Adding model provenance would be a good idea so you can support multiple. I was wondering if DALLE 2 embeds any metadata in the images but haven't had a chance to investigate.
I also think you should consider adding the option to clarify copyright status ... in particular, allow users to dedicate the image to the public domain via CC0 (https://creativecommons.org/share-your-work/public-domain/cc...), or maybe put it under a CC BY license. May be debatable what the legal status of the images is, but a CC0 dedication would at least get rid of any ambiguity.
I'm not sure if Dalle-2 embeds any metadata, but openAI also has this CLIP model https://openai.com/blog/clip/ that lets you get embeddings. So we will be using this for our search functionality. It works amazingly well. Stay tuned for the search functionality. It is super powerful for prompt engineering.
Thanks for pointing out the copyright status. Most of the images in Discover are crawled on the web, and you can see all the sources under each image. However for the images that people post on our platform, we should definitely add the clarification for copyright status.
> I'm not sure if Dalle-2 embeds any metadata, but openAI also has this CLIP model https://openai.com/blog/clip/ that lets you get embeddings.
Hah! Same word, two totally different meaning.
Metadata embedding: store extra information in non-visible fields in the file. For example some phones save the gps location where the picture was made in the image.
CLIP embedding: a vector which describes the semantic content of an image. It is the output of an AI model which maps the image into a semantic space (aka embeds the image in said space.) This vectorial description then typically can be used for image content classification or textual label generation.
For more context about my 'metadata' comment, there is the PNG metadata (should be easy to check) but there is also the watermark in the bottom right. Here is a reddit thread (one of many) investigating the watermark, seems to conclude its innocuous: https://www.reddit.com/r/dalle2/comments/uaq7bf/discovery_da...
I’ve used both DALL-E and stable diffusion extensively, and in my experience DALL-E is more forgiving while stable diffusion is more powerful — with a much higher skill ceiling for writing prompts.
Also, stable diffusion’s open source weights are great for self-hosting which can’t be done with DALL-E.
A great summary. I do feel different prompt engines require quite different prompts to get good results. We will add stable diffusion and midjourney on openart.ai very soon too :)
Is there someone in the illustrator community who is not excited about this meaningful technology? So cool.
You don't have a job anymore. As a start, you will get less money.
Why I have to pay you when GPT 3 can generate prompts for DALL-E?
Send your congratulations to the geniuses which in search of validation and budget optimization are bringing this revolution ahead.
This thing proved for me that the UI design and design in general are the next target.
I have transitioned to pure frontend development, and I am cool for the next three to six years. After this who knows.
On the artistic end. No more digital painting for me.
I will use only analogue mediums, may be someone will pay for a human made picture on the wall....who knows.
Honestly it's not that much different than sampling in music. A few years ago Kayne West tried something similar when instead of paying Aphex Twin for the sample usage of "Avril 14th", he asked John Legend to play basically a carbon copy of the track on a piano[1]. Ultimately he had to pay for the track anyway.
Same here. Did you use my artworks while training your AI model? Cool, pay me licensing rights and residuals. No? Then remove them from your fancy art obfuscator filter.
But how is it different from human artists creating new artworks based on their studies of thousands of previous works, I wonder? No artists have paid royalties just because they went to an exhibition and had a glimpse of other painter’s work, no?
This is basically Art School 101, and the answer can be summed up as:
Another painter would go for contextual and topical inference, specifically trying to understand why some techniques and painting styles garner a specific response. The AI goes for topological inference - there are no other informations beyond the painting data itself. It observes that some artists keep reusing the same colors and proportions and builds over the data by using very complicated parametric analysis. It's still mechanical reproduction, just extremely more complicated than in the past.
And there's nothing intrinsically wrong with this, mind you - going back to my previous musical comparison, Daft Punk built a career with a lo-fi version of this idea, but they never tried to pretend that Harder Better Faster Stronger isn't anything but an extremely, extremely clever flip of Edwin Birdsong's "Cola Bottle Baby". And they paid both licensing and royalties for it.
