Stable Diffusion3 (referred to as SD3) is a powerful text-to-image model with 2 billion parameters, renowned for its efficient inference speed and outstanding generation capabilities. On the evening of June 12, the renowned open-source large model platform, Stability AI, officially released the SD3-M weights, providing users with a free trial opportunity. (The trial video is placed at the top of the article!)
SD3-M Product Entry:https://top.aibase.com/tool/stable-diffusion-3-medium
SD3-M excels in average image generation time, taking only about 2 to 10 seconds, and has relatively low hardware requirements, suitable for various devices such as PCs, phones, and tablets. Additionally, SD3-M has been optimized in collaboration with NVIDIA and AMD for specific hardware, including RTX GPUs, TensorRT, the latest CPUs, and MI-300X, with performance enhancements up to 50%.
The model adopts a Multimodal Diffusion Transformer (MMDiT) architecture, significantly improving image quality, layout, complex text prompts/semantic understanding, and resource efficiency. SD3-M was pre-trained on a dataset of 1 billion images, both public and synthetic, and was fine-tuned on 30 million images for specific art styles and domains, along with 3 million preference images, making it highly effective in understanding user text prompts and embedding text into images.
Through the online demo, the generation effects of SD3-M are stunning. From towering skyscrapers to serene beach scenes, from tropical rainforests to 1950s-style diners, SD3-M demonstrates exceptional image generation capabilities. The free online trial offers users endless possibilities for exploration.
It should be noted that currently, SD3-M is only for academic research and cannot be used for commercial applications. For commercial needs, users must contact Stability AI for commercial licensing.
The open-source release of SD3-M allows users to freely trial this powerful text-to-image model and explore its potential applications across various fields. We hope this initiative will bring more possibilities and opportunities to the development of the artificial intelligence field.
Trial Address: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-medium
