Skip to content
Log inSign up free

Prompting guide

How to prompt Flux 2

Flux 2 prompt guide: how to prompt Flux 2 Dev with subject-first word order, JSON prompts, exact hex colors, and multi-image edits, with examples.

Try Flux 2Flux 2 model page

Editorial product still life
Made with Flux 2Editorial product still life

Quick start

Your first great Flux 2 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

Flux 2 weighs the first words of a prompt most, so lead with the subject, then the action, the style, and the setting and light. Thirty to eighty words suits most images. It reads plain prose and JSON equally well, and it takes exact hex colors tied to objects. It takes no Characters or @mentions: upload images and call them image 1, image 2 in plain words.

The rules that matter most

  1. Put the subject first

    Order by importance: the subject, its action, the style that matters, the setting, then secondary details. The words at the end weigh least.

  2. Name the camera, not "professional"

    "Shot on Hasselblad X2D, 80mm lens, f/2.8, natural lighting" does far more than "professional photo, high quality".

  3. Describe what you want, not what you do not

    There is no negative prompt. Write "sharp focus throughout" instead of "no blur", and "an empty street" instead of "no people".

  4. Tie every hex color to an object

    "Walls in hex #C4725A" works. "Use #C4725A somewhere" is applied unevenly.

  5. Change one thing at a time

    Keep the Seed fixed and change one part of the prompt per run, so you can see exactly what each change did.

Fill in this template

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

Template
[Subject] [doing what], [style], [setting], [light and mood], [camera and lens].
The template, filled in
A weathered fisherman in a yellow oilskin coat hauling a net over the side of a small wooden boat, documentary photograph, on a gray, choppy sea at dawn, low sun breaking through clouds from the left, cold and determined mood, shot on Sony A7IV, 35mm lens, f/4.
  • [Subject] The main thing, first, with its key look: a weathered fisherman in a yellow oilskin coat.
  • [doing what] The key action: hauling a net over the side of a small wooden boat.
  • [style] The style that matters: documentary photograph, oil painting, flat vector illustration.
  • [setting] Where and when: on a gray, choppy sea at dawn.
  • [light and mood] The light with its source and direction, then the mood: low sun breaking through clouds from the left, cold and determined.
  • [camera and lens] The gear that sets the look: shot on Sony A7IV, 35mm lens, f/4.

Prompts to start from

  • A short prompt with a strong look

    Dog wrapped in a white towel after a bath, photographed with direct flash and high exposure, wet fur sharply detailed, editorial raw portrait, harsh flash lighting, intimate humorous documentary style

    Why it works: The subject first, then the light that makes the look, then the genre: under thirty words, every one of them visual.

  • Brand colors on every object

    A modern living room with warm terracotta walls in hex #C4725A, a large L-shaped sectional sofa in deep teal hex #1B6B6F, and amber hex #E8A847 accent pillows, throw blanket, and a velvet ottoman. Light oak floors, a round marble coffee table with dried pampas grass in a glass vase, floor-to-ceiling windows with sheer white curtains letting in soft afternoon light, two potted fiddle leaf figs, and an abstract painting on the main wall in all three accent colors. Clean minimal composition, natural daylight casting soft shadows.

    Why it works: Each hex code is attached to the thing it colors, so the palette lands where it should. A 3:2 aspect ratio suits a room.

  • A retro poster with its quote

    Groovy retro poster with the quote "If you love me let me sleep". Bold 70s typography in deep red and warm pink tones, cream background, a bold orange doodle around the text, funky layout with a playful drop shadow. Style: bold vintage aesthetic, dopamine decor.

    Why it works: The quote comes early, in quotation marks, and the type, the colors, and the layout are each described. Try 2:3.

  • A macro with a label and a gradient

    Realistic macro photograph of a hermit crab using a soda can as its shell, partly emerging from the can, sharp detail and natural colors, on a sunlit beach with soft shadows and a shallow depth of field, blurred ocean waves behind. The can reads "TIDE POP" and has a color gradient from #FF5733 at the top to #33FF57 at the bottom.

    Why it works: A photo, a short quoted word, and a hex gradient on one named object, all in one prompt: the kind of detail Flux 2 holds together.

  • An outfit from three uploads

    The woman from image 1 wearing the cropped denim jacket from image 2 over the white slip dress from image 3, standing on a sunlit Lisbon street. Full body, eye level, 50mm lens, soft afternoon light, natural skin texture, editorial street-style photograph.

    Why it works: Upload the person first and the garments after, and name them image 1, 2, and 3 in plain words: Flux 2 takes no @mentions. Each image has one job.

Advanced

Advanced Flux 2 prompting

Techniques for more control once the basics work.

The settings that matter

Mage runs Flux 2 Dev at Normal steps and Normal guidance, the balance of speed and quality its makers recommend. Leave guidance there. Slow steps, their reference setting, can add a little fine detail at the cost of time.

