Prompting guide
How to prompt Cherry
How to prompt Cherry, Mage's video model: a one-minute quick start, a fill-in template, and expert tips for shots, camera, characters, sound, and edits.
Quick start
Your first great Cherry prompt
Everything a beginner needs in one minute: what to know first, the rules that matter most, a template to fill in, and prompts to copy.
What to know first
A Cherry prompt is a short script, not a photo caption. Say what happens in the order it happens, how the camera films it, and what we hear: Cherry makes sound with every clip. Length, resolution, and aspect ratio are settings, so the prompt only describes what is on screen and on the soundtrack.
The rules that matter most
Write what happens, in order
Lead each beat with a verb (walks, turns, lifts, splashes) and put the beats in the order they happen. A list of adjectives gives a clip where nothing moves.
Give each shot one camera move
Say the shot size and one move: a slow push in, a low tracking shot, a locked-off wide. Two moves in one shot fight each other.
Say what we hear
Name the ambience and the effects. For speech, write who speaks and put the exact words in curly braces: Mara's line (in English, warm): {We made it.}
Keep settings out of the prompt
Choose the length, resolution, and aspect ratio in the settings. Words like "10 seconds" or "4K" in the prompt are read as things to show.
Draft cheap, then finish
Try the motion and timing on Cherry Mini or at 480p, then render the keeper on Cherry 2 Pro. A draft previews movement and timing, not the final look.
Fill in this template
Replace each part in square brackets with your own idea, and leave out a part you do not need.
[Subject] [does what] in [where].
The visuals present [style or mood].
The camera uses [shot size and one move].
The sound includes [dialogue, ambience, effects, or music].A young otter with sleek wet fur slides down a muddy bank into the river on a warm afternoon.
The visuals present a realistic nature documentary in soft natural light.
The camera uses a low shot at water level, following the otter with a slight handheld feel.
The sound includes rushing water, the wet slap of the slide, a splash, and birdsong.[Subject]Who or what the clip is about, with one or two visible details: a young otter with sleek wet fur.[does what]The main action, led by a verb: slides down a muddy bank into the river.[where]The place and the time: a mossy riverbank on a warm afternoon.[style or mood]The look in a few words: a realistic nature documentary in soft natural light.[shot size and one move]How it is filmed: a low shot at water level, following the otter.[dialogue, ambience, effects, or music]What we hear: rushing water, a splash, birdsong. Spoken words go in curly braces after the speaker.
Prompts to start from
A nature moment in timed beats
Realistic nature documentary, natural light. On a warm afternoon on a mossy riverbank, a young otter slides down a muddy chute into the water. The otter has sleek wet brown fur and quick, playful movements. Ferns and smooth river stones line the bank; tall trees blur softly behind. 0-3 seconds: The otter scrambles to the top of the bank, shakes its head, and flops onto its belly at the start of the chute. 3-8 seconds: It slides down fast, splashes into the river, then surfaces facing the camera and rolls onto its back, paws on its chest. Low camera at water level with a slight handheld feel, following the otter. Natural sound only: rushing water, the wet slap of the slide, the splash, birdsong.Why it works: The timed beats tell Cherry 2 Pro what happens when, one camera follows the action, and every sound is named, so nothing is left to chance. Set the length to 8 seconds.
A product reveal
A frosted glass perfume bottle stands on wet black stone in a dark studio. A hand presses the cap: a fine mist bursts out and catches the light, then droplets settle on the stone. The visuals present a premium commercial look with deep shadows and one warm backlight. The camera uses a macro shot with a slow push in. The sound includes the soft hiss of the mist and a single low piano note.Why it works: A product reads best as a start state, one action, and a visible result, and that is all this prompt asks for.
Your Character speaks
Mara (@mara) leans on the counter of an empty diner at night, glances at the door, then looks into the camera with a small smile. Mara's line (in English, low and conspiratorial): {Best pie in town. Don't tell anyone.} The camera uses a slow push in from a medium shot to a close-up. The sound includes the hum of the fridges and a coffee machine hissing in the background.Why it works: The name is bound to the handle once, so the rest of the prompt can say Mara, and the line names its language, delivery, and speaker. Replace @mara with your own Character's handle.
