Skip to main content

​
What is prompting?

Use a prompt to describe what appears in the image, how it looks, and its mood or style. On FLUX 3 Image, the prompt carries everything you want to control. There is no negative_prompt, guidance, or seed field, and bounding boxes, when you use them, go at the end of the prompt string.
Two red pandas on a bamboo beam, one awake and looking at the camera, the other asleep with its head on a bamboo stub

FLUX 3 Image

​
Short prompts and detailed captions

A short prompt works. FLUX 3 Image expands it into a detailed description before it generates the image. Write more only for the details you want to control, such as positions, light, or text. Here is a pair from the FLUX 3 launch post: a short scene prompt, and the detailed caption it was expanded into. This image was generated from the detailed caption plus a bounding box for each element. Hover over the caption to see where each box goes: What the caption adds to the short prompt:
  • Medium first: “High-angle black and white photograph with a grainy film texture” specifies the look before describing the subjects.
  • Positions: “In the right foreground”, “To the left”, “The background” say where each part of the scene sits.
  • Pose and appearance: the men “walk in mid-stride towards the right edge of the building, rendered almost entirely as silhouettes”.
  • Small details: “cars and buses appearing as small dark shapes”, “a faint haze or smoke”, skyscrapers “featuring repetitive grids of windows”.
  • Light and its effect: “long, sharp diagonal shadows … under harsh, directional lighting” and “the left side plunged into deep darkness while the right catches intense sunlight”.
  • Only what’s visible: no “beautiful”, “iconic”, or “masterpiece”. Every phrase names something you could point to in the frame.
Most images don’t need a caption this long. Start with the subject, then add the details you want to control.

​
What can a prompt look like?

A prompt can take many forms. There is no single correct format. What matters is that your description gives FLUX enough to work with.
This template is a useful starting structure, not a strict formula. You can keep it concise, expand individual slots, or skip the ones that don’t apply to your image. The FLUX 3 launch captions follow roughly this order: medium and style first, then the subjects and where they sit, then light, color, and background.

​
Use natural language

Write a clear description of the image. Use sentences to explain relationships: who holds what, what stands behind what, and where the light comes from. Short phrases can specify style, color, and lighting.

​
Text in images

To render text, put the exact words in quotation marks, then say where they go and how they look. Quotation marks separate the words to print from the description around them. These FLUX 3 posters quote each piece of text exactly, say where it goes, and give it a type style. Choose a poster and hover over the caption to see each text box: In the on-time poster, the line break is part of the quote: \n splits “you’ll do” and “everything” into two lines. Text in another script works the same way. The Tokyo poster’s prompt is in English and quotes the Japanese directly: The text reads "東京". Both posters give each piece of text its own box (FLUX 3 Image Editing). Without boxes, write the position in words, as in Below the scene, bold serif typography reads "…". For typography styles and more text examples, see Style, aesthetics & text.

​
Refine as you go

Strong prompts usually come from iteration, not from trying to write the perfect prompt on the first attempt. A practical loop is:
  1. Start with a simple version
  2. Check what FLUX got right and wrong
  3. Adjust one important detail at a time
If the image is close but not there yet, tweak the subject, framing, lighting, or style before rewriting everything. On FLUX 3 Image, the 768sq resolution returns fastest while you work on the wording. There is no seed field, so two runs of the same prompt can differ: look at more than one result before you decide a change didn’t help. Editing works well for this loop. Start from a short prompt, then send each result back with one change. FLUX 3 Image keeps the rest of the image the same, so you can see what each instruction did. Each image below is an edit of the one before it, and the caption is the instruction that made it:

​
Image input

Your prompt isn’t limited to text. FLUX 3 Image takes up to 10 reference images in the images field, each at least 256 × 256 pixels and at most 16 megapixels. When you send images, the prompt says what to do with them: edit, restyle, or combine them into a new composition. There is no separate mode field. In the prompt, refer to references by position: “image 1” is the first entry in images, “image 2” the second. Clarify the purpose of each image. Here, image 1 is the subject and image 2 is the style: For request details, see Multi reference. Editing prompts are covered in depth in Image editing.