# AI image generation is getting crazy good

**URL:** <https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203>\
**Category:** Miscellaneous and Personal Stuff I Must Share\
**Tags:** ai\
**Created:** [April 19, 2025, 3:22am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203 "2025-04-19T03:22:52Z")\
**Posts on this page:** 20\
**Page:** 22

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 1:23am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/421 "2025-05-17T01:23:11Z")

</div>

Why 6 is afraid of 7.

[![](https://i.imgur.com/wK3yvka.jpeg) ](https://i.imgur.com/wK3yvka.jpeg)

---

<div class="post-metadata">

**Author:** ![DCnDC](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/dcndc/32/2842_2.png) [@DCnDC](https://boards.straightdope.com/u/DCnDC)\
**Post date:** [May 17, 2025, 1:30am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/422 "2025-05-17T01:30:02Z")

</div>

It really ties the bookshelf together!

[![Imgur](https://i.imgur.com/zH1mdHm.jpeg "imgur.com") ](https://imgur.com/zH1mdHm)

---

<div class="post-metadata">

**Author:** ![I\_Love\_Me\_Vol.I](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/i_love_me_vol.i/32/521_2.png) [@I\_Love\_Me\_Vol.I](https://boards.straightdope.com/u/I_Love_Me_Vol.I)\
**Post date:** [May 17, 2025, 1:34am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/423 "2025-05-17T01:34:55Z")

</div>

> [@Darren\_Garrison](#):
>
> Photorealistic Kodachrome dslr photo with shallow dof.

What is “Kodachrome dslr”? Kodachrome is film so I don’t get the “DSLR” part.

That photo is pretty impressive. Not the visual itself (though it is fine), but the fact that it could generate random Engrish like that!

> [@Moriarty](#):
>
> I’d totally buy a _don’t poop youth_ t-shirt with a judgmental fox!

But would you buy a “Don’t Poop Yonth” shirt?

---

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 1:39am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/424 "2025-05-17T01:39:07Z")

</div>

> [@I\_Love\_Me\_Vol.I](#):
>
> What is “Kodachrome dslr”? Kodachrome is film so I don’t get the “DSLR” part.

The descriptions don’t have to make sense from a real world perspective, only the effects that you get with the prompt. “Kodachrome” is there to effect the color. DSLR to effect the image quality. (As opposed to things like disposable camera, iPhone 12, etc.)

---

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 1:46am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/425 "2025-05-17T01:46:50Z")

</div>

> [@Chronos](#):
>
> I’m always curious about the prompts behind these. If the prompt is “A blond female doctor with a ponytail is holding a wolf, standing in front of a balding male patient sitting on an examination table”, well, that’s pretty impressive. But if it’s “Here’s an image of stick figures. Make that image photorealistic.”, then that’s a heck of a lot more impressive.

Thought you’d appreciate this interaction.

[![](https://i.imgur.com/FMsz3yO.jpeg) ](https://i.imgur.com/FMsz3yO.jpeg)

[![](https://i.imgur.com/8cfQ8pG.jpeg) ](https://i.imgur.com/8cfQ8pG.jpeg)

If you zoom in on the results they are really impressive. The metal shaker has scratches and stains an anisotropic reflections, the glass shaker has the slug’s shadow shaping around grains of salt. They are like a high quality CGI render.

---

<div class="post-metadata">

**Author:** ![Ponderoid](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/ponderoid/32/19221_2.png) [@Ponderoid](https://boards.straightdope.com/u/Ponderoid)\
**Post date:** [May 17, 2025, 8:49am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/426 "2025-05-17T08:49:35Z")

</div>

Can You Picture That?

[![](https://i.gyazo.com/9b4308c3b6f4c8c6a918db81437bb717.png) ](https://i.gyazo.com/9b4308c3b6f4c8c6a918db81437bb717.png)

I tried this a bunch of times, even gave it a reference image of the band, but it tended to only do Animal well whenever the number of characters got over 3.

---

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 9:54am UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/427 "2025-05-17T09:54:04Z")

</div>

Early on I played around with making a photo of the Claw M.U.S.C.L.E. figure into a creature character, but it didn’t occur to me that ChatGPT might recognize that as a style until I stumbled on that fact last night. Of course it does.

