the two images (click to enlarge!) show how differently the two main portals for text-to-image Ai “think”. i wanted a spider morph into a butterfly. both approaches show a macro shot. DALLE-2 (left) interprets my text prompt in a geometrical way: the two spider resemble the shape of a butterfly. midjourney replaces the body of butterfly with a spider.

both portals offer a selection of four first approaches which i show you below. depending on the seed (a parameter of the random function) the outcomes would be totally different. morphing a spider into an insect is not trivial at all. morphing a → rolls royce car into a crocodile is much easier.

this blog is more or less my official diary where i post things others might be interested in. i started this project 10 years ago, for only one reason: a radio broadcast about self publishing of books. i used the blog entries of a whole year for the → book of 20 copies.

in the meantime the blog has grown to over 1,800 entries. i earn nothing with it, it’s just good to put things there and lay myself to rest = forget about it. today the blog jumped across the mark of 1.5 million views with about 500,000 visitors.

guess what my prompt for this quite bizarre Ai generated object was?: Cup A Tanktop.

↓|

↓|↑

↓|↑

↓|↑

this image is not a photo, but the rendering by #midjourney text-to-image #AI. it is in many respects amazing. for example the light, shadows and reflections, the furniture, the face of the boy, the hair — all perfect. the concert guitar has a few flaws which are not obvious when looking at the image from a distance. the main problem with text-to-image AI are the hands. in this case, the right hand shows AI going one step further: the right hand is playing the guitar with proper finger picking, and we even see the thumb! normally, AI displays hands with three or five or six straight fingers. the boy‘s left hand has the usual problems. buttom line: the AI hands get better, they are much better than a year ago.

typical AI hands:

das bundesverfassungsgericht hält künstliche intelligenz für gefährlich für den datenschutz und den schutz der privatsphäre, vor allem beim bilden von personenprofilen durch die polizei. eines der argumente ist, dass die KI innerhalb von sekunden komplette biografien entwickeln kann.

in den drei beispielen unten antwortet chatGPT auf meine bitte, mir eine kurz-vita über die person maximilian schönherr zu schreiben. wenn ich diesen text auf mich beziehe (viele menschen mit diesem namen gibt es nicht), stimmt in dieser ersten version fast nichts. aber die vita ist so perfekt formuliert und hat ein flair, sodass sie leicht gespenstisch wirkt.

das problem dabei ist, dass man nicht mehr unterscheiden kann, was stimmt und was nicht stimmt. wenn ich zum beispiel den bot bitte, meine lokal berühmte freundin → lioba albus zu beschreiben, stimmt manches, vieles aber nicht. lasse ich ihn meine weltweit berühmte freundin → louise lecavalier beschreiben, stimmt alles. ich vermute, dass chatGPT versucht, der wahrheit nahe zu kommen. aber er deklariert an keiner stelle seine unsicherheit. wenn er mich am 7. mai 1985 auf die welt kommen lässt, stimmt das nicht. falls er jemand anderen mit meinem namen meint, deklariert er das auch nicht.

auch bei einem anderen „seed“, also einer neuen auf zufall basierenden version, bleibt die KI dem thema treu: musiker, komponist. das geburtsdatum wurde vorverlegt, und der lebensmittelpunkt nach berlin bewegt:

bei der dritten version kehrt der bot wieder nach wien zurück:

i fed these words from lana del rey‘s marvellous song “A&W“ into the text-to-image #AI #midjourney. it produced several results, one of which is the one below. i wanted angels and drones in the image. i find no angels. i wanted the image to look like a painting from the chinese han dynasty.

i created these two images with text input and #AI. both images have flaws, but in general they are excellent examples for graphic design, stock photography e. a. i could merge the two images into one which would probably look mysterious, if not horror like. i could instruct the AI to re-render the image with another “seed”, which is basically the starting point of the randomized process. but currently i’m in the process to check out different scenarios, and i need to be cautios with my paid access to the #midjourney bot.

