Having been mildly entertained that ChatGPT doesn't like Donald Trump, I commented to @peppermint24 that I was disappointed ChatGPT wouldn't draw me a picture. Little did I know, ChatGPT has a sibling that was designed to do just that - DALL-E!
Prompted by as many, or as few words as you feel like sharing, DALL-E will generate an image to your specification and subsequently allow you to generate variations of whichever image you like the most. It even allows you to upload your own pictures and will subsequently generate 4 variants.
鈽濓笍 A somewhat bizarre variant of the-toddler-gorilla
As you can see from the image above, it does some things better than others. My original image had a similar background but the-toddler has been replaced by what appears to be a completely random child in a random outfit.
Having pissed about with it for about an hour, I've discovered that I only get 50 Free Credits which is unfortunate and I wish that I hadn't created abominations like this:
鈽濓笍 I think I told it to change the child in the original photo into a Gorilla
馃憤 Things It Does Well
It seems to be very good at cartoons. When prompted to "Draw a cartoon of a man carrying 10 dogs", I was provided with the following 4 variants:
Having said that, whilst proof reading, I've realised that none of the images have 10 dogs.
It's also seems to be fairly good with 3D renders ("3D render of gorilla using a computer"):
I also created a "high resolution 3d render of a lava lamp in the window of westminster":
Which I also thought was a fairly decent effort.
The other thing that I found interesting is that you could ask it to paint something of your choosing in the style of your favourite artist. In my example, I asked for "A framed painting of a gorilla by Doug Hyde" and whilst I've included this in the "Things It Does Well" section, the rather cool output is some way from what I'd consider to be Doug Hyde's style:
When asking for a "Painting of a dog by Doug Hyde", an animal that Doug Hyde has produced in the past, one of the outputs was particularly good:
馃搱 Things That it Could Do Better
It seems to struggle to display locations and faces of people. Whilst my Donald Trump request was said to violate their Content Policy, it did try to create a "Photograph of Harry Kane celebrating a goal on Mars":
Whilst I quite like the 2nd image, Harry Kane, it is not.
Once again, I tried another variant - "Photograph of Harry Kane celebrating a goal on Mars in the style of Doug Hyde, digital art" which I thought generated a better style of image but there was no discernible resemblance to Tottenham's all-time record goalscorer.
As for locations, this is its attempt to "draw a 3d model of a gorilla scoring a goal at Camp Nou (Barcelona)":
There's also no request for the Gorilla to be carrying a walking stick.
I also tried to create a "Photo of an Arsenal fan crying at Goodison Park" and whilst the result looked like something you might find on Shutterstock or Getty Images, it's (obviously) not an Arsenal fan nor is it Goodison Park.
As you can see, I played with it quite a lot and a couple more examples before bringing this article to a close...
I thought I'd test how good it was at creating a movie poster, with my "Harry Potter and The Raiders of the Lost Ark" mashup in mind. The most successful input ("Harry Potter and The Raiders of the Lost Ark" movie poster, digital art") produced the following:
Whilst I thought that the imagery was pretty cool, the random text ruined it somewhat so if I tried to create something like this again, I'd add the text afterwards using Photoshop. It's also worth noting that I didn't crop any of these images - they're resized to fit this post but no part of each image has been removed. This highlights another limitation - all of the images created are square (1024x1024) and trying to define a different resolution doesn't seem to change anything (although I've not read any guides so it might be possible).
The text is where I felt particularly let down though. I asked DALL-E to create a logo for Steemit, first as an oil painting and then with the simple request "Steemit Logo". Whilst the choice of palette is pretty good, it's not the greatest output unless there are plans for a rebrand and a move away from the "Steemit" name:
馃挱 Final Thoughts
I thought it would be cool to start using DALL-E as a novel way for creating images for my Steemit content. Despite some of the output being fairly poor, with a bit of trial and error, I'm sure that it can generate something that I'd be happy to use. It's also quite good fun seeing what it comes up with and which words (e.g. Poo) prompt it into a content violation.
The downside with this approach though is that I've nearly used all of my 50 Free Credits and I'll need to wait another month before getting another 15 (you get 15 credits for Free each month after the initial 50). I could buy more credits but that's not how I want to spend my money.