What is Flux.1 Model? Know All About New AI Image Generator Tool

There is a new champion in the arena of artificial intelligence (AI) image generators, Flux.1. It is developed by  Black Forest Labs, a brand new generative AI startup, that launched on August 1, 2024. The company was founded by a team of researchers who previously contributed to developing Stable Diffusion and invented the latent diffusion technique. 

The mission of Black Forest Labs is “to develop and advance state-of-the-art generative deep learning models for media such as images and videos, and to push the boundaries of creativity, efficiency, and diversity.” The startup has released the FLUX.1 suite of models that change the frontiers of text-to-image generation. 

Microsoft Designer App: How to Use the AI Image Generator Tool for Editing and Creation?

What is the Flux.1 Model?

FLUX.1 models are a suite of advanced text-to-image AI models developed by the genAI startup, Black Forest Labs. These models leverage a hybrid architecture that combines transformer and diffusion methods to generate high-quality images from text prompts. With 12 billion parameters, Flux.1 has three variants- pro, dev, and schnell.

• FLUX.1 [pro]: This is the premium version of the FLUX.1 model, which offers the highest performance in image generation, with excellent prompt accuracy, visual quality, and detail. It is available for commercial use through APIs on platforms like Replicate and Fal.ai. You can reach out to flux@blackforestlabs.ai for a customized enterprise solution.

• FLUX.1 [dev]: This is an open-weight model designed for non-commercial use. FLUX.1 [dev] offers similar quality to the pro version, however, this one is optimized for efficiency. You can access the model weights on HuggingFace and use them on Replicate or Fal.ai.

• FLUX.1 [schnell]: This is the fastest model tailored for local development and personal use. It is openly available under an Apache2.0 license. You can access the weights on Hugging Face. This version is integrated into ComfyUI and compatible with tools like GitHub and HuggingFace’s Diffusers.

All FLUX.1 model variants support a diverse range of aspect ratios and resolutions in 0.1 and 2.0 megapixels, as shown in the following example.

OpenAI Sora vs Kling AI: Differences Between AI Video Generators

Performance and Capabilities

The Flux.1 models have a hybrid structure that follows both transformer and diffusion methods, with a parameter scale of 12 billion. This allows for improved model proficiency and hardware efficiency. Using rotary positional embeddings and parallel attention layers enables Flux.1 models to surpass previous state-of-the-art diffusion models.

Flux.1 models show superior performance in visual quality, prompt adherence, and output diversity. Black Forest Labs claims that the photorealistic images generated by Flux.1 are comparable to OpenAI’s DALL-E 3 and other major text-to-image generators. 

One special feature of the Flux.1 model is its ability to generate precise portrayals of the human body, including hands. Since their inception, AI image generators, including the leading ones, have struggled with portraying human hands. Some generators slice off an extra finger or two, whereas others include unrealistic proportions or awkward poses. Flux.1, however, excels in accurately depicting the intricate details of hands. 

List of 13 FREE AI Image Generators in 2024

Flux.1 Image Generation

We tried the Flux.1 [Schnell] model to generate a few images. We are listing our prompts and the outputs we received from the AI generator below. 

Prompt 1: Fruit cereal spelling out the words “FRUITY LOOPY”, tasty, food photography, dynamic shot

Output:

Prompt 2: A pair of hands holding a box full of donuts, realistic

Output: 

Prompt 3: A fashionable woman walking the streets of the city wearing a red dress and black sunglasses, realistic, city, neon

Output:

Prompt 4: A Ghibli studio style prairie with a small house in the middle, anime, ghibli studio

Output:

Prompt 5: An alpaca wearing a tuxedo, attending the birthday party of a kitten, cartoon

Output:

Prompt 6: A powerful entity watching over the universe

Output:

The Bottom Line

Flux.1 models are genuinely impressive text-to-image generators. Despite being free, the [Schnell] model generated images faster than any other model. We used the the Flux.1 [Schnell] model to generate every image used in this article, including the featured one.

To test its prompt accuracy and coherence, let’s take the example of prompt number 5. You will find the words ‘Happy Birthday’ in perfect spelling written on a sign hanging in the background. There is no gibberish. Also, you can look at prompt number 3, where the woman’s hands actually look like a human’s, free of any anatomical anomaly. 

The model is exceptionally fast and delivers results that exceed what can be expected from a free model. Though it struggles sometimes with anomalies and artifacts, the model still has a better logical coherence compared to similar models. To sum up, Flux.1 models are decent image generators that sometimes struggle with anomalies. Black Forest Labs are now working on SOTA, their suite of competitive generative text-to-video systems.

How to Use Midjourney AI to Create Stunning Images

Common FAQs