DALL-E mini: The AI ​​stuff that meme dreams are made of – or nightmares?

Sephiroth is sad because of dandelions.

Not a day goes by that internet users don’t stumble across 3×3 tile images with strangely distorted, blurry, but sometimes very clearly recognizable motifs. The origin of these images, created by an artificial intelligence called DALL-E, is not only the AI ​​itself, but also the creatively very free-wheeling Internet community that DALL-E via the mini version that feeds the AI ​​with possible and impossible prompts.

This is how DALL-E mini works: You enter a prompt in English (we tried “Varian Wrynn telling his son Anduin a joke” and “Sephiroth crying out loud over dandelions”). DALL-E mini then calculates a little and presents you with the results after a few minutes.




Sephiroth is sad because of dandelions.



Sephiroth is sad because of dandelions.

Source: DALL-E mini




DALL-E: The stuff dreams are made of – or nightmares?

In most cases, an inquiry results in pictures with sometimes absurdly distorted faces. And yet we seem to recognize exactly what we’ve typed – sort of like the effect that causes us to read words correctly if only the first and last letters are correct. In our editorial office, we always call this operational blindness. But what is behind the DALL-E mini and the current hype?




Varian tells Anduin a joke.  Kind of creepy.  But somehow you can tell.



Varian tells Anduin a joke. Kind of creepy. But somehow you can tell.

Source: DALL-E mini






DALL-E mini is (via Spiegel) the stripped-down version of a 27x larger AI: DALL-E 2. This is from Open UI, a research organization funded by Elon Musk and Microsoft. And that’s where DALL-E gets creepy – if the distorted faces don’t already give you goosebumps of the nasty kind.

27 times more powerful = deep fakes?

You have certainly heard of Deep Fakes; Technologies that make it possible to have a virtual image of a real person do or say things that that real person would never do. With really good deep fakes, it becomes almost impossible to distinguish the fake from reality. In other words: Text-to-image AIs such as DALL-E, VQGan, LDM and Google’s Imagen allow realistic impressions – and in the wrong hands this can have fatal consequences.

Reference-www.buffed.de