Another genie that can no longer be put back into a bottle, for better or worse: super high quality AI image generation
The new open Flux model is very good & is free. These are results from my home computer (thanks to https://t.co/CIdN3Ril7D), each took under 10 seconds to make pic.twitter.com/FC51me9MM0
— Ethan Mollick (@emollick) August 4, 2024
The German startup Black Forest Labs, founded by researchers who developed the technology behind Stable Diffusion, has launched its first suite of text-to-image AI models called FLUX.1. The company aims to create advanced generative AI for images and videos. FLUX.1 comes in three versions: a high-end commercial “pro” version, a mid-range “dev” version with open weights for non-commercial use, and a faster open-weights “Schnell” version. The models use a “hybrid architecture” combining transformer and diffusion techniques, scaled up to 12 billion parameters.
An emerging companies & growth team led by Bay Area partner Ben Potter and Orange County partner Nima Movahedi represented the generative #AI startup in the funding round. #LathamTECH https://t.co/HgcsaD8IVK
— Latham & Watkins LLP (@lathamwatkins) August 5, 2024
In comparison tests, the outputs of the two higher-end FLUX.1 models were generally comparable with OpenAI’s offerings in prompt fidelity and photorealism. Notably, FLUX.1 is proficient at generating human hands, a common problem area for earlier image-synthesis models. Black Forest Labs recently secured a $31 million Series Seed funding round led by Andreessen Horowitz, with additional investments from General Catalyst and MätchVC.
The company has also enlisted high-profile advisers, including former Disney President and an AI researcher.
Maybe people realize maybe not, but the new gen model FLUX 1 is from the OG researchers behind stable diffusion technology. Ultimately the inventors of the latent diffusion technique.
hats off to @bfl_ml 🎓
— Linus ●ᴗ● Ekenstam (@LinusEkenstam) August 4, 2024
The company plans to expand into video generation next, announcing that FLUX.1 will serve as the foundation of a new text-to-video model currently in development.
Launch of new generative AI models
This upcoming model will compete with OpenAI, Runway, and Kuaishou offerings. Black Forest Labs stated, “Our mission 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. We believe that generative AI will be a fundamental building block of all future technologies.”
The three German researchers—Robin Rombach, Andreas Blattmann, and Dominik Lorenz—initially worked on their groundbreaking AI projects while at university.
Their affiliation with Stability AI began only after their previous work was published. One area of interest is FLUX.1’s training data. Black Forest Labs has not disclosed the specifics. Still, it is suspected to involve a large, unauthorized image scrape of the internet, a familiar yet controversial practice among generative AI companies.
This practice is the subject of ongoing legal battles, including a notable lawsuit by Getty Images against other AI image generators. FLUX.1 represents an exciting advancement in the field of generative AI, indicative of the continuous evolution and innovation driven by its creators.
