Prompting guide
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.
Quick start
Your first great Lemon 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
Lemon works in one of three ways, set by what you add. With text only, the prompt carries everything: the look, the action, the camera, and the sound. With a first frame, a last frame, or both, the pictures carry the look and the prompt carries what happens. With references (saved Characters and References, images you upload, reference videos), the prompt says what each one is for, then tells the story. Frames and references never mix in one clip. Length, resolution, and aspect ratio are settings, so the prompt only describes what we see and hear.
The rules that matter most
Choose frames or references first
A first and a last frame pin the opening and closing pictures; Characters, References, and reference videos keep people, places, and motion on model. One clip uses one kind or the other, so decide before you write.
Write the beats in order, with the camera
Say who does what, in the order it happens, and how it is filmed: "wide shot, 24mm, camera static", "close-up, slow push in". Lead each beat with a verb.
Say what we hear, and what we don't
Lemon makes sound with the clip and may add speech or music of its own. Put lines in quotes after the speaker, name the sounds, and write "No dialogue." or "No background music." when you want them gone.
Give each shot 2 to 5 seconds
Past about 6 seconds, write numbered shots with times that cover the whole clip. To stop Lemon from cutting at all, make "Generate single shot." the first line.
Put exclusions in the prompt
Lemon has no negative prompt setting. End with one short line of what must not appear: "Negatives: no subtitles, no watermark, no face distortion."
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 order] in [where].
Camera: [shot size and one move].
Sound: [dialogue, ambience, effects, or music].
[Style, color, and mood].A barista with a silver nose ring lifts a steaming cup from the counter, turns, and carries it to the window table in a small corner café at dawn, steam trailing behind her.
Camera: a medium shot tracking left with her at waist height, then settling as she sets the cup down.
Sound: café chatter and a ceramic clink as the cup lands. No dialogue. No background music.
35mm film look, warm amber light against cool blue windows, calm and unhurried.[Subject]Who or what the clip is about, with one or two visible details: a barista with a silver nose ring.[does what, in order]The action as a chain of verbs: lifts a steaming cup, turns, and carries it to the window table.[where]The place and the time: a small corner café at dawn.[shot size and one move]How it is filmed: a medium shot tracking left with her at waist height.[dialogue, ambience, effects, or music]What we hear: café chatter, a ceramic clink. Add "No dialogue." when nobody should speak.[Style, color, and mood]The look in three parts: an art style, a color tone, and a mood. 35mm film, warm amber against cool blue, calm.
Prompts to start from
A short clip in one paragraph
A red fox trots across a frozen lake at sunrise, stops, and pounces nose-first into a snowdrift, then pops up shaking snow from its ears. Low tracking shot at ice level, following the fox, then holding as it pounces. Sound: wind over the ice, the crunch of paws, a muffled thump in the snow. No dialogue. No background music. Nature documentary look, pale gold light, crisp and quiet.Why it works: A short clip needs one paragraph: verbs in order, one camera with a path, every sound named, and the silence stated so Lemon adds no narrator.
Between a first and a last frame
Generate single shot. She tosses the towel over her shoulder and spins once; as she turns, the rows of washing machines fold away and the laundromat becomes a mirrored ballroom around her, chandeliers lighting one by one. The camera circles her slowly at chest height and ends facing her. The hum of the dryers fades into a soft waltz and the click of her heels on the parquet. No dialogue.Why it works: Upload the two stills as First Frame and Last Frame. They carry the look, so every word goes to the path between them, with one camera move as the bridge and no cut.
Your Character speaks
Generate single shot. Mara is @mara, adopting her face, hairstyle and clothing. Mara turns from the rain-streaked window of a dim apartment, walks to the door, and stops with her hand on the latch. She looks back at the camera and says quietly: "I'm ready." Lip sync. Camera: handheld follow at shoulder height. Warm lamp light against cool blue street light. Sound: rain on the glass, a floorboard creaking. No background music.Why it works: One line rules out cuts, the next gives the Character one job, her saved voice speaks the quoted line, and "Lip sync" tightens her mouth to the words. Replace @mara with your own Character's handle.
