#text2image

[aiweirdness]

AI-generated advent calendars

When people post the outputs of text-to-image generators, they tend to gravitate toward the kinds of things the AI seems to be doing well. One of my hobbies is collecting failure cases.

Here's DALL-E2 attempting "A Christmas advent calendar with numbered doors". When it gets to numbers higher than 10 it seems to *mumble mumble*

And here's midjourney (--v 4 for those curious) attempting "A Christmas advent calendar with numbered doors". This model is particularly tuned to fit the aesthetic tastes of people who like fantasy art. And apparently the number 2.

But what if we ask it to open some of the doors?

Dall-E2: "Product photo of a Christmas advent calendar with numbered doors and some open doors showing the scenes inside" (mumbling becomes more cursed)

Midjourney: "Product photo of a Christmas advent calendar with numbered doors and some open doors showing the scenes inside"

Also I made an interactive advent calendar at aiweirdness.com with correctly numbered doors that work!

[nappe-plays-the-sims]

I put in Bella Goth as a prompt in pixray’s AI generated art machine thing and... Well, this is certainly something.

[harmonic-motion]

popcorn alphabet tiger

[the-witcher-parks]

Arrival par Pete Vielhaber Via Flickr : A painting created with Midjourney AI V5.1 Prompt: “A steam locomotive in the countryside, monet style :: Impressionism, detailed brush strokes, mysterious atmosphere, surreal, soft lighting, magic glow --ar 18:9 --v 5.1 “ The light, painterly brush strokes were applied in GIMP using G’MIC artistic filter: Brushify (with a custom brush). With this filter, the brush strokes follow the contours of the image and you can adjust how many directions the brush strokes go in. Nice.

[mysocial8onetech]

Discover the power of Text2Video-Zero, a new approach to text-to-video generation that leverages existing text2image synthesis methods for low-cost and high-quality video generation. Learn more in our latest blog post!

[orum]

I spend the morning playing with and discussing the dall-e 2 algorithm together with @andreasref (who has access to the as yet unreleased version). So many interesting things to think about, such as the relationship between text and image when images are generated from text prompts, it's computer game looking images of humans as a sign of it perhaps being trained with images from games, and how open ai frames it's artist access program and so on, and so on. No doubt this technology, or something like it, will come into widespread use, and affect the images we see, and by extension our outlook on the world in the coming years. speculating on how this might affect us and how we see the world seem pressing to me. Here are a few results of the text prompt "a man under the covers with a signal crayfish under who is also under the covers at night", note the movie or theater play manuscript form that works best, as if "all the world is a stage". #signalcrayfish #dalle2 #text2image (at Captive Portal) https://www.instagram.com/p/CeV9hu2suso/?igshid=NGJjMDIxMWI=

[nappe-plays-the-sims]

AI generated art with Mortimer Goth as prompt.

[harmonic-motion]

when does it all stop

[euanlynn]

[harmonic-motion]

dog throws a fish

[harmonic-motion]

long way from winter

[harmonic-motion]

This is not my beautiful wife

[harmonic-motion]

hogs for Trump

[harmonic-motion]

bird surfing wonderwall

[harmonic-motion]

The test is over

[harmonic-motion]

Disney World albatross