[18:26 Thu,1.August 2024 by Thomas Richter] |
A team of well-known image AI researchers announced today the founding of the new company Black Forest Labs, specializing in generative image and video AIs, and simultaneously released a new image AI. In its simplest form, this AI is even accessible as a free and open-source model. The new FLUX.1 model includes three different variants with various properties and target audiences: FLUX.1 pro, FLUX.1 dev, and FLUX.1 fast, which are expected to deliver very high performance in terms of image quality, stylistic diversity, and scene complexity. ;)
The Black Forest Labs team includes renowned experts in the field of image AIs such as Andreas Blattmann, Andrew Holmes, Axel Sauer, Dominik Lorenz, Frederic Boesel, Patrick Esser, and Robin Rombach, who have worked on the development of fundamental generative AI models including VQGAN and Latent Diffusion, the Stable Diffusion models for image and video generation (Stable Diffusion XL, Stable Video Diffusion, Rectified Flow Transformers), as well as Adversarial Diffusion Distillation for ultra-fast image synthesis in real-time. ;) Three FLUX.1 ModelsFLUX.1 is available in three different models for various requirements, which, according to Black Forest Labs, all feature high image details, precise implementation of text instructions, stylistic diversity, and the ability to depict complex scenes. They also support a wide range of image formats and resolutions. - FLUX.1 pro is the most powerful FLUX.1 model at the cutting edge of technology. It is available for testing via Replicate and fal.ai, and commercial use can be done through the API. - FLUX.1 dev is an open model for non-commercial applications. It has been distilled directly from FLUX.1 pro but is more efficient than a standard model of the same size. The FLUX.1 dev weights are available on HuggingFace, and the model itself can be tested directly at  Replicate and Fal.ai. - FLUX.1 fast: is the fastest model for local use and personal use. FLUX.1 fast is freely available under an Apache2.0 license. The two public FLUX.1 models (dev and fast) are based on a hybrid architecture that combines multimodal and parallel diffusion transformer blocks and have a massive size of 12 billion parameters. The three FLUX.1 versions can be tested for free at Replicate, the weights for FLUX.1 dev, as well as the weights plus code for FLUX.1 fast, can be downloaded at Huggingface and Github respectively – a GPU with a lot of VRAM is required for local use. FLUX.1 - Better than Other Image AIs?Black Forest Labs claims that FLUX.1 redefines state-of-the-art image synthesis and that FLUX.1 pro and dev currently surpass leading models like Midjourney v6.0, DALL·E 3 (HD), and SD3-Ultra in areas such as visual quality, implementation of text instructions, variability of size and aspect ratio, typography, and output diversity. ;) Comparison of image AIs Next: Text-to-VideoPromising is the announcement by the "Black Forest" Labs to next develop cutting-edge text-to-video models that enable the generation and editing of high-resolution videos at great speed. ;) soon with Text-to-Video AI  more infos at bei blackforestlabs.aideutsche Version dieser Seite: Neue Bild-KI FLUX.1 "made in Germany" - besser als DALL-E 3 und Midjourney? |
 |
|
|