A three-shot story from text
A short, warm comedy about a lost kite; quick at first, gentle at the end. Shot 1 [0-3s]: Wide shot, 24mm, camera static. A small boy in a yellow raincoat runs along a windy beach as his red kite snaps loose and sails away. Shot 2 [3-6s]: Hard cut. Medium shot, handheld follow. He chases it down the sand, arms out, and stumbles to his knees. Shot 3 [6-9s]: Close-up, slow push in. A fisherman's weathered hand holds out the kite string; the boy looks up and grins. Sound: wind, surf, gulls, a light ukulele melody entering in Shot 3. No dialogue. Negatives: no subtitles, no watermark, no extra characters.Why it works: One sentence sets the tone and pacing, the three shots cover all nine seconds with no gaps, and each has its own framing. Set the length to 9 seconds.
Advanced
Advanced Lemon prompting
Techniques for more control once the basics work.
Three modes: text, frames, or references
Lemon picks its mode from what you add. Text only: the prompt describes everything. Frames: a First Frame, a Last Frame, or both, and the clip opens or ends on your pictures, so the prompt describes the motion, plot, dialogue, and camera, not the look. References: saved Characters and References by @handle, images you upload, and reference videos; the prompt gives each one a job, then tells the story.
Frames and references are separate modes. A Character or voice mention in a clip with frames adds a reference, and the clip is refused. With frames, describe a person by what the frame shows: "the woman in the green coat".
- To open on an exact still and keep a saved Character, make the still with the Character on an image model such as Mango first, then use it as the First Frame.
- Leave Prompt Extend off when you wrote the prompt: it rewrites your words. Turn it on only for a one-line idea you want Lemon to flesh out.
- Enable Audio is on by default. Turn it off for a clip you will score yourself.
Structure: one line of intent, then the shots
A short clip is one paragraph: who, the beats in order, the camera, the sound, the look. For anything longer than about 6 seconds, or with several people, Lemon follows a layout best: one overall sentence, one line per reference, numbered shots with times, then the sound, the style, and the negatives. Leave out any part you do not need.
Put the pacing in the overall sentence as a short arc: "pacing shifts from steady restraint to explosive release", or "brisk build, sudden conflict, short climax, warm ending". Lemon reads it as the shape of the whole clip.
- Write each shot as if it were its own short prompt: the framing and lens, the action, the expression, the line, and the details that move.
- Stay inside your story: its people, counts, props, events, ending, and lines. Add camera, light, and sound craft, not new plot.
- The length is a setting, but plan how many events fit it.
The layout, to fill in
[One sentence: genre, point of view, mood, and pacing.]
[In reference mode, one line per reference: who or what it is, and what to take from it.]
Shot 1 [0-3s]: [Shot size, lens, camera move]. [Action and expression]. [Line, if any].
Shot 2 [3-6s]: [...]
Sound: [Dialogue language and texture, ambience, effects, music or none].
Style: [Art style, color tone, mood].
Negatives: [Only what must not appear].Why it works: Each line answers one question Lemon would otherwise guess at, in the order it reads them best.
A short drama in four shots
A third-person short drama about a missed train and a kind stranger; restrained at first, warm at the end.
Shot 1 [0-3s]: Wide shot, 24mm, camera static. A young man in a grey overcoat sprints onto an empty night platform as the last train's doors close; his breath fogs in the cold.
Shot 2 [3-6s]: Hard cut. Close-up, 85mm telephoto, shallow depth of field. He bends forward, hands on his knees, and laughs once at himself.
Shot 3 [6-9s]: Medium shot, slow push in. An older woman on the bench beside him holds out half of her sandwich and says: "Next one's in forty minutes. Sit."
Shot 4 [9-12s]: Medium two-shot, camera static. He sits, takes it, and replies: "Thank you. I'm Daniel." The station lights hum.
Sound: quiet platform ambience, a train fading into the distance, soft piano entering in Shot 4.
Style: 35mm film look, cool blue platform light against warm sodium lamps, fine grain.
Negatives: no subtitles, no watermark, no extra characters, no face distortion.Why it works: The overall sentence sets the arc, four shots tile twelve seconds with a lens and a move each, every line has its speaker, and the negatives list only what must not appear. Set the length to 12 seconds.