Seed is the setting to use: keep it fixed while you change one part of a prompt, and that change is the only difference between two images.

  • Leave Prompt Enhance off: it adds words you did not write.
  • With one uploaded image, the result takes that image's shape and the Aspect Ratio setting is ignored. With two or more, the Aspect Ratio and Resolution settings apply.

Word order and length

Flux 2 reads Subject + Action + Style + Context, and it gives the early words the most weight. Order by priority: the main subject, the key action, the style that matters, the essential context (setting, light, time, mood), then the secondary details.

Ten to thirty words is right for exploring concepts and styles, thirty to eighty for most images, and more than eighty for complex scenes with many conditions. English is the most precise, though a prompt in another language can render a local culture more authentically.

  • Use prose for quick, simple scenes, and JSON for production work, scenes with several subjects, and changing one element at a time.
  • Specific gear beats quality words: a camera body, a lens, an aperture, and the light.

JSON prompts for a series

Flux 2 understands a JSON description of the scene as well as it understands prose, and JSON makes it easy to change one thing while everything else stays identical. Paste the JSON in as the prompt.

Build it field by field. Start with one subject, generate, and look. Then change one field per run with the same Seed: add a second subject, change the lighting, swap the background. When the image is right, keep the JSON as the source of truth for the series.

WriteWhat it does
sceneThe setting and the kind of image.
subjectsA list with one entry per subject, each with a description, a position, an action, and optionally a pose and a color_palette.
styleThe medium and the genre.
color_paletteThe colors of the whole image, as a list.
lightingThe source, direction, and softness.
moodThe feeling, in a few words.
backgroundWhat is behind the subjects.
compositionThe layout: rule of thirds, centered, symmetrical.
cameraThe angle, distance, lens-mm, f-number, and focus, as their own fields.

Step one of a product series

{
  "scene": "Professional studio product photography on a polished concrete surface",
  "subjects": [
    {
      "description": "Minimalist ceramic coffee mug with steam rising from hot coffee",
      "position": "Center foreground",
      "color_palette": ["matte black ceramic"]
    }
  ],
  "style": "Ultra-realistic commercial product photography",
  "color_palette": ["matte black", "concrete gray", "soft white highlights"],
  "lighting": "Three-point softbox setup, soft diffused highlights, no harsh shadows",
  "mood": "Clean, professional, minimalist",
  "background": "Polished concrete surface with a seamless studio backdrop",
  "composition": "rule of thirds",
  "camera": {
    "angle": "high angle",
    "distance": "medium shot",
    "lens-mm": 85,
    "f-number": "f/5.6",
    "focus": "Sharp on the steam and the mug"
  }
}

Why it works: One subject and every field filled in. For the next image, change one field (the background, say) and keep the rest and the Seed exactly as they are.

Exact colors with hex

Flux 2 takes hex codes, which makes it the model for brand colors. Attach each code to a specific object: "The car is #FF0000", "walls in hex #C4725A". A code with no object is applied unevenly.

For a product, break it into parts, give each part its own description with its exact hex and "color_match": "exact", and list the brand palette in color_palette. Hex works for type too: "The logo text 'ACME' in color #FF5733."

A product in brand colors

{
  "scene": "Studio product photography on a seamless pale gray backdrop",
  "subjects": [
    {
      "description": "Knit running shoe upper, side view, toe pointing left, hex #1B6B6F",
      "color_match": "exact"
    },
    {
      "description": "Thick foam midsole, hex #F2EDE4",
      "color_match": "exact"
    },
    {
      "description": "Laces and heel pull tab, hex #E8A847",
      "color_match": "exact"
    }
  ],
  "color_palette": ["#1B6B6F", "#F2EDE4", "#E8A847"],
  "lighting": "Large softbox from above left, soft shadow under the shoe",
  "camera": { "angle": "eye level", "lens-mm": 100, "f-number": "f/8" }
}

Why it works: Each part of the shoe carries its own hex and an exact match, so the brand colors land on the right parts.

Editing and combining images

Upload images and refer to them as image 1, image 2, image 3, in the order you added them; Flux 2 does not read @image1 or saved Characters and References. Give each image one job and say how they combine: "the woman from image 1, wearing the jacket from image 2, in the painterly style of image 3, in a rainy city street at dusk".

It handles combining garments into one outfit, placing furniture in a room, product composites, and keeping a character. For an edit of one photo, say the change and what stays.

  • Upload only the images that matter, each with one role.
  • Keep a character across separate images by writing one detailed description and repeating it word for word in every prompt, with the same labeled lines for every panel: Style, Character, Setting, Text, Mood. For later panels, upload the approved first one as image 1.

One panel of a comic

Style: clean ligne claire comic, flat colors, thin even black outlines.
Character: Odile, a tall woman in her forties with a gray pixie cut, round tortoiseshell glasses, a mustard turtleneck, and a long navy coat.
Setting: a rainy tram stop at night, the headlights of a tram approaching.
Text: a speech bubble from Odile reading "Late again."
Mood: dry, tired humor.

