multi-element
One image per element: a person, an outfit, a location, a logo. Say which element comes from which file and how they meet.
Storyboards
An image of a storyboard or a comic page can set the shot order. Ask the model to follow the panels from left to right.
Product orbit
The camera from @Image1, @Image2 and @Image3 sits on a white table in a bright studio. Start on a close-up of the lens, then slowly orbit around it to show the front, the side and the back. Soft shadows, clean reflections, a quiet shutter click at the end.
What to upload
- @Image1Front of the product
- @Image2Side of the product
- @Image3Back of the product
Cafe meet-cute
Inside the cafe from @Image4, the barista from @Image1, wearing the apron from @Image2, slides a latte across the counter to the customer from @Image3. He smiles and says, "You remembered." Warm afternoon light, slow push-in, soft cafe chatter.
What to upload
- @Image1The barista
- @Image2The apron
- @Image3The customer
- @Image4The cafe interior
Video references: motion, camera and effects
A short clip can teach a model things that are hard to describe: a dance, a fighting move, a whip pan, a particular transition. Upload it in Media mode and say exactly what to copy from it. The new clip is rendered from scratch, so you keep the movement while the characters, the place and the style come from your other files and your words.
Template
Copy the [motion, camera move or effect] from @Video1, with [subject] from @Image1, in [new scene].
Motion
Dances, sports moves and gestures transfer well. Keep the full body in frame in your reference and ask for the same in the result.
Camera
Drone reveals, dolly zooms and orbit shots can be copied from a reference clip and applied to a completely different scene.
Effects
Transitions, morphs and stylized effects can be borrowed too. Describe the effect in words as well, so the model knows what you are pointing at.
Borrowed choreography
The dancer from @Image1 performs the routine from @Video1 on a rooftop at sunset, city skyline behind her. Keep her full body in frame, steady wide shot with a gentle push-in on the final pose. Upbeat music with a strong beat.
What to upload
- @Video1A short dance clip, full body in frame
- @Image1Your dancer, full body
Same move, new world
Copy the camera move from @Video1, a fast low glide that rises into a reveal. Use it to fly through a glowing alien forest and rise to reveal a floating city above the trees, styled like the painting in @Image1. Deep hum and soft chimes.
What to upload
- @Video1A clip with the camera move
- @Image1A style or concept painting
Ink transition
Recreate the ink splash transition from @Video1. A portrait of a calm woman dissolves into swirling black ink that reforms as a flock of birds flying over a misty lake. Smooth, elegant motion, a soft whoosh as the ink moves.
What to upload
- @Video1A clip with the transition effect
Editing and extending clips
Each generation is a single clip, up to 15 seconds on the longest models. For longer stories, build them clip by clip. Open a finished clip in your results or in History and press Continue from last frame. A new clip starts from the exact final frame, so the character, the light and the setting carry over, and you only describe what happens next.
To change an existing clip, treat it as a remix. Add the clip as @Video1 in Media mode, describe the change, and keep the rest of the prompt close to the original. The new version is rendered fresh, so small differences are normal.
Template
Continuing from this frame: [what happens next], [camera], [sound].
Describe only what is new
The model already sees the last frame. Spend your words on the next action and the next camera move.
Repeat the style words
Copy the light, palette and lens words from the first prompt into every follow-up so the look does not drift.
Use first and last frame
In Image mode, a first and a last frame give you a planned transition. Chain several to build a sequence with matched cuts.
The next beat
Continuing from this frame: she reaches the edge of the cliff, pulls off her hood and looks out over the misty valley as the sun breaks through. The camera rises slowly behind her into a wide reveal. Wind and distant birdsong, same cold morning light.
Remix the season
Keep the camera move, the timing and the house from @Video1, but turn the season to deep winter: snow on the roof, frost on the windows and warm light glowing inside. Match the snowy sky in @Image1. Quiet wind and a soft crunch of snow.
What to upload
- @Video1The clip you want to remix
- @Image1A photo of the winter look you want
Image prompts
Image prompts follow the same logic without the motion. Describe the subject, the composition, the light, the lens or style, and the color palette. If the image needs words on it, quote them exactly and say where they go.
In Image edit, upload a source image and describe only the change, plus what must stay the same. Most image models also take up to 3 extra reference images for products, faces or styles.
Template
Subject + composition + light + style or lens + palette + "exact text" and where it goes.
Composition first
Say where things sit: centered, rule of thirds, top-down flat lay, wide shot with lots of sky. Layout words matter more than adjectives.
Text that reads
Short headlines in double quotes, with a position and a type style. Ask for one or two pieces of text per image.
Edits that stay put
Start edits with what to keep, like keep the face, pose and lighting, then describe the change in one sentence.
Gig poster
Bold screen-printed gig poster, a giant orange moon rising over a desert highway with a lone van, halftone texture and three-color print look in orange, teal and cream. Big condensed headline at the top reading "Desert Nights", small line at the bottom reading "Live at the Dunes".
Window light portrait
Editorial portrait of an older man with a grey beard reading by a tall window, soft side light, dust floating in the beam, muted green and brown palette. 85mm lens, shallow depth of field, natural skin texture, quiet and thoughtful mood.
Swap the backdrop
Keep the sneaker exactly as it is, same angle, laces and logo. Replace the background with a wet concrete street at night lit by neon signs, add a soft reflection under the shoe and match the colored rim light.
Music prompts and lyrics tags
AI Music reads a short description of the song: genre, mood, instruments, tempo and the kind of voice. A comma-separated list works well. Add your own lyrics in the lyrics box and mark the parts with tags so the song has a clear structure.
Leave the lyrics empty and they are written from your description. Switch on Instrumental for a track with no vocals.
Template
Genre, mood, two or three instruments, tempo in bpm, voice type.
Structure tags
Use [verse], [chorus] and [bridge] on their own lines above each part. [intro] and [outro] help with longer songs.
Match length and lyrics
Two to four lines suit a 30 second jingle. A full song of 2 or 3 minutes needs verses, a chorus and a bridge.
Tempo and voice
A bpm number and a voice description, like warm male vocals or airy female vocals, steer the feel more than adjectives do.
Road trip anthem
upbeat indie pop, jangly electric guitars, handclaps, driving drums, bright and carefree, 120 bpm, energetic female vocals
Lyrics Paste these into the lyrics box
[verse] Windows down and the map on the floor We don't know where, we just want more [chorus] Turn it up, let the highway sing We've got the summer and everything
Focus beat
lo-fi hip hop, dusty drums, warm Rhodes chords, soft vinyl crackle, mellow bass, calm and focused, 72 bpm, instrumental
Best practices and common mistakes
Most disappointing results come from a handful of habits. Fix these and every model on NovaMotions AI gets noticeably better.
Do
- Change one thing at a time between takes, so you know what made the difference.
- Draft in a fast, low cost model or a low resolution, then render the keeper at full quality.
- Keep the seed when you like a result and only want to adjust a detail.
- Write spoken lines in double quotes and say who speaks.
- Match the aspect ratio to where the clip will play, like 9:16 for stories and 16:9 for widescreen.
Avoid
- Stacking many actions into one short clip. Split them into separate clips instead.
- Describing what is already in your first frame instead of what should move.
- Contradicting yourself, like a sunny night or a still shot with a fast camera move.
- Long paragraphs of on-screen text. Keep it to a few words.
- Uploading references without saying what to take from them.
Ready to try it?
Open the flagship workspace, paste one of the examples and change one detail at a time.