Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Imagen, a text-to-image diffusion model

gweb-research-imagen.appspot.com

Post reply on HN
Print view View original

441–450 of 661 posts

Re: Imagen, a text-to-image diffusion model

#441

Post by ravi-delia »

visarga wrote: post #12

Interesting discovery they made > We show that scaling the pretrained text encoder size is more important than scaling the diffusion model size. There seems to be an unexpected level of synergy between text and vision models. Can't wait to see what video and audio modalities will add to the mix.

Basically makes sense, no? DALLE-2 suffered from misunderstanding propositional logic, treating prompts as less structured then it should have. That's a text model issue! Compared to that, scaling up the image isn't as important (especially with a few passes).

Re: Imagen, a text-to-image diffusion model

#442

Post by woeirua »

axg11 wrote: post #384
Veedrac wrote: post #367

I thought I was doing well after not being overly surprised by DALL-E 2 or Gato. How am I still not calibrated on this stuff? I know I am meant to be the one who constantly argues that language models already have sophisticated semantic understanding, and that you don't need visual senses to learn grounded world knowledge of this sort, but come on, you don't get to just throw T5 in a multimodal model as-is and have i…

I firmly believe that ~20-40% of the machine learning community will say that all ML models are dumb statistical interpolators all the way until a few years after we achieve AGI. Roughly the same groups will also claim that human intelligence is special magic that cannot be recreated using current technology. I think it’s in everyone’s benefit if we start planning for a world where a significant portion of the expert…

You should be much more concerned about the prospect of nuclear war right now than the sudden emergence of an AGI.

Re: Imagen, a text-to-image diffusion model

#443

Post by ravi-delia »

thorum wrote: post #338
qclibre22 wrote: post #75

Earlier quoted context omitted.

See the paper here : https://gweb-research-imagen.appspot.com/paper.pdf Section E : "Comparison to GLIDE and DALL-E 2"

Imagen seems better at capturing details/nuance from the prompt, but subjectively the DALLE-2 images feel more “real” to me. Not sure why. Something about the lighting?

That feels about right. Imagen has a better text processing model, so it can tease apart the prompt, but DALLE has a rocking image part.

Re: Imagen, a text-to-image diffusion model

#444

Post by astrange »

tomatowurst wrote: post #427
astrange wrote: post #417

Earlier quoted context omitted.

Train it yourself. Danbooru is a publicly available explicit dataset.

This is not something you can train on a regular AWS gpu-instance without racking up millions of dollars in bills to my knowledge. Dataset isn't an issue its a capex issue.

It’s possible from scratch on not as much personally owned hardware as you’d think but will take a long time, months maybe.

Luckily, training from scratch will hopefully be obsoleted by fine-tuning - if someone else releases a generally capable model then you can turn that into another one for lower cost.

Re: Imagen, a text-to-image diffusion model

#445

Post by JayStavis »

qz_kb wrote: post #437

I have to wonder how much releasing these models will "poison the well" and fill the internet with AI generated images that make training an improved model difficult. After all if every 9/10 "oil painted" image online starts being from these generative models it'll become increasingly difficult to scrape the web and to learn from real world data in a variety of domains. Essentially once these things are widely availa…

Huh, I had never thought of that. Makes it seem like there's a small window of authenticity closing.

The irony is that if you had a great discriminator to separate the wheat from the chaff, that it would probably make its way into the next model and would no longer be useful.

My only recommendation is that OpenAI et al should be tagging metadata for all generated images as synthetic. That would be a really interesting tag for media file formats (would be much better native than metadata though) and probably useful across a lot of domains.

Re: Imagen, a text-to-image diffusion model

#446

Post by whatshisface »

qz_kb wrote: post #437

I have to wonder how much releasing these models will "poison the well" and fill the internet with AI generated images that make training an improved model difficult. After all if every 9/10 "oil painted" image online starts being from these generative models it'll become increasingly difficult to scrape the web and to learn from real world data in a variety of domains. Essentially once these things are widely availa…

I don't think it will "poison the well" so much as change it - images that humans like more will get a higher pagerank, so the models trained on Google Images will not so much as degrade as they will detach from reality and begin to follow the human mind they way plausible fiction does.

Re: Imagen, a text-to-image diffusion model

#447

Post by JohnBooty »

nomel wrote: post #130
JohnBooty wrote: post #83

Earlier quoted context omitted.

Translation: we need to hand-tune this to not reflect reality Is it reflecting reality, though? Seems to me that (as with any ML stuff, right?) it's reflecting the training corpus. Futhermore, is it this thing's job to reflect reality? the world as we (Caucasian/Asian male American woke upper-middle class San Fransisco engineers) wish it to be Snarky answer: Ah, yes, let's make sure that things like "A giant cobra sn…

> Snarky answer: Ah, yes, let's make sure that things like "A giant cobra snake on a farm. The snake is made out of corn" reflect reality. If it didn't reflect reality, you wouldn't be impressed by the image of the snake made of corn.

Pardon? The snake made of corn most certainly does not reflect reality: snakes made out of corn do not exist.

Re: Imagen, a text-to-image diffusion model

#448

Post by rhacker »

qz_kb wrote: post #437

I have to wonder how much releasing these models will "poison the well" and fill the internet with AI generated images that make training an improved model difficult. After all if every 9/10 "oil painted" image online starts being from these generative models it'll become increasingly difficult to scrape the web and to learn from real world data in a variety of domains. Essentially once these things are widely availa…

Look at carpentry blogs, recipe blogs. Nearly all of it is junk content. I bet if you combined GPT and imagen or dalle2 you could replace all of them. Just provide a betty crocker recipe and let it generate a blog that has weekly updates and even a bunch of images - "happy family enjoying pancakes together"

I can see the future as being devoid of any humanity.

Re: Imagen, a text-to-image diffusion model

#450

Post by Daishiman »

axg11 wrote: post #384
Veedrac wrote: post #367

I thought I was doing well after not being overly surprised by DALL-E 2 or Gato. How am I still not calibrated on this stuff? I know I am meant to be the one who constantly argues that language models already have sophisticated semantic understanding, and that you don't need visual senses to learn grounded world knowledge of this sort, but come on, you don't get to just throw T5 in a multimodal model as-is and have i…

I firmly believe that ~20-40% of the machine learning community will say that all ML models are dumb statistical interpolators all the way until a few years after we achieve AGI. Roughly the same groups will also claim that human intelligence is special magic that cannot be recreated using current technology. I think it’s in everyone’s benefit if we start planning for a world where a significant portion of the expert…

These ML models aren't capable of generating novel thinking. They allow for extracting knowledge from an existing network. They cannot declare new ideas, identify how to validate them, and gather data and reach conclusions.
Post reply on HN