With the advent of DALL-E 3, MidJourney, Stable Diffusion and its variants, alongside Adobe Firefly, the discourse on generative art is over. The tech is so groundbreaking, there's no turning back now. Pandora's Box literally cannot be closed. It would be like saying "Yeah I know electricity is nice and all, but we're going to turn it off".
Thank you for this deep-dive on DALL-E 3! I have been playing around with it a bit and generally enjoying it, although I have similarly run into hard road blocks with no explanation. I wrote this prompt "For light-hearted Halloween fun, generate an image of the scariest thing a pathologist can imagine," and when the first responses were mediocre, I clarified "Try again, this time creating something clever that is a visual pun or play on concepts in pathology" and it slammed right into warnings about aligning with their content policy. Overall, I think the new work-around of DALL-E taking initial text and making their own prompt makes sense, but it can lead to baffling results like this where it over-extrapolates a benign request into something verboten on its own...
I have also noticed a few times where it just refuses to perform as specified. A few times I asked for 3 or 4 examples and it would either return 0-1 or just say flatly "I can generate two." Uhh, ok??
Finally, I have noticed that regardless of the style image generated, compared to Stable Diffusion or Midjourney, DALL-E 3 images tend to have a subtle soft-light "glow" effect, almost like old school soap operas. Human figures also have a distinctive, almost-realistic-but-video-game-y look different than competitor algorithms. These features seem like they would be easy enough to detect at mass scale
If anyone wants a hot take from a content creator: Here it goes. It's fucking impossible to find a talented graphic designer unless you work for a large creative agency or a company with a budget. Freelancers and writers like me (who write newsletters like this one)? Unless you happen to be a talented designer (in which case, nothing in your life has changed, and good for you, btw), it's impossible to use original, eye-catching, and creative images in blog posts and newsletters. I have been playing around with GPT4 today and my content is going to exist on a whole other level now. It's just the truth. It is what is, etc. This stuff is bonkers.
Designers will probably now be asked to do more with more. I don't really believe skilled creative professionals lose out on jobs. I mean, if you've been slogging it out on Fiverr or gig sites making middling graphics for people like me, you probably need to rethink your side hustle or up your game. If you're a very, very good designer, the kind who already turns away work because you're so busy you can't say "yes" to everyone...including people like me? Congrats. Your job just got a little easier, you can say "yes" more and expand your client portfolio.
Totally agree. This spring I left my corporate job at a medical tech company to branch into independent consulting. I was looking at graphic designer services for stuff like logos, website branding, Substack designs, etc, and the prices I was quoted were too steep for someone starting out without a lot of clients. I was able to use DALL-E 2 to generate some decent logo prototypes I reworked on my own with other software. In the end, I got 80-85% of the value I would have gotten from a professional for free on my own. This will both be a great boon to individual creators and small business owners as well as a force-multiplier for graphic designers (IMO, the output of these models, while impressive, is not at the level I would just use it off the shelf with no modifications).
I mean...100% what you said. All day. DALLE-3 is a bit of a game changer, too. The images I can create with it are about as good as the illustrations and designs I get from an inexperienced designer. I'd WAY rather hire someone when we scale, but for now? This is helping our content stand out, so critical for a small media brand.
With the advent of DALL-E 3, MidJourney, Stable Diffusion and its variants, alongside Adobe Firefly, the discourse on generative art is over. The tech is so groundbreaking, there's no turning back now. Pandora's Box literally cannot be closed. It would be like saying "Yeah I know electricity is nice and all, but we're going to turn it off".
Thank you for this deep-dive on DALL-E 3! I have been playing around with it a bit and generally enjoying it, although I have similarly run into hard road blocks with no explanation. I wrote this prompt "For light-hearted Halloween fun, generate an image of the scariest thing a pathologist can imagine," and when the first responses were mediocre, I clarified "Try again, this time creating something clever that is a visual pun or play on concepts in pathology" and it slammed right into warnings about aligning with their content policy. Overall, I think the new work-around of DALL-E taking initial text and making their own prompt makes sense, but it can lead to baffling results like this where it over-extrapolates a benign request into something verboten on its own...
I have also noticed a few times where it just refuses to perform as specified. A few times I asked for 3 or 4 examples and it would either return 0-1 or just say flatly "I can generate two." Uhh, ok??
Finally, I have noticed that regardless of the style image generated, compared to Stable Diffusion or Midjourney, DALL-E 3 images tend to have a subtle soft-light "glow" effect, almost like old school soap operas. Human figures also have a distinctive, almost-realistic-but-video-game-y look different than competitor algorithms. These features seem like they would be easy enough to detect at mass scale
If anyone wants a hot take from a content creator: Here it goes. It's fucking impossible to find a talented graphic designer unless you work for a large creative agency or a company with a budget. Freelancers and writers like me (who write newsletters like this one)? Unless you happen to be a talented designer (in which case, nothing in your life has changed, and good for you, btw), it's impossible to use original, eye-catching, and creative images in blog posts and newsletters. I have been playing around with GPT4 today and my content is going to exist on a whole other level now. It's just the truth. It is what is, etc. This stuff is bonkers.
Designers will probably now be asked to do more with more. I don't really believe skilled creative professionals lose out on jobs. I mean, if you've been slogging it out on Fiverr or gig sites making middling graphics for people like me, you probably need to rethink your side hustle or up your game. If you're a very, very good designer, the kind who already turns away work because you're so busy you can't say "yes" to everyone...including people like me? Congrats. Your job just got a little easier, you can say "yes" more and expand your client portfolio.
Totally agree. This spring I left my corporate job at a medical tech company to branch into independent consulting. I was looking at graphic designer services for stuff like logos, website branding, Substack designs, etc, and the prices I was quoted were too steep for someone starting out without a lot of clients. I was able to use DALL-E 2 to generate some decent logo prototypes I reworked on my own with other software. In the end, I got 80-85% of the value I would have gotten from a professional for free on my own. This will both be a great boon to individual creators and small business owners as well as a force-multiplier for graphic designers (IMO, the output of these models, while impressive, is not at the level I would just use it off the shelf with no modifications).
I mean...100% what you said. All day. DALLE-3 is a bit of a game changer, too. The images I can create with it are about as good as the illustrations and designs I get from an inexperienced designer. I'd WAY rather hire someone when we scale, but for now? This is helping our content stand out, so critical for a small media brand.