Why it works: The Character line is pasted word for word into every panel, and only Setting, Text, and Mood change. For later panels, upload the first approved one and add "the woman from image 1".

Text, posters, and infographics

Typography is a strength. Put the exact words in quotation marks, describe the text early in the prompt, and keep each string short. Say where it goes ("below the title, centered"), its style ("elegant serif", "bold industrial lettering", "handwritten script"), its size ("large headline", "small body copy"), and its color, by hex when it matters.

For an infographic, give the title, the subtitle, each section's heading, content, and visual, and a hex color scheme.

An infographic, fully specified

An infographic poster titled "How Coffee Gets to Your Cup" with the subtitle "From seed to sip in five steps", on a cream background #F4EDE1. Five sections from top to bottom, each with a heading and a flat icon: "Grow" with a coffee shrub; "Pick" with a hand holding red cherries; "Process" with beans drying in the sun; "Roast" with a drum roaster; "Brew" with a pour-over. Headings in bold dark brown #4A2C1D sans-serif, icons in terracotta #C4725A and sage #8FA98B. Clean flat vector style with generous spacing.

Why it works: Every word is quoted, every section has its heading and its visual, and every color is a hex on a named element. A tall aspect ratio such as 2:3 suits it.

Camera, light, and style words

Specific gear and named light beat generic quality words. Pick a photo era, a lens, an aperture, and a light, and Flux 2 builds the look from them.

  • Composition words read well: rule of thirds, leading lines, foreground and background layers, worm's-eye, bird's-eye, dutch angle, symmetry, negative space.
  • Styles: teal-and-orange grade, film noir, impasto, Art Nouveau, Bauhaus, concept art, matte painting, graphic novel, polaroid, VHS.
WriteWhat it does
shot on Sony A7IV, clean sharp, high dynamic rangeA modern digital photo.
early digital camera, slight noise, flash photography, candid, 2000s digicam styleA 2000s snapshot.
film grain, warm color cast, soft focus, 80s vintage photoA 1980s print.
shot on Kodak Portra 400, natural grain, organic colorsWarm, gentle film color.
f/1.4 to f/2.8Shallow depth of field: a sharp subject, a soft background.
f/8 to f/16Deep focus: everything sharp.
24mmWide, with room around the subject.
85mmThe portrait lens: flattering, slightly compressed.
anamorphicCinematic frames with oval bokeh and streaked flares.
Rembrandt lightingLight from one side, with a small triangle of light on the shadowed cheek.

Common mistakes

Avoid this, do this instead

  • Avoid: "No blur", "no people", "no text".

    Instead: "Sharp focus throughout", "an empty street", a plain background.

  • Avoid: The subject buried after the style and the mood.

    Cinematic, moody, teal and orange, dramatic lighting, a lighthouse on a cliff.

    Instead: The subject first.

    A white lighthouse on a black basalt cliff, waves breaking below, cinematic teal-and-orange grade, stormy dusk, low sun from the right.

  • Avoid: "Professional photo, high quality".

    Instead: The camera, the lens, the aperture, and the light.

  • Avoid: A hex code with no object.

    Instead: The hex on the object it colors.

  • Avoid: @image1, or a saved Character's @handle.

    Instead: The image uploaded, and "image 1" in plain words.

  • Avoid: Rewriting the whole JSON between runs.

    Instead: One field changed, with the same Seed.

  • Avoid: Prompt Enhance turned on.

    Instead: Prompt Enhance off: it adds words you did not write.

FAQ

Flux 2 prompting questions

How long should a Flux 2 prompt be?
Thirty to eighty words for most images, ten to thirty to explore a concept or a style, and more than eighty for complex scenes. Put the subject first, because the early words weigh most.
Does Flux 2 support negative prompts?
No. Describe the state you want: "sharp focus throughout" instead of "no blur", "an empty street" instead of "no people".
Can I write Flux 2 prompts in JSON?
Yes. Flux 2 reads JSON as well as prose. Paste it in as the prompt; it is the easiest way to change one element at a time across a series.
Can Flux 2 put text in an image?
Yes. Quote the words, describe the text early, keep it short, and give its place, style, size, and color. Posters, ads, magazine covers, and infographics are among its strengths.
How do I keep a character consistent in Flux 2?
Write one detailed description and repeat it word for word in every prompt, and upload the approved first image as image 1 for the next ones. Flux 2 does not take saved Characters.
Can Flux 2 match exact brand colors?
Yes, with hex codes tied to objects: "the sofa in deep teal hex #1B6B6F". For a product, give each part its own hex.

Make your first image with Flux 2

New accounts get 300 free Gems, enough to try Flux 2. No install, no credit card.

Try Flux 2