At AI For Business, we’ve mostly focused on text-based generative AI tools to date. That’s for good reason – most of business is conducted in text, whether it’s reports, tables, charts, numbers, etc. But it’s also because text-based GenAI tools have so far simply been better than AI image-making tools – which have been low quality, hard to guide, or often had that obvious, shiny “AI” quality.
That is, until now.
A few months ago, Black Forest Labs released a series of models called Flux that set a new standard for AI images – not just for open-source image models, but all image models. It can make photograph-quality images with minimal artifacts. It’s highly customizable, enabling you to train it on images or styles of your choosing. Or you can just use the hundreds of customized options already made by the open-source community, completely free.
Best of all – it’s licensed for commercial use, and makes images you’d actually want to use for your business.
Let’s dig in. 👇
An Introduction to Image Generation with Flux
Over the last few years, your choices for commercial AI image generator tools have been fairly limited. A few companies have offered proprietary tools, including OpenAI, Google, Microsoft, and Meta. The results from these were hit-and-miss – style and quality was often difficult to control, and outputs often struggled with key features, like putting arms in the wrong places or making hands with extra fingers.
The open-source options weren’t much better. Over the last two years, Stable Diffusion released progressively better models that allowed anyone to fine-tune them with their desired style or image, but, again, quality results were tricky and the tools available required a high level of technical skill.
Then, in August 2024, Black Forest Labs released their Flux image generation model in three different versions. Flux Pro is the top of the line model, requiring the most resources, but also generating the best commercially-licensable results. Flux schnell (German for “fast”) focused on quick-and-dirty text-to-image generation using a smaller, faster model, with results that can also be used commercially. In between is Flux [dev], intended for non-commercial development and experimentation. All of them can make images that are a giant leap forward from everything that existed before.
Why Use Flux Image Generator Tools?
The Flux AI image generator tools stand out for three main reasons.
First, they are good. Really good. People look like people, with the right numbers of fingers and arms in the right places. Photographs look like photographs, paintings look like paintings, and everything is in it’s right place.
Second, they are good at working text into images. Previous image generation models struggled with text, resulting in jumbled, smeared characters where letters should go. However, Flux models are fairly skilled with text right out of the box, as long as the text to be shown isn’t too long (words or short phrases). Words can be worked seamlessly into objects or backgrounds with the right instruction to the model. Check out this example:
Third, they can be customized very easily. Previous image models either couldn’t be trained, were very finicky to train, or required extensive training with powerful computers to recognize specific shapes or styles. On the other hand, Flux models can be trained using 10-20 examples in a few minutes on a decent desktop computer. The examples can focus on objects, people, styles – any visual element that can be distinctly represented. It can then use these specially-trained LoRa’s (small, focused AI models that run in parallel with the main Flux model) to add the desired elements to your own images, or combine multiple LoRa’s together to mix and match. Here are some examples from a LoRa trained on Meta’s new Orion AR-enabled glasses:
Do you have a product, but are struggling with high-quality marketing photos? Train a Flux LoRa on some images of your product, and you can generate as many images as you like for a few cents apiece.
How to Get Started with Flux AI Image Generation
To make your own images with Flux, you have a few options for getting started.
The easiest way is to use hosted services like Replicate or Fal.ai. Create an account, buy a few credits via credit card, then use their web-based tools for generating images with Flux. All of the images for this post were made using Replicate – the Flux Pro model can be found here, the system for training your own custom LoRa can be found here, or you can explore LoRa’s from the community here. (By the way, these sites are also great resources for other AI-based tools, and can be both called and connected together with a little bit of code; this is a great way to explore what AI can do for your business!)
You can also run it yourself, either on your own computer (although it does best if it’s mid-to-high powered machine, like a gaming PC), or with any number of cloud-based providers. For example, Pinokio provides a one-click Windows installer for both Flux and Flux LoRa fine-tuning. You can also download the models directly from HuggingFace and set up your own system (although this path is not for the faint of heart, and requires a lot of technical know-how).
Whichever path you choose, I highly recommend just jumping in and experimenting. In particular, explore different photographs or illustrations, different painting styles, different people or objects, different levels of detail, and more. Get a feel for how the models work, then train a LoRa on your own products or pictures and explore different scenes, framing, and more. You are only limited by your imagination (and just a little by your wallet).
Become your company’s AI expert in under 30 minutes a month by signing up for the Executive Summary newsletter by AI For Business.
If you liked this post, have future posts delivered straight to your inbox by subscribing to the AI For Business newsletter. Thank you!