FLUX.2 [klein] - Fast Generation Guide

FLUX.2 [klein] - Fast Generation Guide

FLUX.2 [klein] is our fastest model family, delivering state-of-the-art image quality with sub-second inference times. It's designed for applications requiring real-time generation.

Two Variants

Running Locally

Both [klein] variants are available for local deployment via Hugging Face:

• [klein] 4B - Apache 2.0 license allows full commercial use locally

• [klein] 9B - Non-commercial local use; commercial use requires API

• Base variants - Undistilled models available for fine-tuning and customization

Prompting Tips for [klein]

Important: [klein] does not include prompt upsampling. For best results:

• Use narrative prompting - Describe scenes like a novelist, not keyword lists

• Be specific - Include details about lighting, composition, style

• Include context - Describe the setting, mood, and atmosphere

Example prompt:

"A weathered fisherman mends his nets on a wooden dock at golden hour. His calloused hands work methodically while seagulls circle overhead. The warm light catches the spray of salt water in the air, creating a hazy, nostalgic atmosphere."

Supported Features

• Text-to-image - Generate from text descriptions

• Image editing - Modify existing images with prompts

• Multi-reference - Up to 4 reference images

• Exact color matching - Use hex codes like #02eb3c in prompts

• Resolution - 64x64 to 4MP (dimensions must be multiples of 16)

What [klein] Does NOT Support

• Prompt upsampling (write detailed prompts manually)

• Grounding search (exclusive to [max])

• More than 4 reference images

API Usage

Endpoints:

• POST /v1/flux-2-klein-4b

• POST /v1/flux-2-klein-9b

For complete API documentation, see docs.bfl.ai/flux_2/flux2_text_to_image