Fal Flux Prompting

fal.ai Flux Image Generation

API Basics

curl -X POST "https://fal.run/{MODEL}" \
  -H "Authorization: Key $FAL_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "...", "image_size": "portrait_4_3", "num_images": 1}'

Models & Pricing

Pricing changes. Verify current prices on fal.ai before cost-sensitive runs. Current snapshot verified against official fal model pages on 2026-04-23. All per-megapixel prices round up to nearest MP unless the model page states a flat per-image price. 1024×768 = 1 MP.

API Schemas by Model

FLUX.1 [dev] — fal-ai/flux/dev

FLUX.1 [schnell] — fal-ai/flux/schnell

Same schema as dev, except:

• num_inference_steps default: 4 (designed for 1-4 steps)

FLUX 1.1 [pro] — fal-ai/flux-pro/v1.1

No num_inference_steps or guidance_scale — model controls these internally.

FLUX.2 [pro] — fal-ai/flux-2-pro

No num_inference_steps, guidance_scale, or num_images.

FLUX.2 [flex] — fal-ai/flux-2-flex

Text-to-image endpoint: fal-ai/flux-2-flex Image editing endpoint: fal-ai/flux-2-flex/edit

Edit endpoint adds: image_urls (string array, required) — input images to edit.

Kontext [pro] — fal-ai/flux-pro/kontext

Uses aspect_ratio instead of image_size. Uses image_url (singular) not image_urls.

FLUX.1 LoRA — fal-ai/flux-lora

Same schema as dev, plus:

LoraWeight object: {"path": "url-or-path", "scale": 1.0} (scale range 0-4)

Kontext LoRA — fal-ai/flux-kontext-lora

Same schema as Kontext, plus:

Uses resolution_mode instead of aspect_ratio. Default output format is png.

Response Schema (all models)

{
  "images": [{"url": "https://...", "width": 1024, "height": 768, "content_type": "image/jpeg"}],
  "seed": 12345,
  "prompt": "the prompt used",
  "has_nsfw_concepts": [false],
  "timings": {}
}

Standard Size Presets

Prompting Reference

Distilled from the guides listed at the bottom. When in doubt, consult the originals.

Prompt Structure

Flux weighs earlier tokens more heavily. Structure prompts as: Subject → Action → Style → Context [BFL, fal.ai]

• ✅ "Luxury leather handbag, displayed on marble surface, soft directional lighting, warm amber tones"
• ❌ "Create an image with soft lighting and warm tones featuring a leather handbag"

Prompt Length

• schnell token limit: 256 tokens. dev/pro: 512 tokens. [fal.ai how-to]
• Sweet spot: 40-50 words for most use cases [fal.ai how-to]
• Under 10 words triggers internal expansion from training data [fal.ai how-to]
• Over 200 words gets internally summarized, potentially dropping details [fal.ai how-to]

Natural Language, Not Keywords

Flux uses T5/Mistral text encoders. It wants sentences, not tag lists. [fal.ai how-to]

• ✅ "A woman in a red silk dress standing barefoot on a sandy beach at sunset, warm golden light behind her, shallow depth of field"
• ❌ "woman, red dress, beach, sunset, bokeh, 8k, masterpiece"

Weight syntax like (emphasis)++ or (word:1.5) has no effect — the model ignores it. [fal.ai how-to]

No Negative Prompts

Flux is guidance-distilled and does not support negative prompts. Describe what you want, not what to avoid. [BFL, fal.ai how-to]

• ✅ "sharp focus, crisp detail, accurate hands, natural proportions"
• ❌ "blurry, low quality, bad hands, deformed"

Camera & Lens References for Photorealism

Naming specific camera systems shapes output character. [BFL, fal.ai how-to]

• "Shot on Canon EOS R5, 85mm lens at f/2.8" → shallow DOF, specific skin tones
• "Shot on iPhone 16" → casual, candid smartphone aesthetic
• "Hasselblad X2D, medium format, natural light" → editorial quality, fine tonal gradations
• "Shot on Kodak Portra 400" → natural grain, organic colors

Lens focal lengths: 14-24mm wide/dramatic, 35-50mm natural, 70-85mm portrait, 100mm+ telephoto compression. [fal.ai prompt guide]

Depth of field: f/1.4-f/2.8 shallow blur, f/4-f/5.6 moderate, f/8-f/16 deep/sharp. [fal.ai prompt guide]

Lighting

Describe how light interacts with the scene, not just the condition name. [fal.ai how-to]

• ✅ "warm golden sunset light streaming through the window, casting long shadows across the hardwood floor"
• ❌ "golden hour lighting"

HEX Color Codes

Associate hex codes with specific objects. [BFL]

• "apple in color #0047AB"
• "walls in hex #C4725A, sofa hex #1B6B6F"
• Gradients: "vase is a gradient, starting with color #02eb3c and finishing with color #edfa3c"

Vague references like "use #FF0000 somewhere" produce inconsistent results. [BFL]

Text Rendering

Flux renders text well with clear specifications. [BFL, fal.ai how-to]

• Use quotation marks: "A chalkboard sign that says 'TODAY'S SPECIAL: LAVENDER LATTE'"
• Specify font properties: "Bold serif font in dark green, centered on a cream-colored banner"
• Case sensitivity: ALL CAPS in prompt → ALL CAPS output [fal.ai how-to]
• 2-5 words renders most reliably; full sentences have higher error rates [fal.ai how-to]
• Clean backgrounds improve legibility [fal.ai how-to]

JSON Structured Prompts (FLUX.2)

For complex multi-subject scenes, Flux 2 accepts JSON: [BFL, fal.ai how-to]

{
  "scene": "A dimly lit jazz club in 1960s New York",
  "subjects": [
    {
      "type": "musician",
      "description": "man in a charcoal suit, salt-and-pepper beard",
      "pose": "playing an upright bass with closed eyes",
      "position": "foreground left"
    }
  ],
  "style": "Cinematic film photography",
  "color_palette": ["#2C1810", "#D4A574", "#8B4513"],
  "lighting": "Single warm spotlight from above, smoke catching the light",
  "mood": "Intimate and contemplative",
  "composition": "rule of thirds",
  "camera": {"angle": "slightly low angle", "lens": "50mm"}
}

Use JSON for multi-subject precision. Natural language works equally well for single-subject images. [BFL]

Painterly / Non-Photographic Styles

• Flux has a photographic bias — painterly styles may require effort [sandner.art]
• Lower inference steps (8-10 instead of 20-30) can help with painterly results [sandner.art]
• Flux does not closely copy specific artist styles [sandner.art]
• Describe techniques directly: "Oil painting with soft diffused lighting, subtle chiaroscuro, rich color palette with thin translucent layers, controlled brushwork" [sandner.art]

Resolution Limits

• Minimum: 64×64. Maximum: 4 MP (e.g. 2048×2048). [BFL]
• Dimensions must be multiples of 16. [BFL]

Composition

Flux understands compositional terms: rule of thirds, golden spiral, triangular arrangement, diagonal energy. [fal.ai prompt guide]

Prompting Guide Sources

• Official BFL prompting guide (Flux 2): https://docs.bfl.ml/guides/prompting_guide_flux2
• fal.ai Flux 2 prompt guide: https://fal.ai/learn/devs/flux-2-prompt-guide
• fal.ai how to use Flux: https://fal.ai/learn/tools/how-to-use-flux
• Art styles in Flux (tested gallery): https://sandner.art/prompting-art-and-design-styles-in-flux-in-forge-and-comfyui/
• Flux style test gallery: https://enragedantelope.github.io/Styles-FluxDev/