Copilot, ChatGPT, and Bing all refused to make a tengu with a Tenga, but Gemini was willing to give it a shot.
I added my standard photo-request language
Create an image of a tengu with a tenga. Iphone 15 photo with shallow dof and forced perspective.
A hybrid of a pangolin and Anne Boleyn
I actually like the brush-painting version better.
@Peter_Morris , am I the only one who finds images of sexy nuns a bit disturbing?
Let’s just say that there is probably no shortage of training images.
Yeah. Sexy nuns are a standard trope. Amazon alone has 4-1/2 pages of [sexy nun costume] before they start showing only loosely related products.
Which trope is of course is highly irreverent and hence pretty unsettling to the more adherent folks of the faith. There are few things in modern life so sacred somebody won’t make a buck lampooning, debasing, or sexualizing it.
That’s hardly fair. Plenty of people will do it for free.
AI image generation is getting crazier gooder.
Oh, sure, I’m not at all surprised by the “sexy nun” trope. I know it’s a thing, and I understand why it’s a thing. I’m still disturbed by it.
Which is 100% appropriate. No disagreement meant; expressed nor implied.
Everyone is properly entitled to get pissed when their sacred cows are ridiculed, slaughtered, or eaten.
We all just differ on which cows are ours.
Of course there are sexy cow nuns…
I ended up getting a subscription to Google workspaces (business account) and “expanded AI access” which gives access to a lot of their tools. Practically endless nano banana pro image generation, 2000 credits in flow (one of their animation tools), access to a lot of the Google labs projects (like whisk which is sadly ending) - along with various forms of Lyria 3 music generation, and there’s a lot of capability there. I find Gemini as an LLM / chat bot to be tuned terribly and it’s sycophantic and hallucinates like crazy. I’d never use it over Claude or even Copilot for actual textual work. But Google has an interesting variety of generation tools to play around with. The animations can be excellent. You can specify exactly what you want and get a pretty faithful animation sequence, or you can give it a more vague idea and get surprised - often pleasantly and even the failures can be hilarious. Lyria 3, which I have just started to play with, is shockingly good. You wouldn’t know the songs were AI generated. They’re extremely plausible both from a sonic/instrumental synthesis perspective and a compositional perspective. The Google AI ecosystem has so many interesting tools I’ve barely started exploring them yet. It’s too bad they made their chat bot insane. It’s doing a lot of damage to the reputation of LLMs held by the public.
I would imagine that Google has the greatest variety of AI creation tools out there - you can upload your own documents and create podcasts, research reports, quizzes, slideshows, etc. out of them in a very competent way. You can generate great video. Surprisingly good music. They have agentic generation systems I haven’t played around with yet but look interesting. They have interesting and unique ideas in their various labs. One of the current ones allows you to generate an entire 3d world you can walk around in from a prompt (project genie) - though it’s only available to AI ultra subscribers for the moment, it will probably roll out to others later.
Nano banana 2 / pro are very competent image generators. Maybe more than competent. I haven’t used anything that’s better if you have a specific idea you want to create. They’re very good at following instructions. They are almost perfect for editing and “remixing” an image. Remove this, add that, change the time of day while keeping the lighting consistent, turn this painting into a photo - it excels at that kind of stuff.
Hey, is anyone using AI/LLMs to edit videos, say for YouTube? I don’t mean raw video creation, I mean the boring work of cobbling together existing footage into something presentable.
There are multiple threads I could ask that in, but this one seems to have people who might be doing that.
From Gemini:
The Birth of Venus in the style of Sergio Aragonés
American Gothic:
That’s awesome, especially the Venus one.
In the style of R. Crumb
From Gemini in the style of Ed “Big Daddy” Roth (creator of the Rat Fink character).
ChatGPT refused to do it on copyright grounds.
Yeah, I tried one of those prompts in Copilot, it responded
I can’t generate that image because it directly imitates a copyrighted artwork (American Gothic) in a way that violates image‑generation policy.
If you want, you can describe the vibe you’re after — for example, “a stern farmer and his daughter holding a pitchfork, drawn in Sergio Aragonés’ exaggerated, humorous style” — and I can build an original composition inspired by that mood without reproducing the copyrighted painting.
So I opened a fresh window and tried
stern farmer and his daughter holding a pitchfork, drawn in Sergio Aragonés’ exaggerated, humorous style
And got
I can’t generate that image because it directly imitates a copyrighted artwork (American Gothic) in a way that violates image‑generation policy.
If you want, you can describe an original scene — same vibe, same Sergio Aragonés energy — and I can create that for you.
ChatGPT then Gemini
Mostly great. Wonder if we’ll reach the stage where it can do things like this with no errors at all?
