FLUX.1 [schnell]

Family

FLUX.1

Type

Foundation Model

License

Apache-2.0 License

Released

2024-08-01

How To Use

Lab Station OS

The simplest way to self-host FLUX.1 [schnell]. Launch a dedicated cloud GPU server running Lab Station OS to download and serve the model using any compatible app or framework.

Direct Download

Download model weights for local inference. Must be used with a compatible app, notebook, or codebase. May run slowly, or not work at all, depending on your system resources, particularly GPU(s) and available VRAM.

Browse Compatible Apps

comfyanonymous /

ComfyUI

Generate images and videos using a powerful low-level workflow graph builder - the fastest, most flexible, and most advanced visual generation UI.

lllyasviel /

Stable Diffusion WebUI Forge

Forge is a platform built on top of Stable Diffusion WebUI to make development easier, optimize resource management, speed up inference, and study experimental features.

bmaltais /

Kohya's GUI

Train your own LoRAs and finetunes for Stable Diffusion and Flux using this popular GUI for the Kohya trainers.

Model Report

black-forest-labs / FLUX.1 [schnell]

FLUX.1 [schnell] is a 12B parameter text-to-image model optimized for 1-4 step generation, using flow matching and parallel diffusion transformers. Part of a model family focused on speed, it supports resolutions up to 2MP while maintaining output quality comparable to traditional diffusion models.

Explore the Future of AI

Your server, your data, under your control

FLUX.1 [schnell] represents a significant advancement in few-step text-to-image generation, combining innovative architecture with high performance and accessibility. As part of the FLUX.1 model family, it specifically targets fast inference while maintaining competitive output quality.

Architecture and Technical Specifications

FLUX.1 [schnell] is built on a hybrid architecture combining multimodal and parallel diffusion transformer blocks, scaled to 12 billion parameters. The model leverages several key technologies:

Latent adversarial diffusion distillation
Flow matching
Rotary positional embeddings
Parallel attention layers

This architecture enables high-quality image generation in just 1-4 steps, making it the most advanced few-step model available at the time of its release. The model's design focuses on hardware efficiency while maintaining competitive performance with closed-source alternatives.

Capabilities and Performance

FLUX.1 [schnell] demonstrates impressive capabilities across various metrics, competing favorably with models like Midjourney v6.0 and DALL-E 3 (HD) despite its optimization for speed. The model supports diverse aspect ratios and resolutions between 0.1 and 2.0 megapixels, offering flexibility for different use cases.

Key performance features include:

High-quality image generation
Strong prompt adherence
Support for various image resolutions
Fast inference times

Model Family and Differentiation

The FLUX.1 family includes three variants:

FLUX.1 [schnell]: Optimized for speed and personal use
FLUX.1 [pro]: Highest performance variant
FLUX.1 [dev]: Open-weight model for non-commercial applications

Limitations and Usage Considerations

The model has several known limitations:

Cannot provide factual information
May amplify societal biases
Occasional prompt matching failures
Sensitivity to prompting style

Licensing and Legal Considerations

FLUX.1 [schnell] is released under the Apache 2.0 license, allowing for personal, scientific, and commercial applications. Usage restrictions prohibit:

Violation of applicable laws
Creation of harmful content
Child exploitation
Disinformation
Non-consensual nudity
Fully automated decision-making affecting individual rights

References

Black Forest Labs Announcement - Official release announcement and technical details
Hugging Face Model Page - Model weights and implementation details
GitHub Repository - Reference implementation and sampling code
Diffusers Documentation - Integration guide for Hugging Face Diffusers
ComfyUI GitHub - Local inference interface documentation