TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

DreamFusion: Text-to-3D using 2D Diffusion

833 pointsby going_hamover 2 years ago

42 comments

modelessover 2 years ago
The most incredible thing here is that this demonstrates a level of 3D understanding that I didn&#x27;t believe existed in 2D image models yet. All of the 3D information in the output was inferred from the training set, which is exclusively uncurated and unsorted 2D still images. No 3D models, no camera parameters, no depth maps. No information about picture content other than a text label (scraped from the web and often incorrect!).<p>From a pile of random undifferentiated images the model has learned the detailed 3D structure and plausible poses and variants of thousands (millions?) of everyday objects. And all we needed to get that 3D information out of the model was the right sampling procedure.
评论 #33028329 未加载
评论 #33028103 未加载
评论 #33027392 未加载
评论 #33041415 未加载
jianshenover 2 years ago
Did we hit some sort of technical inflection point in the last couple of weeks or is this just coincidence that all of these ML papers around high quality procedural generation are just dropping every other day?
评论 #33026400 未加载
评论 #33026506 未加载
评论 #33028038 未加载
评论 #33027541 未加载
评论 #33028896 未加载
评论 #33026758 未加载
评论 #33028017 未加载
评论 #33026574 未加载
评论 #33029019 未加载
评论 #33028353 未加载
评论 #33027102 未加载
评论 #33026879 未加载
评论 #33026537 未加载
评论 #33026367 未加载
评论 #33026773 未加载
评论 #33026312 未加载
birracervezaover 2 years ago
In the make-a-video I said that things are getting more and more impressive by the day. I was wrong, because that was a couple hours ago. They&#x27;re getting more and more impressive by the HOUR.<p>I&#x27;m curious where this will end up in a year. Will it plateau? If so, when?
etaioinshrdluover 2 years ago
Huh, it&#x27;s a pretty similar technique to what I outlined a couple days ago: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=32965139" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=32965139</a><p>Although they start with random initialization and a text prompt. It seems to work well. I now see no reason we can&#x27;t start with image initialization!
评论 #33030092 未加载
评论 #33029773 未加载
评论 #33026552 未加载
评论 #33026797 未加载
drKarlover 2 years ago
Amazing! How long then until we get photorealistic AI generated 3D VR games and experiences in the metaverse?
评论 #33026633 未加载
nailloover 2 years ago
It&#x27;s funny that the authors are &#x27;anonymous&#x27; but they have access to Imagen so obviously it&#x27;s by Google.
评论 #33027620 未加载
评论 #33026504 未加载
评论 #33026909 未加载
评论 #33027642 未加载
评论 #33027483 未加载
评论 #33026487 未加载
jonas21over 2 years ago
Can someone explain what&#x27;s going on in this example from the gallery? The prompt is &quot;a humanoid robot using a rolling pin to roll out dough&quot;:<p><a href="https:&#x2F;&#x2F;dreamfusion-cdn.ajayj.com&#x2F;gallery_sept28&#x2F;crf20&#x2F;a_DSLR_photo_of_a_humanoid_robot_using_a_rolling_pin_to_roll_out_dough.mp4" rel="nofollow">https:&#x2F;&#x2F;dreamfusion-cdn.ajayj.com&#x2F;gallery_sept28&#x2F;crf20&#x2F;a_DSL...</a><p>But if you look closely, the pin looks like it&#x27;s actually rolling across the dough as the camera orbits.
评论 #33027114 未加载
评论 #33030885 未加载
parasjover 2 years ago
Correct link with full demo: <a href="https:&#x2F;&#x2F;dreamfusion3d.github.io&#x2F;" rel="nofollow">https:&#x2F;&#x2F;dreamfusion3d.github.io&#x2F;</a>
评论 #33026347 未加载
评论 #33028411 未加载
samstaveover 2 years ago
As someone who went to college for 3D animation in +*<i>1997*</i>+ AND DESIGNED the datacenter for luca&#x27; presidio complex..<p>where-by learning that Pixar was developed by steve jobs when lucas didnt think there was a future for computer animation... and so steve bought the death star from lucas...<p>That became pixar...<p>AI is going to fucking kill it - what will happen in the next decade will be ANYONE uploading a script to an AI to make a full length movie...<p>AND their will be editing tools as well that are AI driven...<p>Like mentioned by William Gibson<p>*The future is here, its just not evenly distributed yet*
评论 #33032292 未加载
评论 #33031147 未加载
评论 #33038935 未加载
VikingCoderover 2 years ago
We&#x27;re quickly approaching HNFusion: Text-to-HN-Article-That-Implements-That-Idea ...
macrolimeover 2 years ago
This sounds like something that could be made to work with stable diffusion if someone just implements the code based on the paper.
评论 #33026821 未加载
joewhatkinsover 2 years ago
This is crazy good - most prior text-to-3d models produced weird amorphous blobs that would kind of look like the prompt from some angles, but had no actual spatial consistency.<p>Blown away by how quickly this stuff is advancing, even as someone who&#x27;s relatively cynical about AI art.
ml_basicsover 2 years ago
Awesome! I wonder how long it will be until there is an open source implementation compatible with Stable Diffusion
MitPittover 2 years ago
Coincidentally came out the same day as Meta&#x27;s text-to-video. I wonder if Google deliberately held out the release to make a bigger impact somehow?
评论 #33026327 未加载
评论 #33026031 未加载
评论 #33025857 未加载
bmpooleover 2 years ago
hi folks, ben p from the dreamfusion paper here. happy to answer qs for the next ~hour!
评论 #33032332 未加载