A still brought to life
Use @image1 as the first frame. It shows a woman at a rain-streaked café window at night, cupping a mug in both hands, neon reflections on the glass. She lifts the mug, blows across it, and sips, then glances at her phone as it lights up on the table and breaks into a surprised half-laugh. Slow push in from a medium shot to a close-up. Sound: rain on the glass, low café chatter, the phone's short buzz.Why it works: Upload the still, and the first sentence starts the clip on it; the rest only says what changes. Set the aspect ratio to match the still.
Advanced
Advanced Cherry prompting
Techniques for more control once the basics work.
Which Cherry reads which prompt
Cherry comes as four models that read prompts in two ways. Cherry 2 Pro takes the most detailed direction: whole-second timestamps, multi-view character sheets, larger casts, and dialogue in many languages. Cherry, Cherry Pro, and Cherry Mini follow a simpler script built from numbered shots.
Cherry 2 Pro also looks different from the others, so a Cherry Mini draft previews motion, blocking, and timing rather than the final picture. Work out the action cheaply, then write the final prompt for the model you render on.
- Cherry 2 Pro: time beats with whole seconds ("0-3 seconds") or with Shot 1, Shot 2, and leave out quality words you did not ask for.
- Cherry, Cherry Pro, and Cherry Mini: use Shot 1, Shot 2 only, because they ignore second ranges, and end with one short quality line. With several people, add "No two characters share an identical face and outfit."
- Character images: on Cherry, Cherry Pro, and Cherry Mini use one view per image, such as a headshot and a full-body shot, because a turnaround sheet makes twins. Cherry 2 Pro handles multi-view sheets of up to five people.
- Cast size: one to eight people work best on Cherry 2 Pro, and up to twelve less reliably. On the others, group more than four people into images of four or fewer first, on Mango for example.
- Naming: on Cherry 2 Pro, bind a Character to a name once and use the name after that. On Cherry, Cherry Pro, and Cherry Mini, repeat the name with the mention, as in Mara (@mara), in every shot where that person acts.
A Cherry Mini draft, in numbered shots
Mara (@mara) is a young woman in a yellow raincoat; the street is a narrow cobbled lane at dusk.
Shot 1: Mara (@mara) walks toward the camera along the lane, slowly lowering her hood with her right hand as the rain eases. The camera tracks backward in front of her at walking pace.
Shot 2: She stops under a lit shop sign, looks up at the clearing sky, and lets out a long breath, shoulders dropping. Static medium close-up.
High definition, rich detail, cinematic texture. Her face stays stable without distortion; movement is natural and smooth, with no flicker.Why it works: A setup paragraph, numbered shots with one camera move each, slow and specific movement, and the closing quality line: the script Cherry, Cherry Pro, and Cherry Mini read best.
Structure: write a production brief
Write a Cherry prompt like a note to a film crew, not like ad copy: which picture is what, what happens in order, how it is shot, and what we hear. For a simple clip, the four lines of the quick-start template are enough.
When a clip uses references or has several beats, Cherry 2 Pro follows a labeled brief best. Keep the labels in the prompt, give each reference one line with its single job, and end with what must stay the same.
- Say where everyone is relative to fixed things: the door, the counter, the window. Keep the camera on one side of a conversation.
- End with the final state: who stands where and who holds what. Cherry stops guessing once it knows where the clip ends.
- Keep everything in the prompt about the clip itself: no notes to yourself, and no settings.
The labeled brief, to fill in
Goal: Generate [kind of video]. The core subject is [subject]; the principal event is [one-line summary].
Inputs: [Name] is [@handle]; use only the face, hairstyle and clothing. @image2 is [the place]; use only its layout and light, not the people in it.
Positions: [Fixed features, who stands where relative to doors and tables, who holds what.]
Script: At the start: [state]. Shot 1: [...] Shot 2: [...] At the end: [final positions, who holds what].
Keep the same: [Identities and count, clothing, who owns which prop, directions, which voice belongs to whom.]Why it works: Each label answers one question the model would otherwise guess at: what the clip is, what each picture is for, where everyone stands, what happens, and what must not change.
Several shots in one clip
Mark each beat with Shot 1:, Shot 2:, or, on Cherry 2 Pro, with whole-second ranges that cover the whole clip with no gaps: 0-3 seconds, 3-7 seconds, 7-10 seconds. Moments work too: "At the 5-second mark…", "After 3 seconds…".
Budget one beat or shot for every 3 to 5 seconds of the clip. With too little to do, Cherry improvises; with too much, it adds cuts or skips beats. Use exact seconds only when a beat must land at a moment, such as an entrance or a line said on an action, and never time fast repeated actions.
