Luma's Ludicrous Dream Machine

 

Over the past few days I've been trying to create a modern parable based upon the New Testament story...

"The Flight into Egypt" 

I've used both the new Luma Dream Machine and some of Luma's competitors, all to no avail.

What I get back are comical crazy cartoon-like jumbles instead of a a serious representation of my prompt which often includes a picture as well as carefully worded text.

So here's one of my first visual prompts, courtesy wonderful copyright-free Rembrandt!




I fed this image into Luma with some accompanying text to guide it towards my intended output 


"Joseph leads Mary and Jesus through the rubble in the war zone of Gaza"


and here's what came out:





Starting with a different image I tried another test:


An elderly man dressed in historical Middle Eastern clothing leads a donkey which carries his wife and child. They pass through a war zone of a modern city devastated by bombs and rockets. There are children in the background who are sifting through the rubble looking for food and other items.




A few more terrible tests followed which bore little or no resemblance to the text I fed in as prompt, then I got the bright idea to start with a medium wide shot of the war zone and just type in text to prompt Luma to create the group of a middle eastern man leading his wife and child seated upon a donkey...



As the camera continues pulling back towards a wider shot, a middle eastern man is seen entering the shot from the bottom right corner of the frame. He is leading a donkey which carries a woman and a baby wrapped in swaddling cloth. This group is heading up the road towards the person in the red jacket.




WOW! WHAT A CIRCUS!


Luma Dream Machine should be called 

LUDICROUS DREAM MACHINE.


I've found it doesn't know left of frame from right of frame, or top from bottom.

Often I've asked specifically for a character to approach camera but instead it outputs a character walking away from camera.

My sister Chris created an image which I don't wish to show here, but the hand of her central female character had six fingers! Chris didn't ask for that extra finger.

I created a movie of a woman in a palatial residence wearing period costume and I got her to walk away from camera, stop and turn to camera. Then in "extend" I asked Luma for her to give a sort of "Come on" gesture for the POV camera to follow her up the stairs. 

What Luma gave me was a change of costume from her period style costume to an almost naked modern style:




It's so rude that YouTube will not let me embed it!

Well, well, well!

Let's see about that!


I'll finish by saying I would love to buy Luma if it worked properly, but it has so many failings I just can't be bothered. I've wasted many hours so far. All I get for my subscription is a lot of frustration and total disappointment.

pt


Comments

  1. It's kind of reassuring to see that AI or man-made systems are not as good as Man him/herself. We should be aiming to understand our inner selves and how we can aspire to be better, wiser, more compassionate etc., rather than trying to create engines that can/would outsmart us. You, Peter, are still a great artist. You don't need AI to create the works you create. Never have, never will. You are the artist, not some machine. Machines are only tools. These ones are ineffective to say the least and should be discarded.

    ReplyDelete
  2. I'm flattered by what you gave said David, but I'm not ready to throw them away entirely just yet. Imagine if Shakespeare had said, " I am the artist here, I just don't need this silly quill!"

    ReplyDelete
    Replies
    1. His actors would have needed very good memories!

      Delete
  3. Yes indeed! With or without his quills!

    ReplyDelete
  4. I went to a business networking meeting last night and I spoke about this exact topic. None of these generative platforms can produce a final adequate result yet. What I do is use Photoshop to fix the parts of the image separately. In your image I'd have to search for a range of stock images of faces I think would look good, composite it into Photoshop, scale and retouch again, then upload it again to refine it. Or, repaint the whole thing using a "master oil" brush on my tablet, which I could do as I have manual illustration skills. However, all of that seems very counter to the purpose of generative ai art tools as it takes time!!! I'm thinking this is a real advantage for genuine artists because they are still needed to achieve the final vision. Great post Peter!!!!

    ReplyDelete
  5. Thanks for that Anon! What you say makes a lot of sense, it would be totally counter productive to have to do all that extra mucking around.

    On the good side, what they are giving us recently is very promising, but there's still a long way to go. It would be great if Luma gave us a proper feedback line to tell them what went wrong, instead of just "thumbs up" or "down". At least Runway Gen2 does give you a set of options to tick what went wrong, e.g. "not even close to my prompt".

    And as I've discussed with friends, what they are all offering is changing and improving very quickly, but not quick enough for me!
    pt
    pt

    ReplyDelete
  6. I agree with David wholeheartedly, about humans being better artists ! But he also says "machines are tools and these ones are crap", and that makes me think that it's these PRIMITIVE AI engines that maybe can produce great art, just by accident. And by great art I mean something so surreal that it goes beyond what we currently know as surrealism. Accidental surrealist art. But to get these machines to be sophisticated and produce exactly what we want, or what exists so far, well, that's a bit pointless - we ourselves may as well be the artists, it's more fun that way !

    ReplyDelete

Post a Comment

Popular posts from this blog

Maria Korporal: An Introduction to Her Work

Update on Sora, Luma, Runway Gen-2 Ai text and image machines.

Eyeless in Gaza, images from the net.