FLUX.1 Prompting Guide for Composition, Lighting, and Text

Write FLUX.1 prompts as natural-language descriptions of the subject, composition, lighting, and any exact lettering. Start with the scene, then revise one detail at a time. The local examples use Black Forest Labs' downloadable FLUX.1-dev model through Diffusers. Prompting basics, Model card

Start with a subject and a clear visual intention, then revise the parts that matter to the final image. The examples here are original practice prompts and review exercises, not benchmark results or guarantees.

What are the key facts about FLUX.1 prompting?

How should you describe a scene and lettering to FLUX.1?

BFL's prompting guidance recommends natural-language descriptions that identify the scene and relationships between its elements. Use that as a writing method: tell the model what occupies the image and how it is arranged. Prompting basics

For example, start with “A blue ceramic bowl on a linen tablecloth.” Then decide which missing information matters: whether the bowl contains fruit, whether the view is overhead, or whether the image needs space for a caption.

The publisher also recommends quoting exact text that should appear inside an image. Its style guide adds placement and font description as useful instructions for lettering. Prompting basics, Style guide

Try “The words 'MORNING MARKET' appear on a cream paper label attached to the basket.” This gives the lettering an object and location, making the result easier to inspect than a request for unspecified text somewhere in the scene.

What are the limits of FLUX.1 prompt control?

FLUX.1-dev's card warns that outputs can fail to match a prompt and that wording influences the result. Treat each requested detail as something to verify, especially when several objects interact. Model card

BFL guidance spans several FLUX generations. Apply the writing exercises here with FLUX.1-dev's documented settings; FLUX.2-specific limits belong to that model. Prompt building, Dev setup

Negative-prompt behavior depends on the model and pipeline. BFL's general guidance recommends describing the desired visible state, while the simple dev example used here has no negative-prompt argument. Technical guidance, Dev setup

For an uncluttered scene, describe a plain backdrop and a solitary subject. Then inspect whether those instructions were followed rather than relying on a long list of prohibited objects.

How do you write and test FLUX.1 prompts locally?

Establish the generation setup

Complete the Hugging Face repository's access process and install Diffusers with its supporting dependencies. The following adapts the model card's Python example, including CPU offloading and the documented dev guidance setting. Model card

import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()
image = pipe(
    "A blue ceramic bowl on linen, overhead view, soft window light",
    width=1024, height=1024, guidance_scale=3.5,
    num_inference_steps=50, max_sequence_length=512,
    generator=torch.Generator("cpu").manual_seed(42),
).images[0]
image.save("bowl.png")

Use a compatible PyTorch environment for the documented GPU/offload route. The example is a reproducible starting configuration, not a minimum-memory claim or a promise of a particular generation time. Pipeline documentation

Record the model identifier, prompt, seed, dimensions, guidance, and sampling steps. Keep those settings stable while exploring wording, so the experiment has a clear purpose.

Write the composition first

Begin with the subject and its relationship to the frame. For a recipe illustration, an overhead bowl might be appropriate; for an advertisement, a side view with room for copy might fit the task better.

An original composition prompt to try is: “A blue ceramic bowl containing ripe pears sits on the left half of a linen-covered table. The right half shows uninterrupted linen. View the scene from directly above.”

Check subject placement before adding styling. If the bowl occupies the whole frame, revise the framing instruction and compare again. If its position is right, preserve that wording while developing the light and materials.

Add lighting and surface detail

BFL's style guide separates light source, quality, direction, and color. Choose a small set of those instructions instead of using several competing mood descriptions. Lighting guidance

For the bowl scene, try: “Soft daylight enters from the upper left, creating a gentle shadow beside the bowl. The ceramic glaze has a subtle reflection, and the linen weave is visible.”

Inspect whether the requested surface treatment helps the intended design. If strong reflections dominate the fruit, change that instruction rather than adding another broad quality adjective.

Add lettering only when needed

Keep wording short for the first lettering exercise. Ask for a cream card beside the bowl bearing “PEAR SEASON” in dark serif lettering, then check the spelling, placement, and contrast independently.

For production artwork, decide whether generated lettering belongs in the image at all. If editable typography is a requirement, plan to add the final copy in your layout tool and use the generation to establish the scene.

Keep the acceptance criteria tied to the deliverable: a plausible paper label and an editable campaign headline serve different purposes.

Run a revision exercise

Save the initial bowl image, a composition revision, and a lighting revision. Name the files by what changed. Review them together and write one sentence explaining what each revision improved or disrupted.

Then test the same structure on a different subject, such as gardening tools on a bench. Reuse the questions about framing, materials, and text without assuming the previous wording is optimal for every scene.

How does prompting differ across FLUX.1 models?

The FLUX photography pillar develops photographic direction. For a themed exercise, use the sibling Halloween FLUX prompting guide.

What should you know when revising a FLUX.1 prompt?

How should I structure a FLUX.1 prompt?

For FLUX.1, describe the subject and its position in the scene, then add the lighting, style, and materials that matter. BFL recommends clear natural language and changing one important detail between revisions. Prompting basics

Should FLUX.1 prompts be long?

BFL recommends starting with a short FLUX description and adding only details that change the intended image. For FLUX.1-dev, follow its documented pipeline settings when testing that writing approach. Prompt-building guide, Model card

How do I request text inside a FLUX.1 image?

In a FLUX.1 prompt, quote the exact wording and specify its location and font style. Proofread the generated lettering before using the image. Style guide

Can FLUX.1-dev and schnell use the same sampling settings?

FLUX.1-dev and FLUX.1-schnell have different documented Diffusers sampling configurations. Select the checkpoint first and follow its pipeline example before comparing prompt revisions. Pipeline documentation

Sources

• FLUX.1 announcement
• FLUX.1-dev model card
• Diffusers FLUX pipelines
• BFL prompting basics
• BFL prompt construction
• BFL style, lighting, and lettering
• BFL technical prompting guidance
• FLUX.1 Kontext overview

Related guides on PromptZone

• Realistic Photos with FLUX
• Best SDXL Models in 2026
• The Ultimate Guide to Fooocus Image Prompts