Shots, cuts, and one continuous take
Shots should cover the clip end to end, with no gaps or overlaps, at 2 to 5 seconds each: two or three shots for 8 seconds, three to five for 15, more for longer clips. Pick one time format, Shot 1 [0-3s] or Shot 1 (00:00-00:03), and keep it for the whole prompt.
Lemon may cut on its own, especially in longer clips. When you want one unbroken shot, such as a bridge between two frames, a loop, or a long take, make "Generate single shot." the first line.
- Name the transition at the end of a shot ("hard cut", "dissolve"), or open the next one with "continues seamlessly from the end of the previous shot".
- Give a shot one clear action. Fast repeated actions ("shakes his head three times a second") smear.
- Use the same name for a person in every shot, and restate their look in later shots of a long clip so faces and outfits hold.
A long take with no cuts
Generate single shot. A Steadicam long take through a busy night market. The camera starts low behind a girl in a red hoodie as she weaves between food stalls, rises over a cloud of grill smoke, glides past a cook tossing noodles in a flaming wok, and settles in front of her as she stops at a lantern stall and looks up, the lanterns swaying overhead.
Sound: sizzling woks, vendors calling out, a scooter passing, a radio playing somewhere. No dialogue.
Style: neon and tungsten light from the stalls, warm orange against teal shadows, lively.
Negatives: no cuts, no subtitles, no watermark.Why it works: The first line rules out cuts, and the camera path has a start, waypoints, and an end, so a long clip has somewhere to go. Set the length to 15 seconds or more.
Camera, lens, and motion
Lemon reads camera language as written. Pair the shot with a lens and a depth of field ("85mm telephoto, shallow depth of field", "24mm wide-angle, deep focus", "anamorphic widescreen", "extreme low angle"), and give a move its measure: "the camera orbits 120 degrees around him".
Physical detail sells realism: cloth that sways, dust, sparks, steam, reflections in puddles, and light from sources inside the scene, such as a desk lamp or a neon sign.
| Write | What it does |
|---|---|
push in / pull out | The camera travels toward or away from the subject. |
orbit | The camera circles the subject. Give the angle: "orbits 120 degrees". |
handheld follow | The camera walks with the subject, with a slight shake. |
whip pan | A fast sideways turn that blurs into the next view. |
snap zoom | A sudden zoom onto a detail. |
crane descending | The camera sinks from high above toward the subject. |
FPV drone | A fast, swooping flight through the scene. |
Steadicam lateral tracking | A smooth sideways glide alongside the subject. |
Fixed shot, camera static, position unchanged. | Locks the frame completely. |
high-speed photography, 1000fps slow motion | Deep slow motion for splashes, sparks, and impacts. |
fast, then slow, then fast | A speed ramp inside one shot. |
Style, color, and mood
Name the look as a trio, at the start or the end of the prompt: an art style, a color tone, and a mood. Styles: 35mm film, line illustration, clay animation, motion graphics, ink wash, minimalist keynote business style. Tones: high-saturation warm yellow, cyan-blue against gilded gold, low-key high contrast. Moods: tragic and unyielding, light and healing, oppressive and tense.
For a polished finish, add "cinematic, film grain, high dynamic range", or "IMAX-scale spectacle" for scale. One clear trio sets a clip better than a pile of adjectives.
First and last frames
A frame already fixes the subject, the place, and the style, so spend the prompt on what happens: motion, plot, dialogue, and camera. Describing the frame again only adds noise.
With both frames, describe the path between them. When they differ a lot, name how the clip gets there: by a camera move ("the camera rises and swings left") or by the subject ("she runs toward the door").
- A Last Frame works on its own too: the clip ends on your picture, so describe what leads up to it.
- With frames, the clip takes the shape of the frames rather than the aspect ratio setting.
- Timed shots still work with frames, which keeps a long clip on track.
- For a story longer than one clip, save the last frame of each clip and use it as the First Frame of the next.
- For a loop, use the same picture as First Frame and Last Frame, start with "Generate single shot.", and have the motion return to where it began.
