FLUX Image Generation

Instructions

Generate images using FLUX models through the ITI image generation API at http://localhost:7860 (native) or http://stable-diffusion:7860 (Docker network).

Two FLUX backends are available:

Backend	Model	Speed	Cost	License
flux-local	FLUX.2 Klein 4B	Sub-second (GPU)	Free	Apache 2.0
flux-api	Any FLUX model via BFL cloud	2-10s	$0.014-0.07/image	Commercial API

Text-to-image with FLUX API (cloud):

curl -X POST http://localhost:7860/api/txt2img \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A professional product photograph of wireless earbuds on a marble surface, soft studio lighting, photorealistic",
    "backend": "flux-api",
    "bfl_model": "flux-2-pro-preview",
    "width": 1024,
    "height": 1024
  }'

Text-to-image with FLUX Klein (local):

curl -X POST http://localhost:7860/api/txt2img \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A cat holding a sign that says hello world",
    "backend": "flux-local",
    "width": 1024,
    "height": 1024
  }'

Image editing (multi-reference):

curl -X POST "http://localhost:7860/api/img2img?prompt=change+background+to+beach+at+sunset&backend=flux-api" \
  -F "image=@photo.png"

BFL API model selection guide:

Model	Best For	Price	Key Feature
flux-2-pro-preview	Production quality, general use	$0.03/MP	Latest quality
flux-2-pro	Reproducible workflows	$0.03/MP	Fixed snapshot
flux-2-max	Highest quality, real-time info	$0.07/MP	Grounding search
flux-2-flex	Typography, text overlays	$0.06/MP	Best text rendering
flux-2-klein-4b	High volume, budget	$0.014+/img	Sub-second
flux-2-klein-9b-preview	Balanced quality/speed	$0.015+/img	Latest Klein

Prompt engineering for FLUX:

FLUX uses natural language prompts — not the keyword-heavy style of Stable Diffusion:

• Write descriptive sentences: “A golden retriever sitting in a sunlit garden with dappled shadows”
• Text rendering works: ‘A coffee mug with the text “Good Morning” in serif font’
• Hex colors are precise: “A vase with color gradient from #02eb3c to #edfa3c”
• Structured prompting for complex scenes:

{
  "subject": "Luxury watch on dark velvet",
  "background": "deep black, subtle gradient",
  "lighting": "dramatic side lighting, warm highlights",
  "style": "product photography, high contrast",
  "camera_angle": "45 degree overhead",
  "composition": "centered, shallow depth of field"
}

• Multi-reference editing: FLUX can combine elements from multiple source images while maintaining identity (up to 4 images for Klein, 10 for Pro/Max)
• Negative prompts are NOT used — quality is controlled through positive prompting
• Caption upsampling (caption_upsample_temperature: 0.15) can improve output when using Flux2Pipeline (dev model)

FLUX Klein parameter defaults:

Parameter	Distilled (default)	Base (fine-tuning)
num_inference_steps	4	20-50
guidance_scale	1.0-4.0	4.0-7.0
Max resolution	2048×2048	2048×2048

FLUX vs Stable Diffusion — when to use which:

Capability	FLUX	Stable Diffusion
Photorealism	Superior	Good
Text in images	Reliable	Poor
Prompt following	Natural language, precise	Keyword-heavy
Color control	Hex code exact matching	Approximate
Multi-reference editing	Built-in (no fine-tuning)	Requires LoRA/fine-tune
Community LoRAs	Growing	Largest ecosystem
Speed (local)	4 steps (Klein)	30-50 steps
VRAM	~13GB (Klein 4B)	~5GB (SD 2.1)

n8n workflow integration:

Use an HTTP Request node. Add "backend": "flux-api" or "flux-local" to the JSON body:

{
  "prompt": "Marketing hero image for SaaS product launch, modern tech aesthetic",
  "backend": "flux-api",
  "bfl_model": "flux-2-pro-preview",
  "width": 1440,
  "height": 1024
}

Common use cases:

• Marketing asset generation with brand-consistent colors (hex code control)
• Product mockup creation with text labels and pricing
• Social media graphics with readable text overlays
• Blog post featured images from natural language descriptions
• E-commerce product photography variations
• Concept visualization for consulting deliverables
• Character-consistent image sets (multi-reference editing)

Resource awareness:

Backend	Speed (1024×1024)	Memory	Cost
FLUX Klein local (MPS)	~5-15s	~13 GB	Free
FLUX API (cloud)	~2-10s	Negligible	$0.014-0.07/img