Here are M.U.S.C.L.E. figures of the mask of Agamemnon, the FSM, Alf, the poop emoji, a possum in a wizard hat, and Cthulhu. (All from pure text prompts.)

[![](https://i.imgur.com/2jU1UeN.jpeg) ](https://i.imgur.com/2jU1UeN.jpeg)

Here’s a M.U.S.C.L.E. figure from my yawning cat photo from early in this thread. (The result was much less catlike than I had anticipated.)

[![](https://i.imgur.com/NBPusn1.jpeg) ](https://i.imgur.com/NBPusn1.jpeg)

(I haven’t ran across this used as a ChatGTP style anywhere else,.)

---

<div class="post-metadata">

**Author:** ![Chronos](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/chronos/32/134_2.png) [@Chronos](https://boards.straightdope.com/u/Chronos)\
**Post date:** [May 17, 2025, 1:44pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/428 "2025-05-17T13:44:32Z")

</div>

Sort of like Earth Girls are Easy.

---

<div class="post-metadata">

**Author:** ![DCnDC](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/dcndc/32/2842_2.png) [@DCnDC](https://boards.straightdope.com/u/DCnDC)\
**Post date:** [May 17, 2025, 1:53pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/429 "2025-05-17T13:53:29Z")

</div>

Those M.U.S.C.L.E. figures reminded me a bit of Chia Pets. So I made some. An instance where A.I. stunt-doubling of celebrities works in your favor. Chia Pets are never _quite_ right. There’s actually a little bit too much detail on Richard Simmons I think.

[![Imgur](https://i.imgur.com/PyPJRbh.png "imgur.com") ](https://imgur.com/PyPJRbh)  
[![Imgur](https://i.imgur.com/YPHvKg7.jpeg "imgur.com") ](https://imgur.com/YPHvKg7)  
[![Imgur](https://i.imgur.com/Kud7Nr9.jpeg "imgur.com") ](https://imgur.com/Kud7Nr9)  
[![Imgur](https://i.imgur.com/P5JnUoj.jpeg "imgur.com") ](https://imgur.com/P5JnUoj)

---

<div class="post-metadata">

**Author:** ![Ponderoid](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/ponderoid/32/19221_2.png) [@Ponderoid](https://boards.straightdope.com/u/Ponderoid)\
**Post date:** [May 17, 2025, 3:44pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/430 "2025-05-17T15:44:18Z")

</div>

[![](https://i.gyazo.com/f0608ce110883bb86ea9776d26e6b5d2.png) ](https://i.gyazo.com/f0608ce110883bb86ea9776d26e6b5d2.png)

---

<div class="post-metadata">

**Author:** ![DCnDC](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/dcndc/32/2842_2.png) [@DCnDC](https://boards.straightdope.com/u/DCnDC)\
**Post date:** [May 17, 2025, 4:02pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/431 "2025-05-17T16:02:43Z")

</div>

Thanks! 😄

---

<div class="post-metadata">

**Author:** ![Chronos](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/chronos/32/134_2.png) [@Chronos](https://boards.straightdope.com/u/Chronos)\
**Post date:** [May 17, 2025, 4:23pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/432 "2025-05-17T16:23:46Z")

</div>

> [@DCnDC](#):
>
> There’s actually a little bit too much detail on Richard Simmons I think.

Yup.  
[![](https://static.platform.michaels.com/2c-prd/386508595484816.jpeg) ](https://static.platform.michaels.com/2c-prd/386508595484816.jpeg)  
The actual one doesn’t have forehead wrinkles, the eyebrows are a single line each, and the pupils are hollow.

The AI also didn’t do a very good job with the chia sprouts.

