Midjourney

Midjourney V6和Niji6型号新添角色参照功能，用户可调控面部、着装等，实现跨风格图像生成和换装玩法。该功能目前适用于alpha测试阶段的V6版本，未来可能会有变化。使用时，以自创图像为参考可能效果最佳，真人照片可能导致扭曲。简单操作后，用户能保留角色特征并调整场景，反馈积极，部分网友已开始测试。然而，Stability AI员工因尝试抓取提示图片数据导致系统问题的事件被禁止使用Midjourney，这也引发了对AI版权保护的讨论。

Title: "Revolutionizing Text Rendering: Stable Diffusion 3 and its Video Counterpart"Summary: The article delves into the groundbreaking release of Stable Diffusion 3, a text rendering system surpassing expectations with exceptional capabilities. It closely aligns with Sora's architecture, incorporating the diffusion transformer for seamless text generation. The...

Midjourney, a company with annual revenue of $10 million, operates with an exceptional team consisting of only 11 employees, four of whom are pursuing undergraduate studies. The company is notable for its founder's contribution to the AI community through open sourcing Caffe, a pioneering AI framework.

上周四，备受期待的Midjourney V6终于进行了内部测试更新，其显著改进了数据源和算法，生成更逼真、精细的图像，特别是在材质优化和理解用户模糊提示上有了重大突破。V6版本在处理长文本和复杂指令时展现出色表现，尽管与OpenAI的Dalle-3相比略逊一筹，但对中文输入的支持和无需冗余TAG的改进让用户体验显著提升。Midjourney仍面临AI训练版权合规及艺术作品合法性的讨论，不过随着正式版的到来，期待更多功能优化和惊喜。