Two frames, one take
Generate single shot. The barista lifts the steaming cup from the counter, turns, and carries it to the window table as steam trails behind her and the pendant lights sway slightly. The camera tracks left with her at waist height, then settles as she sets the cup down. Café chatter, a ceramic clink as the cup lands. No dialogue. No background music.Why it works: Nothing about her looks or the café: the frames hold those. The prompt is the motion, the camera path that joins the frames, and the sound.
References and Characters
Mention saved Characters and References (outfits, locations, poses, objects, and voice clips) by their @handle, images you upload as @image1, @image2, and reference videos as Video 1, Video 2, in the order you added them. Before Lemon reads the prompt, Mage turns every mention into a number: @mara becomes "character from image 2", or "character from image 2 with voice from [Audio 1]" when she has a saved voice. Your uploads come first, then Characters, then References, each in the order you first mention them.
Give every reference its own line and one job, and say what it lends: "Mara is @mara, adopting her face, hairstyle and clothing." "The scene follows @image1, adopting its layout, architecture and lighting; ignore the person in it." "Take only the camera movement from Video 1; nothing in it appears on screen."
- With several people, add a cast line: "Cast: Mara is @mara; Jun is @jun. Mara and Jun are two different people and must not look alike; each look stays locked to its own reference, with no mixing or swapping."
- Several pictures of one person: say they define one person together and that only one of her appears.
- Borrow only part of a picture: "Take only the color grade and lighting from @image2, not its people."
- Motion from a video: "Replace the skateboarder in Video 1 with Mara, keeping the skateboarding action, path and scene unchanged." When a video lends only motion, say that its look is not used.
- A storyboard sheet: "Use @image1 only as shot guidance; do not animate the storyboard as a single image." Then restate the panels left to right, top to bottom. The shots need not match the panels one to one.
- When a person is used for the wrong role, recount the numbers: uploads first, then Characters, then References.
A Character, an outfit, and a place
An upbeat vertical ad for a rain jacket.
Mara is @mara, adopting her face, hairstyle and voice. She wears @yellowshell. The setting is @harbourwalk, adopting its layout and light; ignore any people in it.
Shot 1 [0-4s]: Medium shot, handheld follow. Mara jogs along the harbour wall as the rain starts; drops bead and roll off the jacket's sleeve.
Shot 2 [4-7s]: Close-up, slow push in. She pulls up the hood, grins at the camera and says: "Let it pour." Lip sync.
Shot 3 [7-10s]: Wide shot, crane rising. She runs on as the sun breaks through behind the moored boats.
Sound: rain on fabric, footsteps on wet stone, a light acoustic guitar track.
Negatives: no subtitles, no watermark, no changes to the jacket's logo.Why it works: Each reference gets one line and one job, the location lends its layout but not its people, and the shots take it from there. Swap in your own Character and References, and set a tall aspect ratio and 10 seconds.
Sound and dialogue
Write each line as the speaker, a speaking verb, a colon, and the quoted words: Mara says: "…", she whispers: "…", Voiceover: "…". Give each speaker their own line, name the speaker every time, and write the words in the language they should be spoken in.
Lemon fills an open soundtrack on its own: without "No dialogue." it may add speech, and without a music line it may add a score. Name the music you want ("a slow piano score that swells at the climax"), or write "No background music, only ambient and action sounds."
- Budget about 2.5 spoken words per second: a 4-second shot holds about 10 words.
- Time a line by putting it inside its shot, or by the clock: "From the 8th second, a woman's voice chants: …"
- A Character with a saved voice brings it along, so just name her as the speaker. For a saved voice clip, write "Jun speaks in the voice of @gravelvoice: …". Turn off Use Character Voices to leave the saved voices out.
- Add "Lip sync" after a line for a tight match between mouth and words.
- Well-described physical events get their own sound. Add sound sentences for the rest: "a whoosh as each color block slides in, a soft click on every number change".
- Let music drive the motion: mention a saved music clip and write "Choreograph the dance to the rhythm, speed and swells of @beat."
