🌟 本期要闻:
https://alpha.midjourney.com |
新模型具有更好的提示理解能力
提示词连贯性及知识层面上实现了显著提升
支持文本绘制功能,值得注意的是,英文文本的绘制准确度和效果优于中文
两种新的放大器(subtle 和 creative 模式)
注: Midjourney v6 alpha 示例
https://sites.research.google/videopoet/?ref=aiartweekly |
谷歌最近揭示了该大语言模型生成视频的潜能。他们开发的简洁且有效的建模方法,名为 VideoPoet,能将任意自回归语言模型或 LLM 转变为一个高效的视频生成器,有能力创造出含音频的高质量视频。
GAvatar:带有隐式网格学习的动态3D高斯化身
https://nvlabs.github.io/GAvatar/?ref=aiartweekly |
https://research.nvidia.com/labs/toronto-ai/AlignYourGaussians/?ref=aiartweekly |
https://vidtome-diffusion.github.io/?ref=aiartweekly |
https://pi-animator.github.io/?ref=aiartweekly |
https://ubisoft-laforge.github.io/character/mosar/?ref=aiartweekly |
https://haar.is.tue.mpg.de/?ref=aiartweekly |
https://kim-youwang.github.io/paint-it?ref=aiartweekly |
https://peter-kocsis.github.io/IntrinsicImageDiffusion/?ref=aiartweekly |
https://szymanowiczs.github.io/splatter-image.html?ref=aiartweekly |
https://wenbin-lin.github.io/RelightableAvatar-page/?ref=aiartweekly |
https://dreamtalk-project.github.io/?ref=aiartweekly |
📰 其它动态:
pixelSplat:
实现可扩展的通用 3D 重建技术
https://pixelsplat.github.io/?ref=aiartweekly MAG-Edit:
在复杂场景中进行局部图像编辑技术
https://mag-edit.github.io/?ref=aiartweekly SCEdit:
跳过连接编辑高效且可控的图像扩散生成模型
https://scedit.github.io/?ref=aiartweekly TRepaint123:
通过渐进式可控 2D 重绘技术,实现单图快速转化为高质 3D 生成
https://junwuzhang19.github.io/repaint123/?ref=aiartweekly CrossDiff:
文本指导下的自我评估 3D 头部生成技术
https://wonderno.github.io/CrossDiff-webpage/?ref=aiartweekly HCBlur:
深度混合相机去模糊技术
http://cg.postech.ac.kr/research/HCBlur/?ref=aiartweekly
⚒️ 工具与教程:
https://github.com/cumulo-autumn/StreamDiffusion?ref=aiartweekly |
https://github.com/tzco/VolumeDiffusion?ref=aiartweekly |
https://github.com/wysoczanska/clip_dinoiser?ref=aiartweekly |
https://github.com/MengyuWang826/SegRefiner?ref=aiartweekly |