DALL·E now available in beta • Hacker News Board

Re: DALL·E now available in beta

#51

Post by bergenty » Wed, Jul 20, 2022, 5:08 PM UTC

jawns wrote: post #14

One of the commercial use cases this post mentions is authors who want to add illustrations to children's stories. I wonder if there is a way for DALL-E to generate a character, then persist that character over subsequent runs. Otherwise, it would be pretty difficult to generate illustrations that depict a coherent story. Example ... Image 1 prompt: A character named Boop, a green alien with three arms, climbs out of…

You cannot. But a workaround would be to say something like “generate an alien in three different poses— running, walking, waving”

Then use inpainting to only preserve that pose and generate new content around it. It’s definitely not perfect.

Re: DALL·E now available in beta

#52

Post by nharada » Wed, Jul 20, 2022, 5:09 PM UTC

zitterbewegung wrote: post #36

nharada wrote: post #28

Something I haven’t seen anyone talking about with these huge models: how do future models get trained when more content online is model generated to start with? Presumably you don’t wanna train a model on autogenerated images or text, but you can’t necessarily know which is which.

This should be a step in cleaning your data to begin with. If you don't know the providence of your data then you shouldn't be even training with it. Getting humans to refine your data is the best solution right now and many companies and researches go with this approach.

But how would you know? A random string of text or an image with the watermark removed is going to be very hard to distinguish generated from human written.

Re: DALL·E now available in beta

#53

Post by scifibestfi » Wed, Jul 20, 2022, 5:09 PM UTC

ccmcarey wrote: post #6

That's about 10x as expensive as it should be

What are you basing that on? What should the price be? The training and generation are probably expensive.

Re: DALL·E now available in beta

#54

Post by WalterSear » Wed, Jul 20, 2022, 5:10 PM UTC

rvz wrote: post #29

brycethornton wrote: post #3

I'm blown away by this: "Starting today, users get full usage rights to commercialize the images they create with DALL·E, including the right to reprint, sell, and merchandise. This includes images they generated during the research preview." I assumed this was going to be the sticking point for wider usage for a long time. They're now saying that you have full rights to sell Dall-E 2 creations?

It also means there will possibly be another renaissance of fully automated, mass generated NFTs and tons of derivatives and remixes flooding the NFT market in an attempt to pump the NFT hype again. It doesn't matter, OpenAI wins anyway as these companies will pour hundreds of thousands into generated images. It seems that the NFT grift is about to be rebooted again, such that it isn't going to die that quickly. But…

NFTs were never limited by artwork availability - they are limited by wash-trading ability.

Re: DALL·E now available in beta

#55

Post by nutanc » Wed, Jul 20, 2022, 5:10 PM UTC

pqdbr wrote: post #45

nutanc wrote: post #40

Earlier quoted context omitted.

And I just used it to create cover art for a book published in Amazon :) https://twitter.com/nutanc/status/1549798460290764801?s=20&t...

What was your prompt?

"girl with a cap standing next to a shadow man having a speech bubble, digital art"

Re: DALL·E now available in beta

#56

Post by FrenchDevRemote » Wed, Jul 20, 2022, 5:11 PM UTC

zitterbewegung wrote: post #36

nharada wrote: post #28

Something I haven’t seen anyone talking about with these huge models: how do future models get trained when more content online is model generated to start with? Presumably you don’t wanna train a model on autogenerated images or text, but you can’t necessarily know which is which.

This should be a step in cleaning your data to begin with. If you don't know the providence of your data then you shouldn't be even training with it. Getting humans to refine your data is the best solution right now and many companies and researches go with this approach.

You can't use humans to manually refine a dataset on the scale of GPT-3 or DALL-E

Clip was trained on 400,000,000 images, GPT is roughly 180B tokens, at 1-2 tokens per word, that's 120,000,000,000 words.

Re: DALL·E now available in beta

#57

Post by scifibestfi » Wed, Jul 20, 2022, 5:12 PM UTC

hhmc wrote: post #4

bulbosaur123 wrote: post #2

Quoted post unavailable.

So you actually _wanted_ images that perpetuate the biases of the world?

How do you remove bias as long as humans are in the loop? Aren't they just swapping one bias for their own?

Re: DALL·E now available in beta

#58

Post by bemmu » Wed, Jul 20, 2022, 5:12 PM UTC

I was supposed to be making a video game, but got a bit sidetracked when DALL·E came out and made this website on the side: http://dailywrong.com/ (yes I should get SSL).

It's like The Onion, but all the articles are made with GPT-3 and DALL·E. I start with an interesting DALL·E image, then describe it to GPT-3 and ask it for an Onion-like article on the topic. The results are surprisingly good.

Re: DALL·E now available in beta

#59

Post by lagrange77 » Wed, Jul 20, 2022, 5:12 PM UTC

Has anyone else had problems with the 'Generate Variations' functions lately? Tried it out first 3 days ago, and it says 'Something went wrong. Please try again later, or contact support@openai.com if this is an ongoing problem.' everytime since then.

Re: DALL·E now available in beta

#60

Post by throwaway0x7E6 » Wed, Jul 20, 2022, 5:12 PM UTC

dharbin wrote: post #32

I find it amusing that they suggest DALL-E, which typically generates lovecraftian nightmare images, for making children's story illustrations.

yeah. dalle is "so bad it's good".

it's great for post-post-ironic memes, but I don't see it being useful for anything else