FLUX Image Generation Skills

Overview

Official first-party image generation skills from Black Forest Labs, creators of the FLUX model family. These skills cover the full FLUX API surface — text-to-image, image-to-image, inpainting, and upscaling — using the production BFL API.

FLUX.1 models are state-of-the-art for photorealism, typography, and instruction following compared to SDXL and earlier diffusion models.

When to Use

• User asks to generate, create, or make an image from a text description
• User wants high-quality photorealistic images or illustrations
• User needs image-to-image transformations (style transfer, variations)
• User wants to inpaint or edit a specific region of an existing image
• User wants to upscale or enhance an existing image
• Don't use for: video generation (use MeiGen-AI-Design-MCP), UI design mockups (use open-design)

Prerequisites

Get a BFL API key at api.bfl.ml:

export BFL_API_KEY=your_key_here

Install the BFL Python client (if using scripted generation):

pip install bfl

FLUX Model Variants

Model	Use Case	Notes
flux.1-pro	Best quality, photorealism	Slowest, highest cost
flux.1-dev	Development and testing	Open weights available
flux.1-schnell	Fast generation	Apache 2.0, fastest
flux.1-fill-pro	Inpainting / outpainting	Requires mask
flux.1-canny-pro	Edge-guided generation	Requires canny edge map
flux.1-depth-pro	Depth-guided generation	Requires depth map

Usage

Natural language — Hermes will invoke the skill automatically for image requests:

"Generate a photorealistic image of a mountain lake at sunset"
"Create an illustration of a robot reading a book, flat design style"
"Make a product photo of a white coffee mug on a wooden table"
"Generate a logo concept for a tech startup called NovaSpark"

Or load the skill explicitly:

/skill black-forest-labs-flux

API Reference

Text-to-image:

import bfl
result = bfl.generate(
    prompt="A serene mountain lake at golden hour",
    model="flux.1-pro",
    width=1024,
    height=1024,
    steps=28,
    guidance=3.5
)
result.save("output.png")

Image-to-image:

result = bfl.generate(
    prompt="Same scene but in winter with snow",
    model="flux.1-pro",
    image="input.png",
    strength=0.75
)

Inpainting:

result = bfl.generate(
    prompt="Replace the background with a starry night sky",
    model="flux.1-fill-pro",
    image="input.png",
    mask="mask.png"
)

Prompt Tips

• Be specific about style: "oil painting", "photorealistic", "flat vector", "watercolor", "cinematic"
• Include lighting: "golden hour", "soft studio lighting", "dramatic side lighting"
• Specify aspect ratio intent: FLUX handles 1:1, 16:9, 9:16, and 4:3 natively via width/height
• Typography: FLUX.1 handles text in images better than most models — specify exact text in quotes

Common Pitfalls

1. Rate limits. BFL API has per-minute and per-day rate limits depending on tier. Add retry logic with exponential backoff for batch generation.
2. Long generation times. flux.1-pro can take 15–30 seconds per image. Use flux.1-schnell for rapid iteration, switch to flux.1-pro for final output.
3. Prompt injection. Avoid user-controlled strings directly in prompts without sanitization if building an app on top.
4. Image size constraints. Width and height must be multiples of 32. Values outside 256–2048 will error.
5. API key scope. BFL API keys are scoped to a workspace — ensure the key has access to the model variant you're calling.

Verification Checklist

• BFL_API_KEY set and valid
• pip install bfl completed successfully
• Test generation returns an image: "Generate a simple image of a red circle on white background"
• Output file is a valid PNG/JPEG that opens correctly
• Model variant matches your quality/speed requirement