There's this article "I Went Viral in the Bad Way" from a write from The Atlantic - https://newsletters.theatlantic.com/galaxy-brain/62fc502abcb.... He got criticized so hard for using an AI-generated in his newsletter instead of paying an artist to do the work. People's concerns are valid and real.
Yet, we can also imagine a future that "lower-level" artistic work can be achieved by AI, like blog article thumbnails, while higher-level artistic work will be done by artists, assisted by AI. Artists can get inspiration from AI artwork and reduce some tedious parts of their workflow.
> ...while higher-level artistic work will be done by artists, assisted by AI. Artists can get inspiration from AI artwork and reduce some tedious parts of their workflow.
I'm having a hard time seeing how increasingly better and freely available AI artwork doesn't just completely replace artists (and more generally how increasing AI capabilities don't just replace any career that does not involve some degree of skilled manual labor, at least until AIs get robust manual capabilities). In particular "higher-level artistic work" would seem to be that work which rises beyond what most artists do today. Prompt engineering is a far cry from making a painting oneself and there are likely many people who enjoy the latter who do not enjoy the former.
> I'm having a hard time seeing how increasingly better and freely available AI artwork doesn't just completely replace artists
I agree that AI engines like DALL-E seem like a threat to many commercial artists, especially those whose name and personality are not closely associated with their works. Magazine illustrators are one example.
For some artists, though, the artist’s status as a human being with a unique identity is a big part of their appeal. That’s one reason why there’s such a huge price difference between, say, a painting believed to be by Van Gogh and a high-quality painting in Van Gogh’s style that is believed to be a forgery.
My daughter is a successful commercial artist, and I have been following the AI advances partly out of concern for her future. As it happens, she has been able to associate her name and personal identity even with her commercial works, and she is skilled at working closely with her clients to produce the kinds of illustrations they want. For the time being, at least, her business seems safe. I wouldn’t want to make a prediction about the long term, though.
I just tried several DALL-E prompts with “in the style of [my daughter’s name],” but the results looked nothing like her work. Maybe OpenAI hasn’t crawled her website [1] yet.
Also, her commercial work usually includes lettering in English (and sometimes in Japanese), and DALL-E is not able to produce words or sentences yet.
> Also, her commercial work usually includes lettering in English (and sometimes in Japanese), and DALL-E is not able to produce words or sentences yet.
This is purely an issue of scaling and is no longer true. Other text-to-image AI have already gotten around that and produce accurate text. See e.g. https://imagen.research.google/
I can imagine a future that the artist could "license" Dalle-2 to generate art in their style and they actually get some commission for it. Like the AI can work as the artist's assistant to produce art at scale while she gets benefits too if it makes sense.
If I judge by the examples posted on OpenArt, Dall-e is currently perfectly happy to produce art "by [artist name]". What would they gain by starting to pay commissions? I doubt that will happen unless some legal action forces them to do it.
Even in that case, how would you decide which artists get the commission and which do not? They can't possibly track down authors of millions of artworks across the globe they scrapped when training the model. It seems to me that paying commissions to anyone would just mean that they acknowledge the questionable copyright status of the artworks produced by Dall-e.
Sorry, whichever way you flip this, Inspiration is used again and again to hide the real intent of this technology: To remove artistic contribution from the corporate expenses list. Period.
And this is just the beginning. Illustrators are the low-hanging fruit, they have no unions to protect them, they were undervalued for ages even before the digital age.
Human made digital art will be competing with synthesis and generative processes for minimum pay, the regular Joe will have no doubt that the AI "Art" is better.
The main problem with this tech is that I have no clear answer of the question: Who benefits from this "advancement"? Artists? Or corporations?
To become a good artist, I have spent all my life practicing from anatomy through composition and color theory. Can you imagine the stupidity of my choice? Any schmuck with a computer and prompt collection will be a master of Arts soon.
The war between Hollywood and Silicon Valley has reached the new high.
