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

DALL·E now available in beta

openai.com

Post reply on HN
Print view View original

281–290 of 579 posts

Re: DALL·E now available in beta

#281

Post by astroalex »

bemmu wrote: post #58

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.

This is amazing! Honestly one of the first uses of GPT3/DALL E that has held my attention for longer than a few seconds.

Re: DALL·E now available in beta

#282

Post by napier »

harpersealtako wrote: post #267
andybak wrote: post #87

Earlier quoted context omitted.

I also got access a couple of weeks ago and I can't fathom how anyone could be underwhelmed by it. What were you expecting?

Dalle seems to only have a few "styles" of drawing that it is actually "good" at. It is particularly strong at these styles but disappointingly underwhelming at anything else, and will actively fight you and morph your prompt into one of these styles even when given an inpainting example of exactly what you want. It's great at photorealistic images like this: https://labs.openai.com/s/0MFuSC1AsZcwaafD3r0nuJTT , but i…

GPT3 has seen similar lobotomization since its initial closed beta. Current davinci outputs tend to be quite reserved and bland, whereas when I first had the fortunate opportunity to experience playing with it in mid 2020, if often felt like tapping into a friendly genius with access to unlimited pattern recognition and boundless knowledge.

Re: DALL·E now available in beta

#283

Post by ausbah »

danielvaughn wrote: post #251
nsxwolf wrote: post #234

Earlier quoted context omitted.

I'm already bored of it. When you have everything, you have nothing.

I'm sure the novelty wears off. But I'm already coming up with several applications for it. On the personal side, I've been getting into game development, but the biggest roadblock is creating concept art. I'm an artist but it takes a huge amount of time to get the ideas on paper. Using DALLE will be a massive benefit and will let me expedite that process. It's important to note that this is not replacing my entire c…

>I'm an artist but it takes a huge amount of time to get the ideas on paper.

this is what I really like about DALLE-mini, it's ability to create pretty good basic outlines for a scene. it's low resolution enough that there's room for your own creativity while giving you a good template to spring off from. things like poses, composition of multiple people, etc.

Re: DALL·E now available in beta

#285

Post by SquareWheel »

ask_b123 wrote: post #278
SquareWheel wrote: post #95

Earlier quoted context omitted.

Heads up: I think you meant "in vain" rather than "in vail". However, a similar phrase is "to no avail" which also means that something was not successful.

I think you meant "in vain" rather than "in vein".

I sure did! Thank you, I've corrected that now.

Re: DALL·E now available in beta

#286

Post by smusamashah »

xnx wrote: post #76

I fully expect stock image sites to be swamped by DALL-E generated images that match popular terms (e.g. "business person shaking hands"). Generate the image for $0.15. Sell it for $1.00.

They won't. DALL-E images are mostly not as high quality. The high quality stuff which everyone has been sharing is result of lots of cherry picking.

Re: DALL·E now available in beta

#287

Post by jowday »

Sad to say I've been dissapointed in DALLE's performance since I got access to it a couple of weeks ago - I think mainly because it was hyped up as the holy grail of text2image ever since it was first announced.

For a long while whenever Midjourney or DALLE-mini or the other models underperformed or failed to match a prompt the common refrain seemed to be "ah, but these are just the smaller version of the real impressive text2image models - surely they'd perform better on this prompt". Honestly, I don't think it performs dramatically better than DALLE-mini or Midjourney - in some cases I even think DALLE-mini outperforms it for whatever reason. Maybe because of filtering applied by OpenAI?

What difference there is seems to be a difference in quality on queries that work well, not a capability to tackle more complex queries. If you try a sentence involving lots of relationships between objects in the scene, DALLE will still generate a mishmash of those objects - it'll just look like a slightly higher quality mishmash than from DALLE-mini. And on queries that it does seem to handle well, there's almost always something off with the scene if you spend more than a moment inspecting it. I think this is why there's such a plethora of stylized and abstract imagery in the examples of DALLE's capabilities - humans are much more forgiving of flaws in those images.

I don't think artists should be afraid of being replaced by text2image models anytime soon. That said, I have gotten access to other large text2image models that claim to outperform DALLE on several metrics, and my experience matched with that claim - images were more detailed and handled relationships in the scene better than DALLE does. So there's clearly a lot of room for improvement left in the space.

Re: DALL·E now available in beta

#288

Post by educaysean »

bequanna wrote: post #279
educaysean wrote: post #275

Earlier quoted context omitted.

I just tried it out and it looks like DALL-E isn't as inept as you imagined. Exact query used was 'A profile photo of a male south korean CEO', and it spat out 4 very believable korean business dudes. Supplying the race and sex information seems to prevent new keywords from being injected. I see no problem with the system generating female CEOs when the gender information is omitted, unless you think there are?

Isn’t the diversity keyword injection random? My point is that it is pointless. If you want an image of a person included, you can just specify it yourself.

> If you want an image of a person included, you can just specify it yourself.

I agree wholeheartedly. So what are we arguing about?

What we're seeing is that DALL-E has its own bias-balancing technique it uses to nullify the imbalances it knows exists in its training data. When you specify ambiguous queries it kicks into action, but if you wanted male white CEOs the system is happy to give it to you. I'm not sure where the problem is.

Re: DALL·E now available in beta

#289

Post by MWil »

I've been on the waitlist since April 16th. Would have loved to have played around with the alpha but now clearly my ability to experiment and learn to use the system to cut down on expenses is extremely limited.

Re: DALL·E now available in beta

#290

Post by TaupeRanger »

CobrastanJorji wrote: post #100
TaupeRanger wrote: post #21

Earlier quoted context omitted.

You can't do that. I can't see this working well for children's book illustrations unless the story was specifically tailored in a way that makes continuity of style and characters irrelevant.

As an aside, Ursula Vernon did pretty well under the constraint you described. She set a comic in a dreamscape and used AI to generate most of the background imagery: https://twitter.com/UrsulaV/status/1467652391059214337 It's not the "specify the character positions in text" proposed, but still a neat take on using this sort of AI for art.

Nice example and very well done. But yeah, very niche application unfortunately.
Post reply on HN