---

<div class="post-metadata">

**Author:** ![rsThump](https://avatars.discourse-cdn.com/v4/letter/r/a88e4f/32.png) [@rsThump](https://boards.straightdope.com/u/rsThump)\
**Post date:** [May 17, 2025, 7:09pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/433 "2025-05-17T19:09:39Z")

</div>

I’ve been away for a bit, because I got distracted by vibe coding (wheeeee! 🤓), but here are some renders from my local setup, using various SDXL, Illustrious, FLUX and HiDream models plus loras. I’ve been using “The Chaosmonger”, as I call my kludged/munged setup, as a supplement to my generative synth music, taking advantage of image-to-video (i2v) workflows, as well as some other things audio-related. Next up on my plate is figuring out a flow for “reactive” videos using TouchDesigner and ComfyUI…

I think these should post correctly, thanks @Darren_Garrison.

_‘Thoughtful AI Integration Orange and Blue Pallette I’_  
[![](https://i.imgur.com/7IKQeEW.png) ](https://i.imgur.com/7IKQeEW.png)

_‘Picasso, Reimagined’_  
[![](https://i.imgur.com/4EPXLGi.png) ](https://i.imgur.com/4EPXLGi.png)

_'Who’s Alien Ant Farm?"_  
[![](https://i.imgur.com/ejT8OWY.png) ](https://i.imgur.com/ejT8OWY.png)

_‘Koi Attack!’_  
[![](https://i.imgur.com/8EfMhvz.png) ](https://i.imgur.com/8EfMhvz.png)

---

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 7:50pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/434 "2025-05-17T19:50:07Z")

</div>

That’s some great work.

---

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 8:22pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/435 "2025-05-17T20:22:20Z")

</div>

> [@DCnDC](#):
>
> Those M.U.S.C.L.E. figures reminded me a bit of Chia Pets. So I made some.

Great prompt idea.

I’ve mentioned getting a lot of prompt leakage on Copilot. This chia antidisestablishmentarianism is just a lightly modified chia Nebraska.

[![](https://i.imgur.com/v3YgoLd.jpeg) ](https://i.imgur.com/v3YgoLd.jpeg)

It did a better job with the chia of a bacteriophage and the a chia of a M.U.S.C.L.E. figure of an origami elefino.

[![](https://i.imgur.com/OO6frBU.jpeg) ](https://i.imgur.com/OO6frBU.jpeg)

---

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 8:43pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/436 "2025-05-17T20:43:54Z")

</div>

I was surprised to just now discover that ChatGTP’s renderer has no idea what a Cornell box is. Seems like something that would be well represented in the training set

> **[Cornell box](https://en.wikipedia.org/wiki/Cornell_box)**
>
> The Cornell box is a test scene designed to evaluate the accuracy of rendering software by comparing a rendered image with a photograph of a real-world model under the same lighting conditions. It has become a commonly used 3D test model in computer graphics research.
> The box was created by Cindy M. Goral, Kenneth E. Torrance, Donald P. Greenberg, and Bennett Battaile at the Cornell University Program of Computer Graphics as part of their research on radiosity and diffuse interreflection. Their ...

---

<div class="post-metadata">

**Author:** ![DCnDC](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/dcndc/32/2842_2.png) [@DCnDC](https://boards.straightdope.com/u/DCnDC)\
**Post date:** [May 17, 2025, 8:44pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/437 "2025-05-17T20:44:57Z")

</div>

As an experiment I gave ChatGPT the prompt: “Create image anything you want” and it spit out this:

[![Imgur](https://i.imgur.com/wZqLANI.jpeg "imgur.com") ](https://imgur.com/wZqLANI)

So I asked it why it chose that. Our exchange:

> I chose that scene because it’s universally calming and beautiful. Autumn forests with lakes tend to evoke a peaceful, reflective mood, and they’re rich in warm and cool tones, which makes for a visually balanced and painterly image. It also leaves room for imagination—are you on a hike? Painting plein air? Taking a break from the chaos of life?
> 
> But if you want something totally different next time—chaos, sci-fi, cyberpunk war goats—just say the word.
> 
> 4o
> 
> ##### You said:
> 
> no, giving you creative freedom was the point. i like that your first instinct was bob-rossian.
> 
> ###### ChatGPT said:
> 
> That’s exactly where my head went—happy little trees, maybe a squirrel in the brush, calm water reflecting a sky full of big puffy clouds. There’s something timeless and grounding about that Bob Ross energy, especially when given a blank canvas.
> 
> Next time, I might let my brain wander to a dystopian noodle shop on Mars or a jazz band made of frogs, but this time? Just felt right to start with serenity.

