byte

According to AI Technology Review, ByteDance is secretly preparing to set up a large model research institute and actively recruiting talents. People familiar with the matter said that external AI experts have joined the large model research institute and report directly to Zhang Yiming. Since last year, ByteDance has successively announced the progress of related work on large models. According to previous reports by IT Home, in August 2023, ByteDance's self-developed underlying large model "Skylark" was launched, and then the AI dialogue product "Doubao" was launched. In May this year, ByteDance released the "Doubao Large Model" family and launched a price war, claiming to be "99.3% cheaper than the industry." …

Recently, ByteDance released a model called AnimateDiff-Lightning, which has impressive performance in video generation. It only takes 4-8 steps of reasoning to generate very good quality videos, which is undoubtedly a major technological breakthrough for the video production industry. Paper address: https://arxiv.org/html/2403.12706v1 The AnimateDiff-Lightning model also works very well with Contorlnet, which means that video conversion...

Recently, ByteDance released a new tool called ResAdapter, which can solve the problems of limb abnormalities and image collapse when Stable Diffusion (SD) generates super large images and non-training resolution images. In addition, ResAdapter is also compatible with existing IPadapter and Controlnet models. With the development of text-to-image models such as Stable Diffusion and personalization technologies such as DreamBooth and LoRA, we are now able to create...