Prompt Guide: Flux.2 [Klein]

Flux.2 [Klein] is a highly responsive, compact image generation model family developed by Black Forest Labs. Powered by an advanced T5 text encoder, its deep semantic understanding enables precise, instruction-level control over style, composition, text rendering, and exact colors.

This guide details the prompting techniques required to unlock the full creative potential of the Flux.2 [Klein] model family.

---

1. Prompting with Natural Language

Flux.2 performs best when your prompt reads like a clear, natural description of the image you want to generate, rather than a "tag-salad" or chaotic collection of comma-separated keywords.

• Explain Relationships: Use verbs and prepositions to describe how elements in the image relate to each other (e.g., "A wooden mug resting on a rustic table next to a steaming bowl of soup" instead of "wooden mug, table, soup, rustic").
• Describe Intent: Clearly explain what should be in the image, how it should look, and what mood or style it should convey.
• Refine Iteratively: Start with a simple description, evaluate the output, and add details to adjust the subject, lighting, or framing one step at a time.

---

2. Working Without Negatives: The Replacement Strategy

Flux.2 [Klein] does not rely on a separate negative prompt field. Instead of trying to list what the model should exclude (which often confuses the text encoder and causes those exact elements to appear), use the Replacement Strategy to describe the positive visual state of the scene.

When you catch yourself writing a negative phrase, convert it into what you actually want to see:

Think visually. Describe what is present in the scene to occupy the space rather than instructing the model on what to omit.

---

3. Style, Aesthetics & Mood Annotations

FLUX.2 can generate photorealistic images, complex illustrations, or fusion styles. You can apply specific camera, era, and film stock descriptions, or use explicit mood annotations at the end of your prompt.

Photorealistic Eras & Film Stocks

To achieve a distinct photographic look, specify the era, camera, or film stock rather than using generic quality descriptors like "photorealistic":

Style & Mood Annotations

To maintain a consistent aesthetic, you can append style and mood tags at the end of your prompt using this pattern:

[Scene description]. Style: Country chic meets luxury lifestyle editorial. Mood: Serene, romantic, grounded.

---

4. Exact Color Control with Hex Codes

One of Flux.2's most powerful features is its ability to match exact colors using standard Hex Codes. This is invaluable for brand-accurate marketing, product designs, or color-themed illustrations.

To use hex colors, specify the hexadecimal code directly in your prompt describing the color of specific elements or gradients:

• Single Color Precision: A sleek luxury eyeshadow palette with 6 pans, top row color #B76E79, middle row color #CD7F32, bottom row color #800020
• Color Gradients: A glass vase on a wooden table, the color of the vase is a gradient starting with color #02eb3c and finishing with color #edfa3c

---

5. Structured Prompting (JSON Prompts)

Flux.2 [Klein] natively understands structured visual descriptions. Rather than writing a single chaotic paragraph, you can format your prompt as a JSON-like structured block to organize layout, lighting, style, and camera attributes separately.

Structured Prompt Example:

{
  "subject": "A minimalist glass bottle containing orange essential oil",
  "background": "a smooth, light concrete tabletop against a clean pastel wall",
  "lighting": "soft, warm natural morning light entering from a side window",
  "style": "clean product photography, extremely detailed",
  "camera_angle": "eye-level, centered, macro lens",
  "composition": "shallow depth of field, soft shadows"
}

Copy and paste the structured JSON format directly into the prompt box. The model's transformer is trained to interpret these structured keys with extreme precision.

---

6. Legible English Text Rendering & Typography

Flux.2 [Klein] is exceptionally good at rendering legible English text inside generated images. To render characters accurately, follow these three rules:

1. Enclose in Quotation Marks: Always wrap the exact text to be rendered in double quotes (e.g., "COFFEE").
2. Describe Placement: State precisely where the text should appear in the composition (e.g., "The text 'JAVA' appears in bold letters centered above the doorway").
3. Specify the Typography Style: Describe the font characteristics, such as "neon", "serif", "sans-serif", "raised chrome", or "chipped paint".

