A recent buzz online has been generated by highly realistic images of people seemingly at a conference, discussing on stage. These AI-crafted photographs look so authentic that it's hard to distinguish that the individuals are not real. The software behind this new level of photorealism is FLUX.1, an open-source AI image creator launched by researchers from Black Forest Labs.
FLUX.1 presents a challenge to existing image AI models like Midjourney in its ability to depict persons and humans features with precise accuracy. The founders of Black Forest Labs are ex-engineers from Stability AI, known for their work on diffusion models. They have designed FLUX.1 using a mix of transformer and diffusion techniques in a 12 billion parameter model. Compared to Stability's models around 800 million-8 billion parameters, FLUX.1 shows an enhanced capability to render realistic imagery.
Early feedback online has praised FLUX.1's realism, especially in its depictions of hands, limbs and multi-frame videos. The makers claim it can generate high resolution images across different aspect ratios and resolutions. Black Forest Labs is now working on an open-source text-to-video AI as they strive to push the boundaries of photorealistic content creation. FLUX.1 in its free ‘Dev' and ‘Schnell' variants has integrated with development platforms like Hugging Face and AI hosting services. Its creators aim to unlock new frontiers of precision and speed in visual AI.