Text on screen
Short titles and numbers render well in motion graphics and logo pieces. Quote the exact words, and give their place and treatment: The title "Q3 RESULTS" slides in at top center in bold white sans-serif.
- For a product, ask that its label "stays sharp and legible, without deformation".
- Anywhere else, rule text out: "No subtitles, no watermark, no on-screen text." Dialogue scenes especially can grow burned-in subtitles.
A motion-graphics chart
Generate single shot. A minimalist motion-graphics chart on a deep navy background: three teal bars grow upward from left to right while the numbers "12%", "18%" and "27%" count up above them in white sans-serif, and the title "Growth in 2026" fades in at top center. A whoosh as each bar rises, a soft click on each number change. No dialogue.
Negatives: no other text, no watermark.Why it works: Every word on screen is quoted with its place and style, each motion has its own sound, and the negatives keep any other text out.
Editing and extending a clip
Add the clip you want to change as a reference video, and it becomes Video 1. Make one change per run with an edit verb (add, remove, replace, change to) and a keep clause: "…; her movement and the rest of the frame stay unchanged." Point at a moment with times: "From 4s to 6s in Video 1, …". Set the length to the clip's own length, in whole seconds.
To extend, say where the clip sits and what comes next: "Using Video 1 as the opening, continue: …", or "as the closing", or "as the middle segment, extend forward by two seconds … and backward by three seconds …". Restate the person's look, clothes, and setting so they hold across the join, and set the length to the clip plus the new seconds.
Change one thing, keep the rest
Edit the video: in Video 1, change the man's white T-shirt to a black leather jacket that creases with his shoulders as he turns. His face, hair and movement, the background, the lighting and the camera path stay unchanged.Why it works: One change, with the physical detail that makes it believable, and a keep clause that names everything else.
Continue a clip
Using Video 1 as the opening, continue: the girl in the cream knit hat and red mittens stops on the snowy forest path, rubs her hands together, then looks straight into the camera and breaks into a warm smile. She raises one hand to catch a falling snowflake as the camera slowly pulls back to a medium-wide shot of the sunlit path. Golden backlight, soft lens flare, shallow depth of field, warm and gentle.Why it works: It says where the source sits, restates her look so it holds across the join, and adds only the new beats.
Common mistakes
Avoid this, do this instead
Avoid: Unwanted cuts in a clip meant to be one take.
Instead: "Generate single shot." as the first line.
Avoid: Two people who merge or look like twins.
Instead: A cast line that maps each name to its reference, and "two different people, must not look alike".
Avoid: Frames and Characters in the same clip.
Instead: One mode per clip: make the opening still with your Character first, then use it as the First Frame.
Avoid: A soundtrack left to chance.
Instead: "No dialogue." and "No background music." whenever you want silence or only natural sound.
Avoid: Too much in one shot, and rapid repeated actions.
A man shakes his head three times a second, jumps, spins, waves, and runs out of the door, cinematic.Instead: One clear action per shot, 2 to 5 seconds each.
Shot 1 [0-3s]: Medium shot, camera static. The man shakes his head slowly, then laughs. Shot 2 [3-6s]: Hard cut. Wide shot, handheld follow. He grabs his coat and runs out of the door.Avoid: A negatives line full of quality words and repeated positives.
Instead: Only what must not appear: no subtitles, no watermark, no extra fingers.
Avoid: Describing what a frame already shows.
Instead: The motion, plot, dialogue, and camera between the frames.
Avoid: Faces or outfits drifting in long clips.
Instead: The look restated in later shots, and "no costume changes, no face distortion" in the negatives.
Avoid: More words than a shot can say.
Instead: About 2.5 words per second: 10 words for a 4-second shot.
FAQ
Lemon prompting questions
How long should a Lemon prompt be?
How do I keep a character consistent in Lemon?
Can Lemon do text?
Does Lemon support negative prompts?
Can I use a first frame and my Characters in the same Lemon clip?
How do I stop Lemon from cutting between shots?
Why did Lemon add music or a voice I did not ask for?
Can Lemon edit or extend a video?
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 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.
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 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 Lemon
New accounts get 300 free Gems, enough to try Lemon. No install, no credit card.
Try Lemon