Typography Styles Reference Table

Text Rendering Tips

• Front-load Text Descriptions: Place descriptions of text near the beginning of your prompt for higher accuracy.
• Keep Text Short: Short text strings render with much higher accuracy than long sentences.
• Brand Colors: Combine hex codes with text styles: "The logo text 'ACME' in color #FF5733".

---

7. Lenses, Lighting & Style Fusions

To elevate your generations from generic AI art to professional-grade creative outputs, apply specific photography, lighting, and style constraints:

Camera & Lenses

• Be specific with hardware descriptions: Shot on Hasselblad X2D, 80mm lens, f/2.8 or Canon 5D Mark IV, 35mm f/1.4 lens.
• Describe depth of field: shallow depth of field, razor-sharp focus on the subject, background softly blurred.

Lighting Setups

• Rembrandt Lighting: Triangle of light on the face for dramatic, classic portraits.
• Split Lighting: High contrast lighting where only half the face is illuminated.
• Chiaroscuro: High contrast between deep shadows and bright spotlights.
• Practical Lighting: Relying on light sources visible in the scene (e.g., neon signs, desk lamps, LED strips).
• Environmental Light: Specify soft window light, golden hour, or cool moody blue hour light.

Style Fusions

You can merge contrasting artistic concepts with a unifying color palette:

An ancient Greek marble statue, infused with cyberpunk neon lighting, holographic overlays, and electric blue and magenta glow effects, set against a dark futuristic background.

---

8. Image Editing with FLUX.2

FLUX.2 supports natural language image editing, allowing you to modify existing images using single or multiple references.

Single-Reference Editing

Single-reference editing allows you to modify a single image using natural language instructions. The model excels at identifying the subject and modifying the scene according to your positive description.

Key Use Cases & Prompt Examples:

• Background Replacement: Specify a new background to swap out the old one.
  • Example: "Replace the background with a sun-drenched sunset beach, keeping the subject in focus"
• Style Transfer: Convert the style of an image into another art medium or look.
  • Example: "Turn this photograph into an impasto oil painting with visible palette knife strokes"
• Object Manipulation: Add, remove, or replace elements within the frame.
  • Example: "Replace the bicycle with a brown horse standing in the same spot"
• Color & Material Swaps: Control colors precisely using hex codes or change materials.
  • Example: "Change the color of the shirt to #800020" or "Transform the wooden coffee mug into polished silver"
• Lighting & Environment: Modify seasonal details, weather, or time of day.
  • Example: "Add snow falling on the ground and trees, changing the season to winter while keeping the house unchanged"
• Text Editing: Replace or insert typography.
  • Example: "Change the neon text on the sign to read 'OPEN NOW' in warm orange glow"

---

Multi-Reference Editing

Multi-reference editing lets you combine multiple reference images to compose new scenes, transfer styles, or maintain character consistency.

Key Use Cases:

• Scene Compositing: Place a subject from one image into a setting from another image.
  • Example: "Place the cat from image 1 on the sofa in the room shown in image 2"
• Style & Pattern Transfer: Map the artistic style or texture from one reference onto the subject of another.
  • Example: "Redraw the car in image 1 in the impasto style of image 2" or "Apply the animal print pattern from image 2 onto the ceramic plate in image 1"
• Product & Fashion Composites: Combine styled clothing references into a single outfit, or place product images into styled mockups.
• Character Consistency: Reference facial details or clothing from input images to maintain character identity across multiple generated scenes.

---

9. Prompt Templates

Use these pre-configured templates as starting points in the Prompt Editor:

Professional Corporate Portrait

A medium-shot portrait of a smiling adult woman in her late 30s, professional business casual clothing, shot on Sony A7IV, 85mm lens, f/1.8, warm window lighting from the side, clean modern office background softly blurred, realistic photography, no logos, no text.

Flat Vector Illustration

A flat vector illustration of a person reading a book on a green park bench, stylized green trees in the background, limited color palette of warm orange and cool teal, minimal shading, clean modern design, no outlines, simple unmarked background.

Studio Product Shot

A studio product shot of a luxury watch, metallic gold bezel, dark leather strap, placed on a black obsidian surface, dramatic Rembrandt lighting, dark charcoal background with a subtle gradient, extremely sharp details, no text, no branding.