The "geniuses" which scraped all the stock websites and created this monstrosity, in my view, must be named "The Spirit Crushers" from now on.
But hey, it is too late. The genie is out of the bottle. So who is next?
I stopped my successful art career years ago over different reasons.
But now I see my choice in a different light.
> they were undervalued for ages even before the digital age
Why are they? Why are teachers, nurses, policemen, etcetera undervalued?
I think this is because these jobs don't scale. You won't be able to show progression through time where you churn out 1, 2, 4, 8, 16, ... artworks/week. Nor can a schoolteacher aspire to teach 30, 100, 300, 1000 kids a year.
Our economic system is addicted to growth. S&P 500 must increase 4% a year or we're all doomed.
The ridiculousness of this in a post-peak-everything world is glaringly obvious.
I'm sorry you chose a form of human expression that is incompatible with infinite exponential economic growth.
if you are under the impression that working artists were suffering from any shortage of inspiration before sam altman decided we shouldn't have jobs anymore then you are sorely mistaken.
When Sars2 come two and a half years ago (the end of 2019), I rearranged my whole life, restructured my company, moved to rural area. The whole world is changed, socially, economically and mentally. The same people who accused me of overreacting now have several vaccinations and track themselves regularly.
>I agree with Elon Musk that when we build A.I. without a kill switch, we are “summoning the demon” and that humans could end up, as Steve Wozniak said, as the family pets. (If we’re lucky.)
Since when "speeding up" is a marker of quality?
I see someone writing prompts, not drawing or painting at all:)
So exciting. Right?
Hey Michelangelo, you are deadly slow, we will give you this speedup brush to meet the Quarterly Report requirements from the Pope. If you don't comply, the local bard will describe with words better than you.
From the comment section of your post:
>
MAH1RO
2 weeks ago
To be honest: This is the most DEPRESSING technology for my personal life.
Yes, sure. AI will never be as 100% perfect as the absolute top-artists. But even just the 90% that the AI can produce are enough for most companies or whatever. So they can get cheap art and many "real" artists are going down - or are FORCED to use the AI too and just do the little polishing in the end to make a living. To me thats not what art is. I want to create art. Not let a program do 90% for me and then just optimize it a little.
And the other thing is: This programs use MILLIONS of artworks without consent. So everyone can nearly copy your art with an AI like this with just a few alterations.
Yes, humans will always create art. But stuff like this lowers the value of "real" art and the time you invested in it so much, it breaks my heart...
There are tons of artists who value speed even without managers or deadlines. I can see animators making great use of AI to:
1. Quickly create storyboards that convey the mood and style of the scenes
2. Sketch out rough ideas for characters, environments
3. Generate the in-between frames of animations
AI will be used as a tool, it's the artists job to judge the output, guide it, and tweak it as necessary. Just like a movie director. Except unlike directors who need an entire team of workers, AI will empower independent artists who will now be able to create entire animations on their own.
You are an artist, I assume, or not?
Why you don't invent AI financial revolution to remove the banks from the financial system and just say to them: I agree that AI can help banks with scaling up.
The funny part is that this already happens, your carbon-social index is waiting for you to be judged by AI. Not by human. And the banks will still be here.:)
This is such a silly argument. People have been drawing and photoshopping non real depictions of lion elephant hybrids or whatever else for ages. It's already "clogged".
this is great. it's really helpful to see what impact certain keywords and phrases have on the output.
would it be possible for the site to allow toggling between Dalle-2 and the recent Stable Diffusion model release?
edit: I mean, letting you filter between images from Dalle-2 and Stable Diffusion, not actually running the same prompt through the models, which this site obviously does not do
given that the latter has weights available for download, and apparently fits in a consumer GPU, there might soon be more demand for that than for Dalle-2...
This is such a brilliant idea! Most of the images on https://openart.ai/ are from Dalle-2, but we will definitely add Stable Diffusion images thanks to your advice. Thanks!
So much comes from the structure of the prompt. I've played a lot with dalle-mini, because that's all I had access to, and it's interesting to take prompts from this page (OpenArt) and try them over in dalle-mini.