vier spams vom wochenende. solche „sex“-nachrichten sind seit einigen jahren seltener geworden. deswegen lohnt es sich, sie aufzuheben. die angegebenen web-links sind alle falsch; sie führen wo ganz anders hin, nämlich auf webseiten, die dem opfer viren unterjubeln, mit allen möglichen, oft drastischen folgen. interessant finde ich, dass die fake-adressen alle http zeigen, statt dem seit längerem üblichen https.

i tried a couple of variations of the top left image, two of which you see below.

i asked #AI to show me a rural midday scenery with the four Beatles, with long hair, riding on their bikes. for a first attempt not bad at all. ringo, followed by paul, john and george.

before that i had asked #midjourney AI to visualize john lennon on a bike in a messy abbey road studio. the bot delivered an amazingly complex, chaotic scene with someone (not john) on the drums and certainly no bicycle:

for these two images i asked AI (midjourney) to create two tables, one with communists, one with fascists. guess which is which:

i started this song with nonsense in mind. however, it turned out much “deeper”. the chorus is merely a wordplay, but in the context of the rest, it gains a bit of weight.

We put on a mask, hide our true selves / Let’s pretend everything is alright / Deep down we all struggle and fight / And hope to find some peace inside / So, let’s be real, let’s be raw / Embrace the flaws and all the scars / For we all are imperfect stars / That’s what makes us who we are / Nobody’s perfect, everybody’s fine / Everybody’s perfect, nobody is fine / Everybody’s perfect, everybody’s fine / Let’s put on our masks / Hide our true selves / And pretend everything is alright / In this world of black and blue / We often feel so lost and confused / Our struggles pile up / Leave us defeated and used/ Nobody’s perfect, everybody’s fine / Everybody’s perfect, nobody is fine / Everybody’s perfect, everybody’s fine

the chorus appears twice, and in both instances my daughter sings the second half. there are 5 audio tracks in this song, due to the use of vocal harmonies. for example, i doubled parts of my voice in the verses using a slight formant change. the snappy bass and the main pad come from izotope iris 2, the drums from toontrack EZdrummer 3. the hammond organ is the B4 by native instruments. for the vocals i used waves harmony. this plug-in is not easy to handle with chord changes, but fun to use.

the watercolour painting above was created with (my) text input and the help of midjourney AI.

a text to image experiment using midjourney Ai. i wanted a small subwoofer and lots of schoolkids. the subwoofer is not exactly small, but definitely soundproof.

below are a couple of variations, all done by the Ai within seconds.

i created this image from text. midjourney Ai did a marvellous job here. below you see the first four variations. i picked the buttom left for upscaling.

this is my first song with quite extensive use of automated vocal harmony. i used waves’ harmony for the polyphonic voice handling. the lyrics are about a triangle relationship, a train stopping and the middle of life, the middle of a maze. the image above was created using text-to-picture Ai.

There’s a wall between you, your lover and me. / There’s a well between your lover and me. / Wait a minute: The train stops! / Wait a minute: The train! / But when I turn round, I see the mirror, / And in the mirror there is a face / Of the girl I never knew. / A girl I’ve never seen before, / I’ve dreamed of so many times. / Yes, I’ve told you there’s a wall / Between you, your lover and me. / (That’s not quite true.) / There’s a wall between you and me, / Your lover and me / In this bigotomy (?) of our life. / We’re in the middle of our lives. / We’re in the middle of our days. / We’re in the middle of our maze. / But when I turn round, I see the mirror, / And in the mirror there is a face / Of the girl I never knew. / We’re in the middle of our days. / We’re in the middle of our maze.

this is a photo i fed into #dalle2, an artificial intelligence portal:

below is one of four Ai variations. does Ai understand the concept at all?

yet another variation:

i composed this song about thin (and still) air today. the chorus ends with love in a cave. the piece has rap and rock elements. apart of singing i played all instruments on my midi keyboard. c minor. iris 2 is the major VST instrument. the drum sounds come for EZdrummer 3, most of the drum patterns from unison audio.