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

Spent $15 in DALL·E 2 credits creating this AI image

pub.towardsai.net

Post reply on HN
Print view View original

131–140 of 153 posts

Re: Spent $15 in DALL·E 2 credits creating this AI image

#131

Post by yunwal »

aetherson wrote: post #126
TillE wrote: post #123

Earlier quoted context omitted.

It's not the whole problem, but it looks like the "digital art" bit is dominating the style in a way you maybe didn't intend.

Here is the same prompt minus "digital art." https://imgur.com/a/aVbSxHe You will still note that none of them are far shots, that no depicted character actually has fully green skin, and one of the four has nothing even remotely like leaves for hair. I mean, is it better? Sure. They're less ugly, though none of them are what I'd call great results. But they also aren't really doing a basically competent job of fulfi…

I think Dalle's ability to produce good images out of the gate is pretty limited, but I've found that using the fill-in feature along with existing images from google and photoshop, I can pretty much get anything I conceptualize with about 20 minutes of work and like 10 prompts.

It's not fully removing humans from the equation, but you can take something that used to take days and make it a 20 minute operation.

Re: Spent $15 in DALL·E 2 credits creating this AI image

#132

Post by yunwal »

spywaregorilla wrote: post #117
Workaccount2 wrote: post #76

Earlier quoted context omitted.

"We let our graphic designer go so we could onboard a AI Prompt Engineer" "How much are we paying him?" "About $225k plus bonus and equity" "And how much was the graphic designer paid? "$55k" "..."

How long does it take the prompt engineer to make a design though?

Not a professional graphic designer, but did some graphic design classes and photography/photo editing classes in college, and still do it as a hobby.

Things that at one time took days can now be done in minutes with some skillful use of Dall-E + Photoshop. IMO, any image editing software that incorporates a similar technology will take over the market and it'll be one of the most important features in any graphic designer's toolkit.

A talented graphic designer who can also use a dall-e like tool is worth at least 5x the pay of one who can't (although I don't think we're going to get a "prompt engineer" title, it's really not that difficult a skill to pick up for people who already do image editing).

Re: Spent $15 in DALL·E 2 credits creating this AI image

#133

Post by yunwal »

qeternity wrote: post #113

Purely economic take: I’m sure that as knowledge builds over time, people will get more efficient at prompt generation, but the $15 in credits ignores the cost of the time spent to build the final prompt. I wonder how this compares to a junior graphic designer in terms of TCO.

The future is graphic designers who can proficiently use Dall-E. You can't get what you want easily with just a prompt, but you can also have it modify existing photos, so Dall-E + Photoshop is very powerful

Re: Spent $15 in DALL·E 2 credits creating this AI image

#134

Post by itisit »

ma2rten wrote: post #127
samspenc wrote: post #109

Earlier quoted context omitted.

Fascinating, are there any other similar products in this same category as DALL.E and Craiyon?

Not products, but Google Research already published papers about two different, better models: https://parti.research.google/ https://imagen.research.google/ The models themselves are not public however.

Wow. Those models, particularly Imagen, are of an entirely separate calibre. None of that psychedelic foggy memory swirl that is characteristic of the space. I can see why Google Research is hesitant to release them.

Re: Spent $15 in DALL·E 2 credits creating this AI image

#135

Post by aetherson »

yunwal wrote: post #131
aetherson wrote: post #126

Earlier quoted context omitted.

Here is the same prompt minus "digital art." https://imgur.com/a/aVbSxHe You will still note that none of them are far shots, that no depicted character actually has fully green skin, and one of the four has nothing even remotely like leaves for hair. I mean, is it better? Sure. They're less ugly, though none of them are what I'd call great results. But they also aren't really doing a basically competent job of fulfi…

I think Dalle's ability to produce good images out of the gate is pretty limited, but I've found that using the fill-in feature along with existing images from google and photoshop, I can pretty much get anything I conceptualize with about 20 minutes of work and like 10 prompts. It's not fully removing humans from the equation, but you can take something that used to take days and make it a 20 minute operation.

Yeah, the fill-in feature is great.

Re: Spent $15 in DALL·E 2 credits creating this AI image

#136

Post by uoaei »

bredren wrote: post #64
krisoft wrote: post #14

> it was difficult to find images where the entire llama fit within the frame I had the same trouble. In my experiment I wanted to generate a Porco Rosso style seaplane. illustration. Sadly none of the generated pictured had the whole of the airplane in them. The wingtips or the tail always got left off. I found this method to be a reliable workaround: I have downloaded the image I liked the most. Used an image editi…

I had similar problems trying to get the whole of a police car overgrown with weeds. https://imgur.com/a/U5Hl2gO I was testing to see how close I could get to replicating a t-shirt graphic concept I saw. I had been using ~"A telephoto shot of A neglected police car from the 1980s Viewed from a 3/4 angle sits in the distance. The entire vehicle is visible but it is overgrown with grass and flowery vines" This process…

Original: https://foreveryonecollective.com/products/abolition-is-crea...

Re: Spent $15 in DALL·E 2 credits creating this AI image

#138

Post by speedgoose »

astrange wrote: post #104
bpye wrote: post #89

Earlier quoted context omitted.

I suspect this is a joke, but I did find that it was a little overzealous with the filtering. I was trying to get someone (not a specific person) shouting or with an angry expression, and a few prompts I came up with were blocked. Not banned though.

I kept getting a scene with "two people holding hands" blocked, it allowed "two people kissing" and then when I tried "and wife" instead of "two people" it banned me. (They unbanned me when I emailed them though.) Oddly, the ones it blocked were more sfw than several others it allowed, but of course I don’t know what the outputs would’ve been…

I’m guessing they have a filter on the prompt text, but also one on the generated pictures.

I got blocked a few times with very non sexual prompts, and I suspect that the AI was a bit horny when it interpreted them.

Re: Spent $15 in DALL·E 2 credits creating this AI image

#139

Post by andybak »

scifibestfi wrote: post #116

> Tip: DALL·E 2 only stores the previous 50 generations in your history tab. Make sure to save your favourite images as you go. This is kind of funny. DALL·E is one of the most impressive pieces of software, but such a basic feature like history is curiously underpowered.

History is much bigger than 50 now. 1500 is so if I recall correctly.

Re: Spent $15 in DALL·E 2 credits creating this AI image

#140

Post by numpad0 »

bredren wrote: post #64
krisoft wrote: post #14

> it was difficult to find images where the entire llama fit within the frame I had the same trouble. In my experiment I wanted to generate a Porco Rosso style seaplane. illustration. Sadly none of the generated pictured had the whole of the airplane in them. The wingtips or the tail always got left off. I found this method to be a reliable workaround: I have downloaded the image I liked the most. Used an image editi…

I had similar problems trying to get the whole of a police car overgrown with weeds. https://imgur.com/a/U5Hl2gO I was testing to see how close I could get to replicating a t-shirt graphic concept I saw. I had been using ~"A telephoto shot of A neglected police car from the 1980s Viewed from a 3/4 angle sits in the distance. The entire vehicle is visible but it is overgrown with grass and flowery vines" This process…

These are trained with pairs of image and caption text, so they work better with text inputs that resemble description for paintings than with simple descriptions or with William-Gibsonian hyperspecified description-text, though it's tempting to do the latter two.

https://imgur.com/a/YB5StlE

Post reply on HN