For example, just paste this (From an OpenArt example):
"A young girl with long wavy black hair is riding a motocycle, she is riding through sunset, she has long black boots. Perspective is from back. Pixel art."
The results, in dalle-mini, are so much better than anything I've ever gotten out of it.
And, then, try variation on the theme:
"A young boy with short blonde hair is playing a stratocaster, he is playing in front of a garden stream, he is wearing blue jeans. Perspective is from front. Watercolor."
The expression "OpenArt" sounds like a good fit for true art, as opposed to a form of synthetic visual production of visual arts mimicking art. In a way is hijacking the notion of arts out of humanity.
Fair proposition, but then what's the place of humans there? How aren't they deprived of participation in creation and forced to stay as mere consumers in there?
Precisely, that's why we need a bit of help from philosophers and people that has deep historical knowledge of Art.
From my naivety, if art started as a form of human expression, purely synthetically produced expressions are the antithesis of art. We can discuss what that is instead.
I think many people who aren't artists lack the skills to express their creativity and imagination, in a visual way like the artists do, and that's how this kind of AI tool can help with.
But you're talking about high-cultural art thresholds, which is a valid topic but not the core of the concept.
Any kid is an artist in a way, every caveman that painted its cave was an artist, you seducing your wife/girlfriend/significant other are doing art. Some silly memes are or aren't?
That's why I repeat. We need a remix of technologists and philosopher's raising this discussion up.
I wonder how it interprets something like “two lesbians”. There isn’t necessarily anything a picture of might show to indicate that two women are lesbians, unless something intimate is happening. So, I wonder how the program interprets that. Does it just replace lesbian with woman, or is there actually something to indicate that a woman is a lesbian that we might not notice (gaydar?) but a program that has looked at billions of photos has noticed? Does it know us better than we know ourselves?
Beautiful examples... but does anyone else dislike infinite scrolling websites? It feels like a dark pattern intended to maximize time spent on the site.
I very much doubt this. One of the clues is the reproduction of artifacts that have nothing to do with the original prompt - for example the Getty images logo.
In this case, I think that the hand and its shadow are both part of a single original artwork. If they weren't, there would be two other explanations I could see: one, that the AI understands lighting and 3d somewhat, and applied the shadow accordingly, or two, that there are many such images that have a hand a shadow in a similar place, and therefore the AI has it as an association of sorts. I find both of these explanations less likely than the original theory, which would be that the hand and the shadow are part of a single original work that was, even if not copied verbatim, then used as unmistakable influence for the result.
Has anybody seriously compared Dalle2 with Midjourney? I know that midjourney is a relative no-name but personal testing with the same prompts give me better results (matches expectations) on midjourney.
A quick summary:
Midjourney’s images are usually more aesthetically pleasing than DALL·E, and the model is still adaptable and responsive to stylistic prompts. That makes it a great tool if you want to generate a lot of pleasant images quickly.
Dalle-2 is capable of delivering a wider range of visual styles. As a result, you’re much more likely to craft a surprising or amusing result, or create a never-before-seen image. It’s less likely the image will seem ‘AI-like’. You can also feel more confident that other users haven’t generated very similar images.
> Don't solicit upvotes, comments, or submissions. Users should vote and comment when they run across something they personally find interesting—not for promotion. [0]
> > Don't solicit upvotes, comments, or submissions. Users should vote and comment when they run across something they personally find interesting—not for promotion.
I read the GP as asking for an upvote feature on the site, rather than asking for upvotes here on HN.
Adding model provenance would be a good idea so you can support multiple. I was wondering if DALLE 2 embeds any metadata in the images but haven't had a chance to investigate.
I also think you should consider adding the option to clarify copyright status ... in particular, allow users to dedicate the image to the public domain via CC0 (https://creativecommons.org/share-your-work/public-domain/cc...), or maybe put it under a CC BY license. May be debatable what the legal status of the images is, but a CC0 dedication would at least get rid of any ambiguity.