AI Don't Wanna Go Anymore, Picturesidrawn, Digital, 2026
AI Don't Wanna Go Anymore, Picturesidrawn, Digital, 2026
I agree it shouldn't be used for art/ creative stuff - but adding to that my experience of AI for anything related to repairs/maintenance is the results are riddled through with errors so its not even usable for that purpose..
Well, yes, repair manuals, mushroom identification guides, etc. are "self-correcting", in a way… ⚡🍄
… if you have enough expendable family members, that is.
@katzenberger @vfrmedia @cmconseils I do not think there will be any difference whether the manual is written by AI or a low paid employee trying to get home before the rush hour. I never read the manual, anyway….
“Please hand me the red wire…”
“The mushroom with the blue spots looks nice. Let’s eat that one.”
“Is this the trigger?”
“Don’t need no stinkin’ manual.”
But poetry and AI should never meet.
@patrick @cmconseils Even the repair manual used to have a human designing it*. And there's art even in the mundane.
*and if you want it to be accurate, it probably still should. 😁
@freelikegnu Well, we mock that, or we mock Comic Sans...
or best of all, both at once.

Wow and well done to the pasta vlogger for making the most of their trip out.
Definitely an introvert's dilemma - all the effort to go out and then noone to meet.
Perhaps good because you didn't have to socialise after all, or a waste of money as you could have ordered a takeaway and stayed home?
@cmconseils Tough question: Which is worse, AI ‘art’ or ‘typography’ in Comic Sans? 🤪
@cmconseils Unless there really *is* a flatulist performing.
Because I would need to see that.
Yes to this.
Bring back the creative use of Clipart, guys! Some old victorian drawings or hippie flowers played around with gives the novice an hour of creative fun. It looks unique. And it has more soul than a twelve fingered karaoke singer wearing a tight tube top and is staring blankly at a wall whose corners don't converge in the distance.
@divVerent @dec23k @cmconseils
You can tell it has no idea of what a karaoke bar looks like, no concept in reality.
I figured if generative AI were to pull together such an image it would at least begin with stolen social media images of darkly lit bars.
I guess we're lucky it didn't generate something that looked like a 12 yr old in skimpy clothing.
@MyWoolyMastadon @dec23k @cmconseils To give it credit, initially it was so dark I could not even see any lines that would be supposed to converge. I had to ask it to up the lighting. See my prompt that I put verbatim in the alt text.
So that error is on me.
I still got no lines converging issue, but the wall corner in the back is oddly slanted. So it halfway succeeded.
I really was hoping for it to cover a room's corner with something so it can then let the three lines not converge. But no dice.
@divVerent @dec23k @cmconseils
Excellent attempt, my friend.
Maybe it failed with the non convergence of lines to the horizon because you asked it to. Someone might be working hard on the backend to remove that telltale sign of it being generative AI.
And yes. I would believe the first attempt was too dark, another AI tell. It probably even had the wrong shadows due to the lighting as well.
@MyWoolyMastadon @dec23k @cmconseils No, that telltale sign also didn't appear without me asking for the lines not converging.
I presume the way they fix it is by describing what they want to look for and avoid in the system prompt (so whatever I write to get it to fail would have to be "stronger" written than the system prompt). I am actually surprised I got extra fingers to work.
At the same time, all these system prompt additions make it impossible to create many _interesting_ things. As an example, I asked it for a version of M.C.Escher's Waterfall picture, but underwater (with air bubbles moving in a cycle) and the necessary adjustments so it can work (primarily the air bubbles have to flow _below_ the structure, the paths have to be closed at the top and open at the bottom, and of course the camera position be adjusted so one can still see how it works). It did not even get close. It indeed can't make any impossible geometry except by accident, or by 1:1 replicating existing drawings.
What's the point then? That would have been a _really cool_ thing... but no. It can just create generic crap. Good for a meme or two, but had the box of the Pandora never been opened, we'd still be happy with MS Paint style memes.
@MyWoolyMastadon @cmconseils I am actually appalled that the _usual_ method of detecting AI slop isn't working.
The red lines, all parallel in 3D space, actually do converge sufficiently well in a single point.
Do note the mistake with the baseboard in the white ellipse.
Also, the rear wall is slanted on the left but doesn't seem to have any slant on the right. That's rather specific to this image though.
Even though it failed here, showing the method anyway to help people spot the slop easier.
Sadly I'm seeing lots of AI images in use for podcast thumbnails. If you thumbnail is AI, then I'm not going to listen.
@cmconseils Very good, but needs more irony.
Just kidding, I like yours more.