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。

We asked 100 humans to draw the DALL·E prompts

138 点作者 echen大约 3 年前

19 条评论

esjeon大约 3 年前
I think it&#x27;s quite easy to distinguish b&#x2F;w DALL-E and human.<p>The biggest problem w&#x2F; current DALL-E is that it&#x27;s really really bad at details, unless it&#x27;s a stereotypical one. DALL-E delivers pretty good <i>impression</i>, but almost always fails to fill in the last mile. This is because DALL-E currently doesn&#x27;t know what &quot;perfection for human&quot; is like..<p>Also, often times, DALL-E doesn&#x27;t have a definitive style, meaning it composes images with different styles in one place. This is quite noticeable in the second image for &quot;Teddy bears, mixing sparkling chemicals as mad scientists&quot;. The bubbles are obviously not in the same style. This hardly happens for humans because, welp, when someone draws a picture, one person draws with a limited set of tools in a single style. DALL-E itself is not subject to such limitations.
评论 #31363964 未加载
评论 #31364386 未加载
评论 #31365377 未加载
评论 #31365448 未加载
评论 #31367714 未加载
评论 #31366686 未加载
评论 #31365006 未加载
评论 #31368429 未加载
CarrieLab大约 3 年前
My intuition is that humans will continue to make art that takes advantage of technological advances, just like they always have.<p>The modern process of producing music would basically be unrecognizable to anyone 40 years ago — it&#x27;s completely intertwined with technology, and far more automated. Yet music is as important as ever, and amazing music is being made (will politely side-step the pitfall of debating whether music was better 40 years ago!)<p>So I&#x27;m excited to see how visual artists incorporate tools like Dall-E into their artistic process.
评论 #31362395 未加载
评论 #31363984 未加载
评论 #31362567 未加载
cercatrova大约 3 年前
It&#x27;s relatively easy to tell which are AI made due to the weird artifacts they have that human drawings would not have.<p>For example, the first image has a basketball where the lines don&#x27;t line up correctly, a visual artifact that lets you immediately know it was an AI. Human artists would have made all the lines line up correctly.
评论 #31363434 未加载
评论 #31363622 未加载
评论 #31365075 未加载
评论 #31363371 未加载
nirui大约 3 年前
&gt; What about now?<p>Well, that picture captioned &quot;DALL·E or Dalí?&quot; is almost certainly drawn by an AI, because most humans have learned that you don&#x27;t draw weird shapes anywhere near ones crotch, it&#x27;s just something a half-decent human won&#x27;t do.<p>Joke aside, the AI generated pictures are amazing. See those radishes? Most human-drawn radishes are happy. And those generated ones are happy too, even without explicit instruction to tell the AI to do so. I guess it really captured what humans collectively want.
评论 #31363504 未加载
评论 #31364485 未加载
sriku大约 3 年前
I had an odd reaction reading the threads. People considering it worth posting that it is easy to tell which one is AI and which one is by a human .. are doing so because the Dall-e ones are damn good and not because te difference is too obvious, in which case nobody should bother.
评论 #31366667 未加载
cwillu大约 3 年前
Interesting that the human-drawn lab bears are nearly all clearly mad scientists out to conquer the world, while the AI bears are merely earnest… clowns I think?
评论 #31363388 未加载
评论 #31364353 未加载
评论 #31364533 未加载
评论 #31363149 未加载
评论 #31363623 未加载
nautilus12大约 3 年前
Subjectively I find the human versions are almost always better even the poorly drawn ones, which isnt surprising, but even though DALLE is able to emulate elements of style it still seems to be lacking some gestalt or ability to connect on an emotional level. There&#x27;s no accounting for taste it seems.
shannifin大约 3 年前
I find it fascinating and hilarious that DALL-E produces basketball-looking planets! (Or planet-looking basketballs?)
egypturnash大约 3 年前
I find myself wondering: were any of these hundred humans professional artists, was there money accompanying this request to draw these prompts, and if so how much?
评论 #31365870 未加载
userbinator大约 3 年前
<i>Imagine a future where every homemade bedtime story can be turned into an IMAX-quality cartoon…</i><p>Alternatively, imagine one where every website is flooded with mass-produced bland &quot;art&quot; like Corporate Memphis... we know what SEO spam looks like today, and I&#x27;ve been occasionally seeing what looks like AI-generated images show up in image searches. This feels like the latter --- good enough to look the part at a brief glance, but clearly &quot;off&quot; if you take a closer inspection (see comments about the lines on the basketballs.)
评论 #31367399 未加载
webmaven大约 3 年前
What I want to see next is Dall-E produced variations from all the &quot;human&quot; versions in this post. That would illuminate Dall-E&#x27;s limitations in terms of mimicking original styles.<p>P. S. Has anyone demonstrated actual style transfer with Dall-E? As in, provide an image by one artist and telling it to draw it as another artist would? Something along the lines of &quot;Vincent Van Gogh&#x27;s self portrait as drawn by Jack Kirby&#x27;.
评论 #31362737 未加载
folli大约 3 年前
For me, the DALL-E generated imagery looks better than what an average human can draw, but there is still a gap to the skill of a true artist. However, commissioning artists is expensive, DALL-E less so.<p>So there&#x27;s definitely pressure on illustrators on upwork, fiverr etc.
评论 #31364014 未加载
grumbel大约 3 年前
The two reasonably good looking human versions of baseball-astronaut are copy&amp;paste of clipart or at least used it for reference:<p><a href="https:&#x2F;&#x2F;www.freepik.com&#x2F;free-photos-vectors&#x2F;cat-astronaut" rel="nofollow">https:&#x2F;&#x2F;www.freepik.com&#x2F;free-photos-vectors&#x2F;cat-astronaut</a><p><a href="https:&#x2F;&#x2F;stock.adobe.com&#x2F;de&#x2F;images&#x2F;id&#x2F;267579488" rel="nofollow">https:&#x2F;&#x2F;stock.adobe.com&#x2F;de&#x2F;images&#x2F;id&#x2F;267579488</a>
Philomath大约 3 年前
I really wanted to give this a try and guess, but unfortunately, there&#x27;s the DALL-E watermark in the AI images, so I couldn&#x27;t even try.<p>It would have been cool to also add the watermark to the non DALL-E images and then do the disclosure of what is what.<p>For those that don&#x27;t know, the watermark is the bottom right rainbow squares.
elzbardico大约 3 年前
Glad to know that the AI has humanity enough not to use Corporate Memphis style.
rjh29大约 3 年前
The DALL-E ones are really impressive technically but I laughed at, and went &#x27;aww&#x27; at a few photos and they were always the human ones. For now.
darepublic大约 3 年前
the Dall-e pictures are good, interesting, and feel the most human from any ai produced content I&#x27;ve seen. I don&#x27;t get the same sense when I read ai generated text, even the best examples, although those results may be impressive too. Perhaps something about the visual pieces more convincingly imparts the sense of another fellow mind
montanha大约 3 年前
The easiest way to distinguish b&#x2F;w DALL-E and Humans is looking to the freakish hands generated by the previous
montanha大约 3 年前
It&#x27;s pretty easy to find