FLUX.1: Black Forest Labs Unveils a Cutting-Edge Image Generation Model

FLUX.1 delivers state-of-the-art image generation performance with exceptional prompt tracking, visual quality, image detail, and output diversity. It maintains similar quality and rapid responsiveness while being more efficient than standard models of the same size.

What makes FLUX.1 stand out from the crowd?

FLUX.1 models excel in prompt adherence, visual excellence, intricate image details, and diverse outputs. Here are some standout features that have truly impressed us：

Exquisite Composition. FLUX.1 excels in understanding and executing complex instructions, placing elements precisely within an image. For instance, given the prompt: 'Two adorable spiders, one wearing a black hat and the other a maroon hat, both adorned with flowers, are hosting a miniature tea party. On a leaf, there's a small table and a teapot, captured in a macro photograph.' FLUX.1 flawlessly recreates this intricate scene:
Text Handling! Unlike older models that often confuse similar letters, FLUX.1 excels at managing tricky words with repeated letters. This makes it exceptionally suited for tasks requiring precise text design. For example, consider this photo of a blackboard in an old classroom. The blackboard has the chalk-written phrase, 'Let's make something really beautiful together,' with a red chalk heart at the end. Sunlight streams through the window, and FLUX.1 accurately captures this entire scene:
Realistic Hands. While hands have always been a challenge for AIs, FLUX.1 makes significant strides in this area. It generally produces hands with the correct number of fingers in the right places. Though not flawless, it represents a substantial improvement over previous models. FLUX.1 consistently outperforms any other open text-to-image model we've tested, setting a new standard in generating lifelike hands:

Model Introduction

This release primarily includes three models in the FLUX.1 series:

FLUX.1 [pro]: The premium version of FLUX.1, offering state-of-the-art image generation capabilities with exceptional prompt adherence, visual quality, image detail, and output diversity.
FLUX.1 [dev]: FLUX.1 [dev] is an open-guided distillation model intended for non-commercial use. Distilled directly from FLUX.1 [pro], FLUX.1 [dev] achieves similar quality and prompt adherence while being more efficient than standard models of the same size. It is available for non-commercial use.
FLUX.1[schnell]: The fastest model, designed for local development and personal use. FLUX.1 [schnell] is publicly available under the Apache 2.0 license and is supported by ComfyUI for direct use.

Detailed Architecture

All publicly available FLUX.1 models utilize a hybrid architecture that combines multimodal and parallel diffusion Transformer blocks, scaling up to 12 billion (12B) parameters. We have enhanced previous state-of-the-art diffusion models by employing a training method based on flow matching, which is a general and conceptually simple approach to generative modeling, with diffusion as a special case.
Additionally, the introduction of rotary positional embeddings and parallel attention layers has significantly improved both model performance and hardware efficiency. A more comprehensive technical report will be released in the near future.

Next Steps

FLUX.1 is an incredible model showcasing unprecedented intelligence and performance. Click https://app.aitubo.ai/models to experience it for free! By following this link, you'll be among the first to explore the powerful features and innovative capabilities of FLUX.1. Don't miss out—dive into a new level of smart technology today!