- Write each shot as its framing in square brackets (shot size, camera position, angle, move), then the action as a chain with what the hands do, then the exact line in curly braces, then an end state.
- One camera move per shot. Name a cut by what triggers it: "cut on the word guest", "at the 6-second mark, a whip pan right into the next room".
- Close a shot on a clear state: "End with the dagger flat on the rug, no hand on it."
An action sequence in timed beats
Rooftop chase at sunrise, saturated color.
0-3 seconds: [Wide shot, locked off] A cyclist in a red jacket sprints across a flat rooftop toward the edge.
3-6 seconds: [Extreme close-up, low angle] The front wheel leaves the edge against the orange sky.
6-10 seconds: [FPV, following] The bike lands on the next roof and skids to a stop, and the rider punches the air, the city spread out below.
The sound includes wind, the hum of the tires, a hard landing thud, and a distant siren.Why it works: The ranges cover all ten seconds with no gaps, each beat has its own framing and one move, and only the jump gets detail. Set the length to 10 seconds.
A dialogue scene with two saved Characters
Goal: Generate a tense dialogue scene in a candlelit tent at night. Hatim offers bread to a guest who hides a dagger.
Inputs: Hatim is @hatim; use only his face, head wrap and cloak. The guest is @guest; use only his face, head wrap and travel cloak.
Positions: Hatim sits on the west side of a low rug, the guest on the east side, a clay plate of bread between them. The camera stays south of them.
Script:
Shot 1: [Medium two-shot from the south, locked off] The guest's right hand rests on a sheathed dagger under his cloak. Hatim tears bread and holds out a piece. Hatim's line (in English, warm): {Eat. You are my guest.}
Shot 2: [Close-up on the guest, slow push in] His eyes move from Hatim's face to the bread. He lets go of the hilt and takes the bread with that same hand. Hatim keeps his mouth naturally closed.
Shot 3: [Medium two-shot from the south, locked off] Both men eat in silence. End with the bread in the guest's hand and the dagger still sheathed.
Keep the same: Two men only, seats unchanged, the dagger stays sheathed. No subtitles. Sound: candle crackle, cloth rustle, wind against the tent.Why it works: Fixed seats and a camera that stays on one side keep the scene readable, every shot has one move and a clear end, and the listener keeps his mouth closed so only the right person speaks.
Camera and motion
Named camera techniques work as written. For anything niche, give the term and what it looks like on screen: "Rack focus: the foreground goes out of focus while the woman behind it comes into sharp focus."
A camera move needs a subject and a path: the move, who it centers on, where it starts, which way it goes, and where it arrives.
On Cherry 2 Pro, describe most action briefly ("they trade blows across the deck") and save the detail for the one or two moments that matter; never repeat an action. On the other Cherry models, spell movement out: the body part, how far, how fast ("slowly raises her right hand to shoulder height"), and how one action flows into the next. Slow, continuous movement looks better there than sprints and flips.
- Show feelings as visible behavior, two or three cues each: sad is a lowered head, trembling shoulders, and welling tears; anxious is checking a watch, tapping, and darting eyes; relieved is a long breath out, dropping shoulders, and a faint smile.
- Cherry 2 Pro overreacts to extreme words such as "fanatical" or "utterly shocked" and can give people glowing eyes. Use plain words ("amazed"); if it happens anyway, add "normal human eyes, no glowing eyes".
- Name a style outright: "2D Japanese anime style". Realistic reference photos pull a stylized clip toward live action, so restyle the still first, on Mango for example.
- For a product, write a start state, one operation, and its visible result, one per stage.
| Write | What it does |
|---|---|
one-take | A single unbroken shot with no cuts. |
dolly zoom | The camera moves while the lens zooms the other way, so the background stretches or squeezes behind the subject. |
FPV | A fast, swooping drone view, as if flying through the scene. |
bullet time | Time nearly stops while the camera sweeps around the subject. |
speed ramp | The action slows down or speeds up in the middle of the shot. |
whip pan | A fast sideways turn that blurs into the next view; a good way to cut. |
locked off | The camera does not move at all. |
handheld | A slight natural shake, as if someone is holding the camera. |
References and Characters
Mention saved Characters and References (outfits, locations, poses, objects, and voice clips) by their @handle, and images you upload as @image1, @image2. Before Cherry reads the prompt, Mage turns every mention into a numbered picture: @mara becomes "character from image 2", and a Character with a saved voice becomes "character from image 2 with voice from Audio 1". So say in words what each picture is for.
