How to prompt FLUX.2 Pro
Black Forest Labs · BFL Playground / API · Guidance v2, verified Sep 23, 2026 against 2 official sources
How to structure your prompt
Natural language in priority order, because FLUX.2 weights what comes first: main subject, key action, critical style, essential context (setting, lighting, time, mood), then secondary details. Usually 30-80 words; 10-30 for quick concepts; 80+ only for complex scenes. For photorealism name a camera, lens and film stock or era rather than 'professional photo'. Attach each hex color to a specific object. Put exact lettering in quotation marks with its placement, typography, size and color. Describe the positive alternative instead of exclusions (for example 'sharp focus throughout' instead of 'no blur'). Set dimensions in the BFL interface or API.
Length and syntax
Up to 2,000 characters. Longer prompts are not better if the extra words do not change what appears on screen.
Exact text in quotation marks. Hex colors written as 'color #RRGGBB' or 'hex #RRGGBB' directly after the object they apply to. Structured JSON prompts are also understood, but this card produces natural language. No parameter flags.
Prompt skeleton
[Main subject] [key action], [critical style], [setting, lighting, time and mood], [secondary details]. [Camera, lens and film stock, for photos]. [The text "..." in [typography, size, color], placed [position].]
What it’s best at
- photorealism and product photography
- legible text in images
- precise hex color control tied to objects
- editorial layouts, posters and infographics
- prompts in the native language of the scene
Known limitations
- no negative prompts; exclusions must be described positively
- hex codes not tied to a specific object give inconsistent results
- reference-image editing is a separate workflow not covered by this card
Avoid in your prompt
- negative prompts or 'no X' phrasing
- unrequested typography
- unsupported parameter flags
- hex codes not attached to an object
- burying the main subject late in the prompt
Other image models
Summarised from the provider’s official documentation by Promptuned. Results vary by model version and settings. FLUX.2 Pro is a trademark of its owner; Promptuned is not affiliated with Black Forest Labs.