๐ŸŽ‰๐ŸŽ‰ Turn your prompts into realistic images with Realistic AI!Try for Free
Image AI

How to Write Flux Prompts: Natural Language, Layout and Text

Flux follows full sentences, spatial layout and quoted text. Learn the prompt structure, settings and examples that get the best results.

By the AI Prompt Generator TeamUpdated September 27, 20263 min read

Flux, from Black Forest Labs, understands language very differently from older image models. It uses a large T5 text encoder alongside CLIP, which means it can follow long, natural sentences, place objects where you describe them and render short text accurately. Prompts written in the old "keyword soup" style waste that ability. This guide shows how to write prompts that play to Flux's strengths.

Write sentences, not tag lists

Compare:

Tag style (works poorly)

woman, cafe, rain, window, coffee, cinematic, 8k, masterpiece, best quality

Natural language (works well)

A young woman sits by a rain-streaked cafรฉ window, holding a cup of coffee
with both hands. Warm light from inside contrasts with the blue evening
street outside. Shot on 35mm film with a shallow depth of field.

Flux reads the relationships in the second prompt โ€” who is holding what, where the light comes from โ€” and composes the scene accordingly. Quality tags such as masterpiece or 8k have little effect and can be dropped.

A reliable structure

  1. Subject and action โ€” the main thing and what it is doing.
  2. Composition โ€” foreground, background, left/right, camera angle.
  3. Setting and atmosphere โ€” place, time, weather, mood.
  4. Style or medium โ€” photograph, oil painting, 3D render, flat illustration.
  5. Lighting and color โ€” light source, palette.
  6. Technical details (for photos) โ€” lens, film stock, depth of field.

Aim for roughly 30โ€“80 words. Very short prompts give generic images; very long prompts can drop details from the end.

Describe spatial layout explicitly

Flux is good at following positional instructions. Use them:

In the foreground, a ceramic bowl of ramen on a wooden counter. In the
background, a blurred neon-lit Tokyo street at night. The camera is at
table height, looking slightly up.

Words like in the foreground, on the left, behind, above, in the corner are far more effective than they are in Stable Diffusion 1.5.

Rendering text

Flux can put readable text into images. Put the exact text in quotation marks and describe where and how it appears:

A vintage enamel shop sign that reads "OPEN LATE" in bold red serif
letters, mounted on a weathered brick wall, soft afternoon light.

Tips:

  • Keep text short โ€” a few words renders more reliably than a sentence.
  • Describe the font style: bold sans-serif, hand-lettered script, neon tube letters.
  • If a letter is wrong, regenerate with a new seed before rewriting the prompt.

There is no negative prompt โ€” describe what you want instead

Standard Flux pipelines do not use a negative prompt. Instead of "no people", describe the positive version: "an empty beach at dawn". Instead of "not blurry", specify "sharp focus, crisp details".

Choosing a Flux model

  • Flux Schnell โ€” fastest, good for drafts and exploration.
  • Flux Dev โ€” higher quality, open weights for non-commercial use, popular for LoRAs.
  • Flux Pro / newer releases โ€” best prompt adherence and detail, available through APIs.
  • Flux Kontext โ€” edits an existing image from a text instruction ("make it winter", "change the jacket to red").

The same prompt style works across all of them.

Settings to know

  • Guidance โ€” Flux Dev works best at a lower guidance than Stable Diffusion; around 3โ€“4 is a common default. Higher values increase contrast and saturation but can look artificial.
  • Steps โ€” about 20โ€“30 for Dev; Schnell is designed for 1โ€“4 steps.
  • Aspect ratio โ€” choose it before writing; a vertical 9:16 composition should be described differently from a wide 16:9 scene.

Examples

Photography

A close-up portrait of an elderly beekeeper in a white veil, smiling as
bees rest on her gloved hand. Late afternoon sun filters through apple
trees behind her. Shot on 85mm, shallow depth of field, natural colors.

Illustration

A flat vector illustration of a fox reading a book under a tree, autumn
leaves falling, limited palette of orange, teal and cream, clean lines,
children's book style.

Poster with text

A minimalist travel poster for Kyoto showing a red torii gate on a misty
hillside, the word "KYOTO" in large white sans-serif letters across the
top, muted pastel colors, mid-century print texture.

From idea to Flux prompt

The Flux Prompt Generator expands a short idea into a descriptive, natural-language prompt tuned for Flux. Copy the result into your Flux tool of choice, or send it straight to an image generator.

Try the Flux Prompt Generator

Descriptive natural-language prompts for Flux. Free, no sign-up.

Open generator

Keep Learning