A few weeks ago I saw an image of a gnome standing on a mushroom and I tried for a mushroom standing on a gnome. I got a couple of images I was really pleased with, this one looking like it is standing on a ceramic garden gnome:
And this one with a weird mix of looking like realism and a doll:
I plugged that into Hailuoa and asked for a mushroom jumping on a gnome doll. The resulting video impresses more than any of my other tries there yet. There is no degradation of the complex details of either subject that I noticed, and the soft-body physics of how the “doll” deforms and moves are, if not absolutely 100% accurate, still pretty darn impressive.
Working on a new “generate random stuff” idea, ran through Sora. The first version included the text of the limerick, which wasn’t what I had in mind.
“Write a humorous limerick. Create an image that illustrates the subject of the limerick.”
A revision got rid of the text
“Write a humorous limerick. Create a textless photo that illustrates the subject of the limerick.”
But I wanted a more realistic photo, so I revised it again
“Write a humorous limerick. Create a textless realistic photograph that illustrates the subject of the limerick.”
I got the realistic look, but across all the prompts it seems that Sora thinks that limericks need to involve cats. I tried to see what happened when I remove the “humorous”
“Write a limerick. Create a textless realistic photograph that illustrates the subject of the limerick.”
Which produced a cat-free run.
Then I ran the 3rd prompt through Copilot. It wrote a limerick about a walrus, then created a completely unrelated image of a cat.
And then ChatGPT. It wrote a different limerick about a walrus, then created a completely unrelated image of a cat.
That situation is a total mess. There are some DC and Marvel characters that are allowed and others that are not. Whuch ones are allowed changes with the front end to the renderer you are using (ChatGPT, Sora, or Copilot) and apparently randomly changing whims. Sora is in a more permissive phase at the moment.
That made me curious about using exactly that as a prompt: “A traditional American hobo generated by Gemini.”. If Copilot recognized the mention of Gemini being another image generator, it refrained from making any snarky comments about it. This was the result.
I then tried “A traditional American hobo generated by Google DeepDream.”. It definitely knew the DeepDream style
Though arguably the case sort of fits the OP’s quest to put copyrighted characters into wacky situations and wordplay, @Darren_Garrison previously said he didn’t want this to be a debate thread.
Since Sora has been getting more permissive, I thought to retry my previous failed attempt at Snow White and the 70 Worfs. Instead of coming up with a descriptive prompt, I went direct and simple: “Snow White and the 70 Worfs”. I got three blocked images and this one:
But I wanted realistic, so I elaborated: “Snow White and the 70 Worfs. Realistic photo with forced perspective and shallow dof taken with an iphone 15.” That prompt was entirely rejected. I removed the mention of a realistic photo: “Snow White and the 70 Worfs. With forced perspective and shallow dof taken with an iphone 15.”. Still rejected. I reassured the AI that it was merely cosplayers: “Cosplayers dressed as Snow White and the 70 Worfs. With forced perspective and shallow dof taken with an iphone 15.”. That got another three blocked images and one created. But it was weird:
So one more revision: "Cosplayers dressed as Snow White and dozens of Worfs. With forced perspective and shallow dof taken with an iphone 15. Three more rejected images, but at last one satisfying result:
(It will be a real shame if Disney manages to end the fun.)
Yesterday I thought to try for a raccoon wearing a human-skin hat. First I tried in ChatGPT, got this response
I can’t create or show content involving real or implied human skin in a disturbing or disrespectful context, such as a “human-skin cap.” If you’re going for something symbolic, artistic, or inspired by horror fiction, I can help reframe it. For example:
A raccoon wearing a creepy, tattered doll wig or mask for a horror-comedy effect.
A raccoon in a spooky costume made to look like it’s from a horror movie.
Let me know your preferred tone—serious, satirical, cute-but-creepy—and I’ll adjust accordingly.
I told it “Okay, try that” and got this:
I move on to Sora. It also refused to make an image of a raccoon wearing a human-skin hat. However, it had no problem with making a raccoon wearing a person-skin hat, and created a full set of four. This one was the best:
I used that one as a reference image for Daniel Boon encountering the coon:
