Skip to content
Log inSign up free

Prompting guide

How to prompt Mango

How to prompt Mango, Mage's image model: a one-minute quick start, a fill-in template, and expert tips for characters, references, precise edits, and text.

Try Mango 3Mango model page

K-drama key art: two members of the Mage cast under one umbrella on a rainy bridge at night
Made with MangoK-drama key art: two members of the Mage cast under one umbrella on a rainy bridge at night

Quick start

Your first great Mango 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

Mango reads plain sentences and thinks them through before it draws, so precise facts beat piles of adjectives. Say what the image is, who or what is in it and where, then the style and the light. To edit, upload the image and say what stays and what changes. Size and shape are settings, not words in the prompt.

The rules that matter most

  1. Write sentences, not tags

    Subject, action, and setting first, then style, light, and framing. "A girl walks under a parasol along a tree-lined path" gives Mango far more than "girl, parasol, path".

  2. Say what kind of image it is

    Open with the image type when there is one: "A poster for…", "Design a logo…", "An infographic about…". It sets the layout before the details.

  3. Quote the exact words

    Put any text in double quotes, with where it goes and how it looks: the title "NIGHT MARKET" in bold red brush lettering across the top.

  4. For an edit, say what stays

    Start with what must not change, then the one change, then its consequence: the shadow, the reflection, how it blends in.

  5. Use the settings for size and shape

    Aspect ratio and resolution are settings. "16:9" or "4K" in the prompt does nothing.

Fill in this template

Replace each part in square brackets with your own idea, and leave out a part you do not need.

Template
[Image type, if any]: [subject] [doing what] in [setting]. [Key details, each in its place]. [Style or medium]. [Light, with its source and direction]. [Framing or lens].
The template, filled in
A poster for a jazz night: a saxophonist in a brown suit plays with his eyes closed on a small brick-walled stage. Across the top, the title "NOCHE DE JAZZ" in tall cream art-deco lettering; his long shadow falls on the wall behind him. Screen-print style in deep navy and amber. A single warm spotlight from above. Full body at eye level.
  • [Image type, if any] What the image is for: a poster, a product photo, a book cover. Leave it out for a plain scene.
  • [subject] Who or what it is about, with a few visible facts: a saxophonist in a brown suit.
  • [doing what] The action or the pose: plays with his eyes closed.
  • [setting] Where it happens: a small brick-walled stage.
  • [Key details, each in its place] The other elements and where they go: on the left, in the foreground, across the top.
  • [Style or medium] A precise name: screen print, Monet oil painting, editorial photograph, 3D cartoon.
  • [Light, with its source and direction] One light and what it does: a single warm spotlight from above casting a long shadow.
  • [Framing or lens] How it is framed: full body at eye level, a 50mm lens, a close-up.

Prompts to start from

  • A busy scene, placed piece by piece

    A cluttered developer's desk at night. An open laptop shows green code on a black screen. Beside it, a mug printed with the word "DEPLOY" lets off a thin curl of steam. An open notebook shows a hand-drawn Venn diagram of three overlapping circles in gray, blue and light green. A yellow sticky note carries a three-level mind map. A fountain pen lies with its cap beside it, next to a phone lighting up with a new message. A small pot of succulents sits in the far corner. Behind, a bookshelf falls soft and out of focus. A desk lamp on the right casts warm light and long shadows across the desk.

    Why it works: Instead of adjectives, it lists every element by where it is, which keeps a crowded scene in order, and the one lamp sets the light.

  • Your Character in a saved outfit and place

    Ana (@ana) wears @red-coat and stands at the rail of @harbor-pier at dusk, one hand on the rail, looking back over her shoulder toward the camera with a small smile. Wind lifts the hem of the coat. Keep her face, hair and the coat's cut and color exactly as in their images. Photorealistic, 50mm lens at eye level, the low sun behind her rimming her hair, harbor lights starting to glow in the background.

    Why it works: Each mention has one job, the name is bound once, and the light has a source and a direction. Swap in your own Character and References.

  • An edit that leaves the rest alone

    Keep the composition, the marble surface, the lighting and the bottle's position unchanged. Replace the green glass bottle with the frosted white bottle from @image2, matching its label exactly. Its reflection and shadow on the marble follow the same light from the upper left.

    Why it works: Upload the product shot first and the new bottle second. The prompt says what stays, names the one change, and asks for the shadow and reflection that make it look real.

  • A poster with text in Spanish

    A vertical poster for a neighborhood jazz night. A saxophonist in a brown suit plays under a single warm spotlight on a small brick-walled stage, his shadow long on the wall behind him. Across the top, the title "NOCHE DE JAZZ" in tall cream art-deco lettering; along the bottom, "Viernes · 21:00 · Café Lumbre" in small clean sans-serif. Deep navy and amber palette, screen-print texture.

    Why it works: Quoted words with a place and a style come out as written, and Mango 3 writes many languages natively. Pick a tall aspect ratio in the settings too.

Advanced

Advanced Mango prompting

Techniques for more control once the basics work.

Which Mango to use

One prompt carries across the family. Draft on Mango 3 Turbo, then render the keeper on Mango 3 with the same words. Mango 3S and Mango 2 render larger images, and the original Mango stays for matching work made on it.

Mango 3 reasons over your prompt and expands it before it draws. That makes it strong at exact instructions, counts, and positions, and it means precise facts help it more than stacked adjectives.

  • Pinpoint edits with coordinates and native text beyond English and Chinese work on Mango 3 and Mango 3 Turbo only.
  • The original Mango takes no @handles, not even @image1. Upload the images and write "image 1", "image 2" in plain words.

Structure: clear sentences in order

Write coherent sentences: subject, action, and setting, then style, color, light, and composition when the look matters. Name the image type first when there is one.

Stay under about 600 words: past that, details get dropped. For a busy scene, inventory every element by where it is instead of adding adjectives.

  • For a single portrait or product, a comma-separated brief works: "editorial portrait, shot in medium format, shallow depth of field, dramatic studio lighting". Switch to sentences once there are several subjects or a layout.
  • There is no negative prompt. Put an exclusion in a sentence: "Do not include any text."
  • For an image that explains something, use the exact technical terms and the format: "Draw this system of equations and its solution steps on a blackboard: 5x + 2y = 26; 2x − y = 5."

References and Characters

Mention saved Characters and References (outfits, locations, poses, objects) by their @handle, and images you upload as @image1, @image2. Mango sees them all as one numbered list: your uploads first, then Characters in the order you first mention them, then References. Before Mango reads the prompt, Mage replaces each mention with its number, so @ana becomes "character from image 2" and @red-coat "outfit from image 3".

So give every image one role, in words: "Dress the character in image 1 in the outfit from image 2.", "Apply the style of image 2 to image 1.", "Use image 1 as the background."

  • Bind a name once ("Ana (@ana) sits…") and write Ana after that. Repeating @ana pastes the long replacement into every clause.
  • To keep a person exactly on model, mention the Character and upload a close crop of the face: "Ana (@ana), her face exactly as in @image1, …".
  • Never write "the reference" or "the attached photo": name the image by its number and say what to take from it.
  • Call the target of an edit by what it visibly is, never by a pronoun: "the tallest panda", "the person on the left".
  • To merge several references into one scene, write two parts: first what to take from which image ("the face and hair of Ana (@ana), the shape and label of the bottle in @image1"), then the new scene, with where each subject is and what they do. End with "consistent art style and harmonious composition".
  • An @ followed by letters always reads as a handle, so an email address or a handle you have not saved stops the generation before anything is charged.

A clean Character portrait to reuse

Asset: one fictional character identity reference for a live-action short film. One original woman, early thirties, athletic build, warm brown skin, dark brown eyes, short natural curls, small gold hoop earrings. She wears a faded olive bomber jacket over a white T-shirt, dark jeans and white sneakers. One person only, a single portrait from the top of her head through the knees, standing in a slight three-quarter turn with her face nearly frontal and both empty hands relaxed at her sides. Plain warm-gray studio backdrop, soft even light showing both eyes and the fabric textures, natural unretouched skin. Neutral closed-mouth expression. No writing, logos or watermark. No props. No collage or multiple views. This image fixes appearance only.

Why it works: One person, a neutral pose and light, and closing constraints make a clean identity image to save as a Character or to use in video. A 2:3 aspect ratio suits it.

Edits that leave the rest alone

With an image uploaded, the prompt is an edit. Say what stays (the pose, the face, the composition, the light), then the change with a verb Mango is tuned on (add, remove, replace, change, turn into), then its physical consequence: shadows, reflections, how it blends in.

You can also draw on the image before you upload it, with a circle, an arrow, a scribble, or a colored box, and then point at the mark in words.

  • Text edits are edits too: "Change the text on the sign to "OPEN LATE" in the same font."
  • To render a sketch, a wireframe, or a floor plan, name the subject ("a photorealistic modern living room"), say what must match ("the layout and furniture placement exactly match image 1"), and add "Do not include any text or hand-drawn edges from the sketch." If the sketch carries labels, add "Generate based on the text in the image."

An edit guided by marks you drew

Insert a TV where the red area is marked and a sofa where the blue area is marked. Keep the original wooden style. Remove all sketch lines.

Why it works: The colored marks say where, the words say what, and the last sentence cleans up the marks. It works on every Mango.

Pinpoint edits with coordinates

Mango 3 and Mango 3 Turbo read positions on a 0 to 999 grid laid over each image: 0 0 is the top-left corner and 999 999 the bottom-right. To convert from pixels, divide the x position by the image width and multiply by 1,000, and do the same for y.

Put the tag right after the image it points into. A point lets Mango decide how much around it to change; a box sets the exact region and size, from its top-left corner to its bottom-right corner.

  • Name the target when a box holds several things ("the person on the left in @image1 <bbox>…</bbox>"), and box what must not change ("keep @image1 <bbox>700 120 920 360</bbox> unchanged").
  • Use coordinates only when you know them, for example from an image editor's cursor readout. A guessed box is worse than words.
WriteWhat it does
<point>x y</point>A spot. Mango decides how much around it to change.
<bbox>x1 y1 x2 y2</bbox>An exact region, from its top-left corner to its bottom-right corner.

