TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

MagicEdit: High-fidelity temporally coherent video editing

261 点作者 lnyan超过 1 年前

13 条评论

htrp超过 1 年前
Looks like its building on the same concepts as stable video.<p><a href="https:&#x2F;&#x2F;github.com&#x2F;rese1f&#x2F;StableVideo">https:&#x2F;&#x2F;github.com&#x2F;rese1f&#x2F;StableVideo</a>
capableweb超过 1 年前
Interesting enough ByteDance (TikTok&#x2F;Douyin) seems to be behind it.
评论 #37313011 未加载
评论 #37313446 未加载
shantnutiwari超过 1 年前
When I click on the code link, there is no code, just the same images and an assets folder with 1 gif file.<p>Why is this even on HN? There are zero details of how its built, or what the source looks like. Am I missing something?
评论 #37322769 未加载
xnx超过 1 年前
This looks like a nice improvement on current video to video techniques: <a href="https:&#x2F;&#x2F;stable-diffusion-art.com&#x2F;video-to-video&#x2F;" rel="nofollow noreferrer">https:&#x2F;&#x2F;stable-diffusion-art.com&#x2F;video-to-video&#x2F;</a>
pindab0ter超过 1 年前
Site doesn&#x27;t work correctly for me on Edge. Every row has at least one disappear after a fraction of a second when loading the page.
satvikpendem超过 1 年前
Imagine in 5-10 years where, just like people make video games all by themselves or with a small group of people, people can make their own movies that rival Hollywood productions, for a fraction of the cost as there&#x27;s no need to hire anyone. When the output is just pixels on a screen and you can manipulate the placement of every pixel, it&#x27;s really no different to Hollywood making them with real actors and crew members or someone drawing them, as is the case with current animation methods. Now, we can do the same process as drawing each frame but then make it look photorealistic.
评论 #37311294 未加载
评论 #37310986 未加载
评论 #37315877 未加载
评论 #37310855 未加载
评论 #37316821 未加载
评论 #37313715 未加载
评论 #37311341 未加载
评论 #37311838 未加载
评论 #37315451 未加载
评论 #37317767 未加载
评论 #37310925 未加载
Timon3超过 1 年前
It&#x27;s crazy how consistent the Shutterstock logo is in the outpainted examples.
评论 #37311518 未加载
评论 #37311512 未加载
cdchn超过 1 年前
All of the source videos that have people&#x27;s faces seem to be blurred. Are they blurred before transforming or do they blur it just to post the samples on their web page?
评论 #37310513 未加载
评论 #37310665 未加载
dylan604超过 1 年前
what would happen if they didn&#x27;t say &quot;pretty girl&quot;?
评论 #37311621 未加载
评论 #37310949 未加载
justinator超过 1 年前
<i>&quot;a pretty girl, white singlet, dark pants, on the stage&quot;</i><p>Well, glad they kept the 6 fingers on her right hand.
评论 #37315631 未加载
thatguy27超过 1 年前
No code...
评论 #37312251 未加载
评论 #37310427 未加载
jdprgm超过 1 年前
I have been reading a lot of adjacent papers to this recently. Here is a useful collection for anyone interested: <a href="https:&#x2F;&#x2F;github.com&#x2F;zengyh1900&#x2F;Awesome-Image-Inpainting">https:&#x2F;&#x2F;github.com&#x2F;zengyh1900&#x2F;Awesome-Image-Inpainting</a><p>I have noticed an overwhelming trend that the vast majority of authors tend to be Chinese sounding names even when associated with an US based university. Obviously some of those could be Americans as well but it stood out to such a degree I was curious if anyone had any insight.<p>Also if there is code (this projects links to github but it&#x27;s empty), it tends to be basically abandonware once these papers are published with no effort towards commercializing or turning into healthy open source projects for some reason.
评论 #37320553 未加载
brink超过 1 年前
Kendrick Lamar: &quot;I&#x27;m so sick and tired of the Photoshop&quot;<p>Who actually wants to look at this? It&#x27;s a neat trick, but I greatly prefer to look at what&#x27;s real, and I imagine most people outside of this AI hype bubble do too. It&#x27;s implied that the stable diffusion here is making the video better, but by most definitions of the word &quot;better&quot;, it&#x27;s not.
评论 #37317642 未加载