AGEofLLMs.com
Search

Meet Flux: Open Source AI Image Generator

Calculating... Comments

A new, free, and open-source AI image generator from Black Forest Labs - Flux - is looking to be a promising tool for both hobbyists and professionals in the field of AI imagery. Some were even quick to label it a "Midjourney killer".

FLUX AI
FLUX AI

So Flux is a new groundbreaking AI image generator developed by Black Forest Labs, a team composed of former Stability.ai employees. What sets Flux apart is its open-source nature and free availability, making it accessible to a wide range of users, from hobbyists to professionals. This openness fosters community involvement, allowing for rapid development and improvement. Black Forest Labs aims to provide a high-quality tool that democratizes AI image generation, enabling anyone to create stunning visuals without the need for expensive subscriptions or proprietary software.

Flux Performance

Flux has shown remarkable performance in benchmarking tests, reportedly outperforming established models like Stable Diffusion 3, Midjourney V6, and Dolly 3. This superior performance is evident in the quality and realism of the images generated by Flux. The model excels in creating detailed textures, realistic lighting, and dynamic scenes, making it a powerful tool for both artistic and practical applications. The ability of Flux to handle complex prompts and generate high-quality images sets it apart from its competitors, making it an attractive option for users seeking top-tier AI image generation.

Flux Comes in 3 Versions

Flux provides three distinct versions to cater to different user needs:

  • Flux Pro (Commercial Use): This is the top-of-the-line version, offering state-of-the-art performance and capabilities. It is available for commercial use, making it suitable for professional applications such as graphic design, advertising, and film production.
  • Flux Dev (Non-Commercial): This version is intended for non-commercial use and is aimed at developers and enthusiasts who want to experiment with the model's capabilities. It provides a robust set of features for those looking to explore and innovate.
  • Flux Schnell (Fast): Schnell, which means "fast" in German, is designed for speed and efficiency. It offers quicker image generation, making it ideal for users who need rapid results without compromising too much on quality. 12B param rectified flow transformer distilled from FLUX.1 [pro] for 4 step generation.

Flux Examples and Capabilities

FLUX DEV image with text, generated by FLUX
FLUX DEV image with text, generated by FLUX

The model's ability to handle text within images is a standout feature. Flux can generate text with varying fonts and styles, adding a layer of realism and customization to the images. This makes it suitable for a wide range of applications, from creating promotional materials to designing virtual environments.

Flux has demonstrated impressive capabilities in generating realistic and cinematic images.

Here are some of my tests of the same prompt in both models, I could only use Flux Schnell, Dev is likely a bit better.

Realistic, blue hour, a couple in a tense, emotional conversation in the rain on a lonely urban street

Flux vs Midjourney same prompt
FLUX.1 [schnell] vs Midjourney same prompt

Another comparison for this surreal scene prompt:

Photorealistic monkey sitting still in the living room is in focus, seen behind the two people arguing with each other passionately ignoring the monkey

Monkey and people arguing; FLUX vs Midjourney
Monkey and people arguing; FLUX vs Midjourney

In my opinion, it's a really good result overall, but photorealism is still unparalleled in Midjourney. Flux's outputs look more like ChatGPT's Dall-e with improvements. Oh, also, the man's leg is living its own life, hahaha. But check out Midjourney's fingers fail! Anyway...

Flux can also create detailed and dynamic scenes with accurate lighting and textures, making the generated images look natural and lifelike.

Let's test this space prompt, verbatum what I've used in Midjourney:

Wide shot of the alien planet, tracking the spaceship landing, capturing the futuristic and otherworldly environment. Lots of motion and depth of field, futuristic, detailed, lots of light and shadow, ultra-realistic, super high quality, high resolution, 8K

Flux Schnell vs Midjourney, space exploration theme
Flux Schnell vs Midjourney, space exploration theme

A serene, mountainous landscape at twilight with lush greenery and terraced hills. A sleek, luxury car is driving down a wet, winding road lined with palm trees and lanterns. The road leads towards a cozy, warmly lit resort with traditional-style buildings. The misty atmosphere creates a dreamy, peaceful mood, with distant mountains softly fading into the fog.

Car & mountainous landscape in both models
Car & mountainous landscape in both models

These are definitely very promising results!

What's more, Flux has reportedly mastered human hands and fingers - something many current AIs struggle with. I won't be testing this enought times yet to verify, because once in a while hickups like this - as you see - can happen to the best of them, but I'm willing to trust that those images others are posting are the rule not the exception. But let's just run one last prompt, combining the finger test with text test:

Man and woman sitting in the cafe, woman is holding a cup of tea in her hands, and a man is waving at the camera. The sign behind them reads "Ageofllms.com Testing Us"

Oh damn! LOL

Fingers and text test with Flux Schnell
Fingers and text test with Flux Schnell

Flux Current Limitations

While Flux offers many advanced features, it currently lacks some capabilities that are available in other AI image generators. These include upscaling (increasing the resolution of images), inpainting (filling in missing parts of an image), and image-to-image generation (transforming an existing image into a new one based on a prompt). That's one of the features I love in Midjourney where you can reference an existing image as a style or as character. But, given the open-source nature of Flux, these limitations are expected to be addressed quickly by the community. Developers and users are likely to contribute improvements and new features, ensuring that Flux continues to evolve and meet the needs of its users.

How and Where to Try Flux

To try out Flux image creation, you have several options, ranging from online platforms to local installations. Here are the steps and platforms where you can get started:

  • For Quick and Easy Access: Use Hugging Face or Fall.ai.
  • For Advanced Users: Install Pinocchio and ComfyUI locally.
  • For Integrated Workflows: Explore platforms like Wand.

Hugging Face

At Hugging Face you might have limited free credits, but it's a quick way to get started.

Flux models are available throught the following links: Flux Schnell, Flux Dev

Use the provided interface to input your prompt and generate images.

Fall.ai

Fall.ai offers free credits and low-cost pricing plans, allowing you to use the Pro model. But you need a Github account to log in.

Flux Local Installation

Note: This method is more technical and may require a powerful PC.

Pinocchio and ComfyUI:

  1. Install Pinocchio. (Note: The installation process can be complex and may require technical knowledge.)
  2. Download ComfyUI.
  3. Download the Flux models (Dev or Schnell) from the provided URL.
  4. Open ComfyUI and navigate to the workflows folder.
  5. Open the Flux example PNG and drag it into the workflow.
  6. Input your prompt and generate images.

Black Forest Labs Future Plans

Right now, Black Forest Labs is focused on text-to-image generation, but they have plans to move into video generation next. They say that FLUX.1 will be the foundation for a new text-to-video model they're developing. This model will compete with others like OpenAI's Sora, Runway's Gen-3 Alpha, and Kuaishou's Kling, aiming to revolutionize how we create and edit media on demand. "Our video models will allow for precise creation and editing at high definition and unprecedented speed," the company claims.

While some enthusiasts have already hailed Flux as a potential "Midjourney killer," it's best to approach such claims with caution. Midjourney has established itself as a leading AI image generator with a unique aesthetic and workflow. Flux, on the other hand, offers impressive capabilities and is open-source, which could attract a different segment of users. But rather than viewing Flux as a replacement for Midjourney, it's more realistic to see it as an alternative that is providing much needed competition in the field of AI imagery. Both tools have their strengths and can coexist, offering users a choice based on their specific needs and preferences.

Last modified 19 August 2024 at 11:02

Related Posts

Visitor Comments

Please prove you are human by selecting the tree.