Replace two things and protect a third

Replace the floor lamp at @image1 <point>812 430</point> with a tall potted fiddle-leaf fig in a terracotta pot. Replace the painting in @image1 <bbox>310 120 590 360</bbox> with a large black-and-white photograph of ocean waves. Keep @image1 <bbox>0 600 999 999</bbox> unchanged. Everything blends with the room's warm afternoon light.

Why it works: A point for the lamp, a box for the painting, and a box around the floor that must not move, all in the same room photo. Draft it on Mango 3 Turbo, then render on Mango 3.

Text in the image

Put the exact words in double quotes, with their position and style: the title "NIGHT MARKET" in bold red brush lettering across the top. Writing titled Night Market, without quotes, leaves the words to chance.

Say what the image is for, because the layout follows the purpose: "Design a logo for a gaming company. The logo shows a dog playing with a game controller. The company name "PITBULL" is written below it."

  • Mango 3 and Mango 3 Turbo write natively in Russian, Arabic, Filipino, Thai, Turkish, Korean, Malay, Spanish, Portuguese, Indonesian, French, German, Vietnamese, and Japanese, besides English and Chinese. Write the quoted words in their own script. The other Mango models write English and Chinese.
  • Change or resize existing text with an edit: "Enlarge the title to fill the red box and match its style to the saxophone icon."

Style and light

Precise style names beat adjectives: picture book style, children's book illustration, Monet oil painting, linear minimalist icon style, 3D cartoon style, anime key visual, felt wool figure on a small stand, transparent crystal.

For photographs, name the kind of shot: photorealistic, cinematic photographic feel, high-definition commercial product photography, or an editorial portrait shot in medium format with a shallow depth of field.

Describe light plainly and give it a direction: "Sunlight shines from the right side, casting light and shadow on the desk." One light source, one lens, and one style set a mood better than a dozen mood words.

Sets and series

One generation makes one image. For a set that belongs together, such as a storyboard, a campaign, or a character in four places, run one generation per image with the same handles and references, changing only the line that sets the scene.

Asking for "a set of four" in one prompt gives one image, at best a sheet of panels. When a sheet is what you want, ask for it: "one image laid out as a 2×2 grid of four panels, thin white gutters".

Common mistakes

Avoid this, do this instead

  • Avoid: Tag lists.

    girl, umbrella, tree-lined street, oil painting

    Instead: A sentence with the subject, the action, the setting, and the style.

    A girl in a long dress walks under a parasol along a tree-lined path, painted in the style of a Monet oil painting.

  • Avoid: An edit that points at nothing.

    put that one in pink

    Instead: The target named by what it looks like, and what stays.

    Dress the tallest panda in a pink opera costume and headdress, keeping its pose unchanged.

  • Avoid: "Like the reference" or "same as the photo".

    A woman like the one in the reference photo, reading in a garden.

    Instead: The image's number, and what to take from it.

    The woman in @image1, with her face and hair exactly as in that image, reading in a garden.

  • Avoid: Size words in the prompt: "16:9, 4K, ultra HD".

    Instead: The aspect ratio and resolution in the settings.

  • Avoid: "Generate four images of…".

    Instead: Four generations with one change each, or one image laid out as a grid.

  • Avoid: @ana repeated in every clause.

    Instead: Ana (@ana) once, then Ana.

  • Avoid: Coordinates on Mango 3S, Mango 2, or the original Mango.

    Instead: Words, or a mark drawn on the image.

  • Avoid: A dozen adjectives about the mood.

    Instead: The one light source, lens, and style that make that mood.

FAQ

Mango prompting questions

Does Mango support negative prompts?
No. Put an exclusion in a sentence instead, such as "Do not include any text or logos.", and describe what you do want.
How long can a Mango prompt be?
Up to about 600 words before details start to get dropped. Most images need only a few sentences: spend the words on placing elements, not on adjectives.
How do I keep the same character across many images?
Save the person as a Character and mention the same @handle in every prompt, bound to a name once. For an exact match, also upload a close crop of the face and write "her face exactly as in @image1".
Which languages can Mango write in an image?
Mango 3 and Mango 3 Turbo write English and Chinese plus 14 more languages natively, among them Spanish, Japanese, Korean, Arabic, and Russian. The other Mango models write English and Chinese.
Why does Mango ignore "4K" or "16:9" in my prompt?
Mage sends Mango an exact size from the aspect ratio and resolution settings, so those words do nothing. Set them in the settings.
Can Mango make several images from one prompt?
One generation makes one image. Run the prompt again with one change for each image of a set, or ask for one image laid out as a grid of panels.
Do I write prompts for Mango 3 Turbo differently?
No. The same prompt works on Mango 3 and Mango 3 Turbo, so draft on Turbo and render the keeper on Mango 3 with the same words.

Make your first image with Mango 3

New accounts get 300 free Gems, enough to try Mango 3. No install, no credit card.

Try Mango 3