Bind each Character to a name the first time they appear ("Mara (@mara) walks in…"); on Cherry 2 Pro, call her Mara after that, and on the other Cherry models repeat "Mara (@mara)" in each shot she acts in. Give each person two or three fixed visible features, and mention people in the order they first appear on screen: a mismatched order is the usual cause of swapped faces and voices.
- Give every picture exactly one role: "use only the face and clothing", "use only the layout and light; do not use the people in it".
- Send only what the clip uses. One or two Character images, one place, and one voice clip is a strong mix; filling every slot muddies the result.
- Never write names as text on a reference image. Bind names in the prompt.
- To start on a still you like, upload it and open the prompt with its own sentence, "Use @image1 as the first frame.", then one sentence on what that frame shows. Set the aspect ratio to the still's.
- Cherry 2 Pro follows keyframes: upload them in order, open with "Use images 1 to 5 in order as keyframes.", and describe the path between them.
- Cherry 2 Pro follows a storyboard sheet: one clean line-art image of up to 15 panels with little text. It guides the plot loosely, so give the reading order and still write every shot.
- High-resolution AI images used as references on Cherry 2 Pro can leave a fingerprint-like texture on grass and leaves. Downscale them to the output resolution before you upload them.
Sound and dialogue
Cherry reads three kinds of sound by their brackets: spoken lines in curly braces {…}, sound effects in angle brackets <…>, and music in parentheses (…). Keep names out of the angle brackets.
Write each line as the speaker, then the language and delivery in parentheses, then the words: Mara's line (in English, low and steady): {Then we hold the ridge.} Label every line's speaker and keep one spoken language per clip, apart from names. If English might come out in another language, add "Use English".
- Do not repeat or explain the words after a line: that tends to summon subtitles. One delivery note per line, inside the parentheses.
- Listeners "keep their mouths naturally closed". For a shot with no speech: "Characters keep their mouths naturally closed; there is no narration; no subtitles or signs appear on screen."
- With a voice clip, describe the voice in words too ("in the deep, slightly gravelly voice of Audio 1"). The prompt decides the words; the clip gives the timbre, accent, pace, and emotion. Mage keeps only the opening seconds of each clip, so a voice clip should start speaking at once.
- A clip takes up to three sounds: voice clips you mention first, then Characters' saved voices. Turn off Use Character Voices to leave the saved voices out.
- On Cherry 2 Pro, set the sound bed in words: "No BGM; only ambient and action sounds." or "No audio." If music still appears, list its names (music, background music, BGM, score, melody) at both the start and the end of the prompt.
- Words about water, waves, and echoes can add rippling or echo artifacts to the sound. Use them only when you want that sound.
Text on screen
Keep on-screen text short, and say when it appears, where, how it enters, and how it looks: At the 2-second mark the word "OPEN" fades in at the bottom center in white bold sans-serif.
- Ask for subtitles when you want them ("Display subtitles at the bottom center, synchronized with the speech.") and write "No subtitles." when you do not.
- Exact logos and labels come from a picture of them: upload it and write "the label exactly as in @image3".
- Cherry 2 Pro misspells long words. Spell a word out letter by letter, appearing in sequence, or supply it as an image. Captions that must be exact are best added in an editor afterwards.
Editing a video you upload
Upload a source clip and Cherry edits it. On Cherry 2 Pro the result keeps the source's length and aspect ratio, and sources under about 20 seconds work best. Use an edit verb (edit, add, insert, remove, replace, change to) and call the clip "video 1" in plain words: writing "reference video 1" makes Cherry generate a new clip instead of editing yours.
For a replacement, say exactly what changes and that everything else stays. Cherry 2 Pro follows a labeled edit brief best, and the new object should take over the old one's timing and path.
- Edit part of a clip by time on Cherry 2 Pro: "From 4-6 seconds in video 1, change the man's action from drinking coffee to mopping the floor; leave the rest unchanged."
- Edit only the sound: "Remove the background music from video 1; keep the dialogue, lip sync, ambience, action sounds and all visuals." Or translate it: "Translate the spoken dialogue in video 1 into Spanish, with no subtitles. Match the lip movements to the new speech and keep everything else unchanged."