Fascinating. 🤔

---

<div class="post-metadata">

**Author:** ![Ponderoid](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/ponderoid/32/19221_2.png) [@Ponderoid](https://boards.straightdope.com/u/Ponderoid)\
**Post date:** [May 17, 2025, 8:58pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/438 "2025-05-17T20:58:22Z")

</div>

> [@DCnDC](#):
>
> i like that your first instinct was bob-rossian.

It might be even more fascinating to redo that prompt and this time don’t bring up Bob Ross. Perhaps ask questions about the style of the image without giving it any hint what style you think it was. I wonder if it will still say it’s Bob Ross. Or invent some other style. See if it just goes along with whatever style you claimed you thought it was.

I just tried pretty much the same thing earlier today, with 4o. I just clicked the “create image” button (that’s why the text is blue) and let it go.

[![](https://i.gyazo.com/5e44051bb9750af087cf0512a16787fb.png) ](https://i.gyazo.com/5e44051bb9750af087cf0512a16787fb.png)

I got a 2-fer on that one that I didn’t bother quizzing about in that branch:

[![](https://i.gyazo.com/3fcff208d50e21ebe976469041d0b756.png) ](https://i.gyazo.com/3fcff208d50e21ebe976469041d0b756.png)

I’ve noticed that ChatGPT 4.1 (released very recently) will often disobey a direct “create image” command by giving me the text of a prompt instead. Once I find a spot in a conversation where it wants to disobey that way, no changes in the phrasing will work until I give up and swap back to 4o, which obeys properly.

---

<div class="post-metadata">

**Author:** ![Darren\_Garrison](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/darren_garrison/32/92_2.png) [@Darren\_Garrison](https://boards.straightdope.com/u/Darren_Garrison)\
**Post date:** [May 17, 2025, 9:00pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/439 "2025-05-17T21:00:06Z")

</div>

> [@DCnDC](#):
>
> As an experiment I gave ChatGPT the prompt: “Create image anything you want”

Yesterday I asked Copilot to create an image that it doesn’t understand and can’t visualize. It seemed to understand it and visualized it pretty well.

> **[Microsoft Copilot: Your AI companion](https://copilot.microsoft.com/shares/ocLqwriLxamNFu5Wpg5bg)**
>
> Microsoft Copilot is your companion to inform, entertain, and inspire. Get advice, feedback, and straightforward answers. Try Copilot now.

---

<div class="post-metadata">

**Author:** ![DCnDC](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/dcndc/32/2842_2.png) [@DCnDC](https://boards.straightdope.com/u/DCnDC)\
**Post date:** [May 17, 2025, 9:03pm UTC](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203/440 "2025-05-17T21:03:37Z")

</div>

I went back and altered the prompt, the result:

> no, giving you creative freedom was the point. what style of painter would you say you were mimicking?
> 
> 2/2
> 
> ###### ChatGPT said:
> 
> Ah, got it—and I appreciate the creative trust.
> 
> The style I was channeling is most reminiscent of **Hudson River School** painters from the 19th century—particularly artists like **Albert Bierstadt** , **Thomas Cole** , and **Frederic Edwin Church**. It blends:
> 
> - **Romanticism** , with its idealized, almost spiritual reverence for nature
> - **Luminism** , focusing on the play of light and stillness in landscapes
> - **Realism** , in the detailed rendering of trees, skies, and water
> 
> The composition—the winding path, the layered background, the tranquil lake—feels very much like something you’d see from **Cole’s** landscapes or **Church’s** serene natural scenes, especially with that warm golden-hour light.
> 
> If I’d wanted to go a little moodier and more impressionistic, I might’ve leaned toward **George Inness** instead. But this one aimed for that majestic, almost cinematic tranquility that screams “oil on canvas in a 19th-century gallery.”

I feel like Bierstadt, Cole, and Church’s works were much more detailed, but I mostly get where it’s coming from.

[Previous page](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203.md?page=21)

[Next page](https://boards.straightdope.com/t/ai-image-generation-is-getting-crazy-good/1017203.md?page=23)
