FLUX 3 Image.
Describe it like a photographer.

On FLUX 3 Image the prompt carries everything. There is no negative prompt, no guidance slider and no seed, so the words you write are the whole control surface. That makes it simple to learn: describe the finished image the way you would brief a photographer, and add detail only where you want control.

A short prompt works, because the model expands it into a detailed description before it generates. Write more when you care about positions, light, materials or text. Black Forest Labs suggests a reliable order: the image type and medium first, then the subject, the setting, the light, and the framing and lens. Every prompt in this guide follows that order.

The same page does generation and editing. Leave the reference slot empty for text to image. Attach one to ten images and the prompt becomes an instruction: recolor this, replace that, put the product from image 2 into the room in image 1.

1. Image type and medium. One opening sentence that sets the look: "Editorial portrait photograph shot on medium format film", "Full color scan of a screen printed concert poster", "Epic concept art matte painting, ultra wide cinematic shot". This sentence decides what kind of picture you get before anything else is read.

2. Subject. Who or what, how they look, and the one thing they are doing: "a ceramicist in her seventies with cropped silver hair and a rust colored linen work jacket ... clay dusted hands folded on one knee". Visible details that tell your subject apart do more than adjectives.

3. Setting. Where it happens and what surrounds the subject: "shelves of unglazed bowls fade into shadow behind her", "a weathered pale blue wooden table". If two subjects relate to each other, say how.

4. Light. The source, the direction, the quality and what it does: "soft north window light from the left, deep falloff on the right", "late afternoon sun casts long window shadows across a polished concrete floor". This is the single most useful line for photoreal work.

5. Framing and detail. Shot size, angle, lens, depth of field and texture: "shot straight down with a 50mm lens", "tilt shift lens with straight verticals", "grainy tungsten film look with halation". The order is a habit, not a rule: if the frame is the point of the picture, as in an ultra wide shot, open with it.

Attach one to ten images and the prompt becomes an edit. There is no edit mode, mask or strength setting: the instruction says whether you are changing a detail, restyling, or combining images. Each input should be at least 256 x 256 pixels and no larger than 16 megapixels.

Refer to images by position. The first image you attach is "image 1", the next is "image 2", and so on. With the aspect ratio on auto, the output keeps the frame of image 1, so attach the image whose shape you want first. Set a ratio only when the edit should change the frame, for example turning a square product shot into a 16:9 banner, and say what fills the new space.

For a single image edit, use three moves: name the target precisely enough that only one thing matches, say what it changes into, then list what stays. Our product edit above follows it exactly: "Change the dripper and its saucer in image 1 to a glossy cobalt blue glaze (#1f4fd8) ... Keep the glass carafe, the coffee stream, the steam, the water droplets, the lighting and the framing exactly the same."

For several images, give each one a role: a subject to keep recognizable, a product whose shape and label must hold, a setting everything goes into, or a style to apply. Then say how the pieces fit, including whose light wins. Our cabin composite used: "Put the sage green ceramic dripper and glass carafe from image 2 on the walnut table in image 1, beside the stack of books, lit by the warm late afternoon window light of image 1 with a soft shadow on the table. Keep the room, the leather chair, the window view and the camera angle of image 1 exactly the same."

Removing something is an instruction, not a negation: "Remove the cat from the sofa" works as an edit prompt. Season, weather and time of day changes can be very short: "Change this to winter".

FLUX 3 Image on PixelDojo has a Web Grounding switch, on by default. With it on, the model researches your prompt with web and image search before it generates. That is what you want when the picture has to match something that exists: a landmark, a specific product, a current event, a place you have never seen.

Write the real thing by its name and the model does the looking up. Our Ponte Vecchio prompt named the bridge, the city and the river, then spent the rest of the words on what only we could decide: blue hour, the rowing scull, the camera at water level, the 35mm lens.

Grounding adds a research step to the job, so a grounded image can take longer than an ungrounded one. For invented scenes and pure style work there is nothing to look up, and switching grounding off keeps the run focused on your description.