评论 #33032377 未加载
评论 #33030537 未加载
deltasevennineover 2 years ago
What does this mean for our understanding of intelligence?<p>It trivializes it, in my opinion.<p>When asked the question of is lambda&#x2F;GPT-3 and&#x2F;or DreamFusion and it&#x27;s derivatives an aspect of sentience? there&#x27;s always a bunch of people who are repeating the same cliche negative line, of &quot;no, it&#x27;s only attempting to statistically mimic sentience.&quot; I agree with the reasoning.<p>But have we considered the other side of the story? That yes, the mimicry is All sentience actually is. Nothing more.
评论 #33036066 未加载
achr2over 2 years ago
The thing that frightens me is that we are rapidly reaching broad humanity disrupting ML technologies without any of the social or societal frameworks to cope with it.
评论 #33027024 未加载
评论 #33026999 未加载
parasjover 2 years ago
@dang The link should be updated to <a href="https:&#x2F;&#x2F;dreamfusion3d.github.io" rel="nofollow">https:&#x2F;&#x2F;dreamfusion3d.github.io</a>
sirianthover 2 years ago
Is there code for any of these models? Or a collab? Ajay Jain&#x27;s colab doesn&#x27;t work, but I would love to see a colab for this.
评论 #33026281 未加载
评论 #33040365 未加载
评论 #33028416 未加载
LarsDu88over 2 years ago
As someone who dabbles in 3d modeling, this is going to be an incredible resource for creating static 3d objects. Someone ought to come up with a way to convert to mesh better than the Marching Cubes algorithm I&#x27;ve seen applied to most NERFs. The models still lack coherent topology and would probably be janky if fully rigged.
评论 #33031635 未加载
评论 #33040232 未加载
macawfishover 2 years ago
So does this mean I can use DreamBooth to create plausible NERFs of myself in any scenario? The future is looking weird.
评论 #33029143 未加载
MrLeapover 2 years ago
Fun that they had an octopus playing a piano.<p>I made the same thing the old fashioned way. Mine can actually play though. <a href="https:&#x2F;&#x2F;twitter.com&#x2F;LeapJosh&#x2F;status&#x2F;1423052486760411136" rel="nofollow">https:&#x2F;&#x2F;twitter.com&#x2F;LeapJosh&#x2F;status&#x2F;1423052486760411136</a> :P
评论 #33031283 未加载
yargover 2 years ago
Cool.<p>The samples are lacking definition, but they&#x27;re otherwise spatially stable across perspectives.<p>That&#x27;s something that&#x27;s been struggled with for years.
RosanaAnaDanaover 2 years ago
This is getting asymptotic.
评论 #33026301 未加载
WheelsAtLargeover 2 years ago
I&#x27;m not even going to pretend that I have a clue on how this is done. But I&#x27;m wondering if the output can be turned into 3d objects that can be used in any of the 3D modeling software? It would be a game changer in terms of real world product development in both of speed and ease.
评论 #33028732 未加载
spaceman_2020over 2 years ago
These are getting too good, too fast.<p>I&#x27;m excited and scared. The world is going to look very different in 10 years!
visargaover 2 years ago
Futurists have been predicting when we&#x27;ll have stable fusion for decades, but now we suddenly got stable diffusion working. That&#x27;s good too, not what we wanted, but good. We&#x27;re gonna need stable fusion or other renewables to run stable diffusion though. &#x2F;s
kennyloginzover 2 years ago
Pretty neat, wish I could try it out ( maybe I missed a link). Obviously has interesting &#x2F; novel uses, but kind of reminds me of the previous discussion of upscaling audio recordings to the “soundstage” format. I doubt most 2d images want to be 3d ;)
O__________Oover 2 years ago
Unclear to me what is going on, but there’s another URL that lists the authors names. Given it’s possible this change was done for reason, not linking to it, but strikes me as odd it’s still up. Anyone know what’s going on without causing problems for the authors?
评论 #33026521 未加载
coolcaover 2 years ago
This is like magic to me. The pace at which we are getting these tool amazes me.
keepquestioningover 2 years ago
Oh my god, we are done for
samuellover 2 years ago
Gives a new perspective on a classic verse:<p>&quot;For he spoke, and it came to be;<p>he commanded, and it stood firm.&quot;<p>Psalm 33:9, NIV<p>:)
评论 #33028963 未加载
评论 #33027641 未加载
EZ-Cheezeover 2 years ago
btw guys stable diffusion img2img consistently applied frame-by-frame will get us some insane CGI for movies yo<p>&quot;transform this into this realistically&quot;<p>ILM&#x27;s holy grail
arisAlexisover 2 years ago
Is it a light version of script when the AGI comes fast
gershover 2 years ago
Is code available?
airbreatherover 2 years ago
but seems i can only generate models from predetermined inputs, when can i submit my inputs to create a video?
tonis2over 2 years ago
Is there an API for using it myself ?
edgartaorover 2 years ago
I don&#x27;t see a person in the gallery. It&#x27;s capable of generate a 3D model of me with only a photo?
dustedover 2 years ago
that is so amazing! Next up it puts a skeleton in them and animate :o
golemotronover 2 years ago
Anonymously authored research is very ominous.
评论 #33026512 未加载
owenpalmerover 2 years ago
Source?
dangover 2 years ago
Url changed from <a href="https:&#x2F;&#x2F;dreamfusionpaper.github.io&#x2F;" rel="nofollow">https:&#x2F;&#x2F;dreamfusionpaper.github.io&#x2F;</a> to the page that names the authors.