I tested Dall.e 2, an AI that creates images from text

4 replies · 2K views · last reply
fluidistic
Gold Member
Messages
3,934
Reaction score
286
So I tested ChatGPT's cousin, Dalle.e 2. It is said to create images from text, and the website suggest to be very explicit in the description of the image, add many details, to get a good picture.
My mind started to wander... Would I finally get a pic of Einstein battling dinosaurs, or having a dinner with Newton while Trump and Putin are playing chess under the table? Boy did I was disappointed. Terribly disappointed.

Don't waste your time... this software is absolute garbage (for now at least), it's light years behind chatGPT.

Albert Einstein having a dinner with Isaac Newton while dinosaurs are playing chess under the table.
Cartoon of two white-haired figures with bird-like heads seated across a white-clothed table holding a roast fowl and a tray of food, with a pink dinosaur and a grey lizard-like creature in the background.
Music flowing in a river with pink reflections with physics equations on the water
AI-generated image of a dark water surface lit in bright magenta and pink, with blurred reflections and bright white glints on the ripples, showing jagged shapes that look like distorted structures along the top and middle of the frame.
Humanized apes having dinner on a table with Einstein while playing Bitcoin
AI-generated photo of two people in dark suits seated at a white table, eating from plates of food; a gold statue and a pink figurine sit at the center, with framed paintings on the wall behind.
A neandertal fighting an angry tyranosaurus rex with a screwdriver on a busy road
AI-generated image of a road scene where a large dinosaur bends its head toward a stopped white car, with a woman in green pants running on the left and a woman in white trousers watching on the right.
 
  • Haha
Likes   Reactions: mcastillo356
I had the same experience with Defusion Bee. It was about the same as what you are showing. People on DeviantArt are producing wildly better things with Ai, including complex stories with consistent appearances of the characters. I don't know what tools they are using, but it isn't DeviantArt's own Ai. The thing threatened to ban me for my first request, and didn't even say why.
 
  • Like
Likes   Reactions: Office_Shredder
The person who won that art competition said they spent 200 hours trying prompts before getting what they wanted. This isn't a fire and forget tool.
 
Yes. Ai can't create art. But then neither can pencils, cameras, or photoshop. Humans create art.

Digital painting of a young woman in a glass-domed astronaut helmet with orange trim, gazing off to the side against a starry space backdrop with small planets.

Night street scene after rain, with blurred neon lights, car tail lights and streetlamp glow reflected in wet asphalt, and dark cables running across the road.
Digital painting of a grand, gothic-arched library with tall wooden bookshelves, sunlight through arched windows, and tables piled with books and papers.
 
I have spent several evenings trying to configure stable diffusion to run on my computer using CUDA. I have spent even more hours generating over 24k images from different prompts (OK, my computer did, choosing and entering a prompt is like half a minute). I have but a few images that are reasonably close to being worth of showing.

Yes, at this stage it is mostly useless garbage.

Close-up portrait of a woman with long brown hair whose left eye is replaced by a yellow owl eye framed by grey and white feathers, with a mottled brown-and-white feathered wing edge across the top, in front of a blurred green background.


Painting of a pale woman with closed eyes, reddish-pink eyelids and tousled dark hair, wearing a red off-shoulder top against a muted pink and purple background.
 
  • Like
Likes   Reactions: BillTre, Algr and DennisN