- Turn on Keep Original Audio for a visual edit that must keep the clip's own sound.
- A new length, a new aspect ratio, or a continuation is not an edit on Cherry 2 Pro. Start a new clip on the source's last frame ("Use @image1 as the first frame.") or use Lemon. On the other Cherry models, "Extend video 1 forward: …" continues a clip, though faces wear down over chained extensions.
- Use a clean source: subtitles burned into the source come back in the result.
Swap a moving object, keep everything else
Edit: Edit video 1, replacing only the red bicycle and its rider with the dark-gray electric patrol cart from @image1.
Source video: video 1 is the sole editing master and governs the park path, the two people on the bench, the camera, the rider's motion and the event order.
New material: @image1 is used only for the cart's body, color and clear windshield; do not use its background or driver.
Scope: There is exactly one cart throughout. Except for the objects explicitly modified above, all other visible characters, props, and background elements in video 1 remain unchanged and are not to be replaced or removed.
Timing: The cart takes the rider's exact timing, path, speed and occlusion in front of the bench. The bicycle and rider no longer appear.Why it works: It names the one thing that changes, takes only the cart's look from the image, and hands the cart the bicycle's exact timing and path. With Keep Original Audio on, the park keeps its own sound.
Common mistakes
Avoid this, do this instead
Avoid: A still description: adjectives, and nothing happens.
A beautiful woman in a red coat on a neon street at night, cinematic, stunning, moody.Instead: A short script led by verbs, in the order things happen.
A woman in a red coat walks down a neon street at night, stops under a flickering sign, and turns toward the camera as the rain starts. The camera tracks backward in front of her, then holds. The sound includes rain on the pavement and a distant tram bell.Avoid: Two camera moves in one shot.
Instead: One move per shot, and a cut to the next.
Avoid: A picture number followed by a verb: Cherry reads it as a count.
@image2 runs toward the door.Instead: A description or a name, with the mention after it.
The woman in the yellow coat (@image2) runs toward the door.Avoid: Settings in the prompt: "10 seconds, 1080p, 16:9, with sound".
Instead: The length, resolution, and aspect ratio in the settings. Cherry makes sound by default.
Avoid: Second ranges on Cherry, Cherry Pro, or Cherry Mini.
Instead: Shot 1, Shot 2 on those models; whole-second ranges only on Cherry 2 Pro.
Avoid: Quality and negative words on Cherry 2 Pro: "8k, masterpiece, no blur".
Instead: Nothing, or the light and the lens that make the look you want.
Avoid: A whole story pasted into one clip.
Instead: One beat for every 3 to 5 seconds, and a longer story split into several clips.
Avoid: Extreme feelings: "fanatical", "utterly shocked".
Instead: Plain words ("amazed") and two or three visible cues.
Avoid: Every reference slot filled.
Instead: Only the pictures the clip uses, each with one job.
FAQ
Cherry prompting questions
How long should a Cherry prompt be?
Does Cherry support negative prompts?
How do I make a Cherry clip without music?
Why did my characters swap faces or voices?
Which languages can Cherry speak?
Is a Cherry Mini draft a preview of the final clip?
How do I keep a character the same from clip to clip?
Prompting guide
More prompting guides
How to prompt Berry
How to prompt Berry, Mage's reference-faithful video model: a one-minute quick start, a template, and expert tips for references, timing, dialogue, and edits.
How to prompt Blueberry
How to prompt Blueberry, Mage's first-and-last-frame video model: a quick start, a template, and expert tips for transitions, loops, camera moves, and sound.
How to prompt Grok Video
How to prompt Grok Video, xAI's Grok Imagine, on Mage: a quick start, a fill-in template, and expert tips for camera, references, edits, and sound.
How to prompt Kiwi
How to prompt Kiwi, Mage's everyday video model: a one-minute quick start, a fill-in template, and tips for light, camera moves, start images, and sound.
How to prompt Lemon
How to prompt Lemon, Mage's all-round video model: a one-minute quick start, a template, and expert tips for frames, references, voices, shots, and edits.
How to prompt LTX Video
How to prompt LTX Video on Mage: a quick start, a fill-in template, and expert tips for the 120-word rule, start frames, camera moves, and fast drafts.
Make your first video with Cherry 2 Pro
Start free, then run Cherry 2 Pro on Gems or a plan. No install.
Try Cherry 2 Pro