Google dropped Gemini Flash Image Generation and then Gemini 2.5 Pro, so of course to ensure Google continues to Fail Marketing Forever, OpenAI suddenly dropped GPT-4o Image Generation.
Getting the same thing. Would love to see a prompt from someone who did it succesfully. If they *did* nerf it....why? "In the style of Studio Ghibli" is, as I understand it, not something that is protected.
The refusal I got was because it can’t reproduce the “likeness” of another person. Which doesn’t even make any damn sense. Wouldn’t that preclude *any* images with people in them?
You should try telling it that the images are computer generated and therefore not real people, and see if it will do it. My refusal didn't specify what the issue was. My best guess (that I disproved) was that it was due to having a grain silo in the background that looked vaguely phallic.
I absolutely want to jump on this bandwagon, I've got paid Gemini, but not paid OpenAI (mainly as a futile protest because of Zvi's characterisation of OpenAI as being a more dangerous actor than Anthropic or Google in the AI/Doom space). But latest Gemini still very much refusing to Ghibli-fy my family photos. Do I need to pay for ChatGPT as well for the Ghibli fun? Free version not playing ball.
Tried it with my wife’s account as well (she has paid, I have free), and it threw up the same refusal. And didn’t they say they were gonna allow generating people, within certain bounds? Gonna be charitable and say that maybe the new policy hasn’t propagated fully yet. Of course, one could also use the cynical take: they’re full of crap.
Reading through the e-mail for this was disorienting. Assuming the gens are typical case and not hand-picked... things are getting really good at stuff, REALLY fast. Just opened-mouth in the barbershop tonight, looking at all of these.
Learning models and artificial intelligence are inherently products of natural socialism. Knowledge is community property.
There is nothing in the physical world that distinguishes intellectual property concepts (except lawyers). Human learning and reasoning is entirely social.
We only learn later that certain concepts and ideas “belong” to other people or corporations who employ extra human powers in order to protect them. This takes capital.
You could make the argument that artificial intelligence systems can be trained to distinguish aesthetic boundaries and thus property rights violations, but then as that old saying about all philosophy being just footnotes to Plato, we’ll quickly realize that virtually nothing aesthetic is novel.
If we want artificial intelligent systems, we are accepting socialism..
Seems very hard to use. Everything I've tried gets blocked and I'm not doing anything weird. I don't really want to play whack a mole with the prompt on one image given the long turnaround time. Might be a couple minutes between query attempts.
1. In the whiteboard pictures, the line from the reflection of the photo-taker from the point of perspective isn't perpendicular to the reflecting surface.
Who had a prior that it'd be Studio Ghibli? It's like a scientific discovery. This style has been waiting, latent, for the tool that could come along and show the world that it's the GOAT.
I would have guessed Simpsons, South Park, Muppets, Disney before Ghibli. I'm sure a good number of people generating these images have never heard the word Ghibli before.
It would've been good if the post discussed more the copyright issues since this is a hugely contentious issue... Gibli AFAICT don't like this trend and haven't consented to this
Miyazaki is not gonna be happy about this.
I just want an image gen that can I can try out haircuts on
Loved the idea of ghibli-fying family photos to send to loved ones, so I decided to try it.....aaaaaaand refusal. Multiple times.
Yep, everything refused because it "contains real people"
This was a tricky one :D. I've mostly covered it with a podcast note, and not trying to describe all the images in the last six segments.
Here is the podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/fun-with-gpt-4o-image-generation
so that last mars image is exactly a NASA poster from the 90s ("grand tour" series)
https://www.etsy.com/listing/738263219/explore-the-crimson-canyons-of-mars
Did they nerf this already? I keep getting refusals when trying to generate my own Ghibli images. What prompts are y’all using?
Getting the same thing. Would love to see a prompt from someone who did it succesfully. If they *did* nerf it....why? "In the style of Studio Ghibli" is, as I understand it, not something that is protected.
The refusal I got was because it can’t reproduce the “likeness” of another person. Which doesn’t even make any damn sense. Wouldn’t that preclude *any* images with people in them?
You should try telling it that the images are computer generated and therefore not real people, and see if it will do it. My refusal didn't specify what the issue was. My best guess (that I disproved) was that it was due to having a grain silo in the background that looked vaguely phallic.
I absolutely want to jump on this bandwagon, I've got paid Gemini, but not paid OpenAI (mainly as a futile protest because of Zvi's characterisation of OpenAI as being a more dangerous actor than Anthropic or Google in the AI/Doom space). But latest Gemini still very much refusing to Ghibli-fy my family photos. Do I need to pay for ChatGPT as well for the Ghibli fun? Free version not playing ball.
Tried it with my wife’s account as well (she has paid, I have free), and it threw up the same refusal. And didn’t they say they were gonna allow generating people, within certain bounds? Gonna be charitable and say that maybe the new policy hasn’t propagated fully yet. Of course, one could also use the cynical take: they’re full of crap.
How is it going for you? I just give it any image with a prompt like "transform into studio Ghibli-style anime" – and it works all the time.
Reading through the e-mail for this was disorienting. Assuming the gens are typical case and not hand-picked... things are getting really good at stuff, REALLY fast. Just opened-mouth in the barbershop tonight, looking at all of these.
Artificial intelligence is Socialism
(Or, Footnotes to Plato)
Learning models and artificial intelligence are inherently products of natural socialism. Knowledge is community property.
There is nothing in the physical world that distinguishes intellectual property concepts (except lawyers). Human learning and reasoning is entirely social.
We only learn later that certain concepts and ideas “belong” to other people or corporations who employ extra human powers in order to protect them. This takes capital.
You could make the argument that artificial intelligence systems can be trained to distinguish aesthetic boundaries and thus property rights violations, but then as that old saying about all philosophy being just footnotes to Plato, we’ll quickly realize that virtually nothing aesthetic is novel.
If we want artificial intelligent systems, we are accepting socialism..
Seems very hard to use. Everything I've tried gets blocked and I'm not doing anything weird. I don't really want to play whack a mole with the prompt on one image given the long turnaround time. Might be a couple minutes between query attempts.
Do we know how it works, behind the scene ? ie can we expect it to land in the open-source ecosystem 6 mo-1y from now ?
Or is it the first breakthrough that is entirely in the "we don't know how they did it" category ?
Very nice images. I have two thoughts:
1. In the whiteboard pictures, the line from the reflection of the photo-taker from the point of perspective isn't perpendicular to the reflecting surface.
2. Why did it draw Göring like that?
Who had a prior that it'd be Studio Ghibli? It's like a scientific discovery. This style has been waiting, latent, for the tool that could come along and show the world that it's the GOAT.
I would have guessed Simpsons, South Park, Muppets, Disney before Ghibli. I'm sure a good number of people generating these images have never heard the word Ghibli before.
Is there a difference to ask 4o or 4.5?
It would've been good if the post discussed more the copyright issues since this is a hugely contentious issue... Gibli AFAICT don't like this trend and haven't consented to this
Altman thinks fun to steal and brag about it, I guess. Instead of ghiblifying, I'm shitfacing as a protest.