AI art is genuinely stupid!

[Heavy_Metal_Hound]

Recently, during a bout of boredom, while browsing faceache I saw what I assume to be a sponsered post. I cant find it again but. It depicted a man & child holding a hunting rifle next to a tv. The text on the post spoke about how AI has forgotten all about the video game "Duck Hunt". This has inspired me to play about with an AI art generator, specifically "Dream" by Wombo. My prompt was "Child playing "Duck Hunt" on his Nintendo" while I refuse to legitimise this stupidity by actually pressing the " publish" button. Here is my screenshot:

Anyway...I thought that it would be funny to share with guys, and I invite those of you with an AI art generator to make something even more hilariously inaccurate...using the exact same prompt that I have used.

Oh and for those who dont know " Duck Hunt" was a 1984 Nintendo Game, in Which you shoot ducks.

 

[ZJ68]

I use Nightcafe's Stable Diffusion for fun, and it seems to do reasonably well with the latest versions. It still sometimes has issues with hands, how human joints work, the number of limbs most people have, and doesn't appear to know the difference between "catgirl" and "cat lady". Some things it does quite well though. I'm just a hack who types in idiotic prompts like "Cthulhu fights the Kool-Aid Man by Norman Rockwell", but people who know what they're doing with prompts do some amazing stuff.

I couldn't use your exact prompt because it prohibits the use of the word "child", so I substituted "boy" and got this:

 

[Heavy_Metal_Hound]

I use Nightcafe's Stable Diffusion for fun, and it seems to do reasonably well with the latest versions. It still sometimes has issues with hands, how human joints work, the number of limbs most people have, and doesn't appear to know the difference between "catgirl" and "cat lady". Some things it does quite well though. I'm just a hack who types in idiotic prompts like "Cthulhu fights the Kool-Aid Man by Norman Rockwell", but people who know what they're doing with prompts do some amazing stuff.

I couldn't use your exact prompt because it prohibits the use of the word "child", so I substituted "boy" and got this:

View attachment 1754912

sorry about the slow reply, I seem to have posted this while drunk only to forget all about it . This is actually rather good, it just shows how much of a roll of the dice ai art is. Or perhaps the importance of using a good quality ai art generator. Still I have on occasion got a good picture out of my rather dim witted ai art generator. My profile picture for example.

 

[emwolf]

I use Nightcafe's Stable Diffusion for fun, and it seems to do reasonably well with the latest versions. It still sometimes has issues with hands, how human joints work, the number of limbs most people have, and doesn't appear to know the difference between "catgirl" and "cat lady". Some things it does quite well though. I'm just a hack who types in idiotic prompts like "Cthulhu fights the Kool-Aid Man by Norman Rockwell", but people who know what they're doing with prompts do some amazing stuff.

I couldn't use your exact prompt because it prohibits the use of the word "child", so I substituted "boy" and got this:

View attachment 1754912

I REALLY want to see Cthulhu fights the Kool-Aid Man by Norman Rockwell.

 

[Whisky]

Now I want to play duck hunt. It’s been probably 35yrs since I’ve played.

 

[mcee_sharp]

Now I want to play duck hunt. It’s been probably 35yrs since I’ve played.

"Kids don't get outside no more and hunt ducks" - Grumpy 80's Grandpa

 

[m949]

I'll play too; the results are just too hilarious.

This is actually the third AI image creation for this prompt, the others were "un-post-able"; the AI does have an issue with the anatomy.

Here the duck does the hunting with some kind of fishing rod, the boy has some sort of partially assembled hand gun, and both of them seem to "hunt" the Nintendo.

This AI Model runs locally on my machine, and it is fast. We are talking about 20 seconds to create this image. I use Draw Things as a client software and various AI Models which run locally. This particular image was created with the following parameters:

Model: SDXL Turbo (8-bit) - Stable Diffusion XL Turbo
LoRA: SDXL Offset 1.0 (SDXL Base) - Low-order Rank Adaptation; applies separately trained additive models to a Stable Diffusion base model.
Image to Image (the image is generated based on previous generated image; less faithful to text prompt)
Steps: 100 (more steps require more computing power and therefore more time for image generation; Max. number of steps in Draw Things is 150)
Sampler: Euler Ancestral (SD samplers are methods for collecting samples for target distribution.They work by generating a random image and then repeatedly subtracting the predicted noise, resulting in a clean image.)

I'm myself rather new to this, but I have gotten some good results. The important thing to keep in mind is that the prompt should be as detailed as possible. In this example I stuck to the exact prompt, as this was the challenge here. However, I found that additional instructions are helpful: 4K, UHD, volumetric lighting, sharp focus, etc. are all good additions to any AI image generation prompt. Also, the AI would almost never get the image right in the first iteration. Fine tuning the prompt, choosing the right AI Model, even the sampler choice, and the number of steps would generate very different images. It takes maybe around 20 to 30 images to get acceptable results which are close to the intended prompt.

Remember we are not talking about real Artificial Intelligence here, we are a long shot away from that.

 

[ZJ68]

I REALLY want to see Cthulhu fights the Kool-Aid Man by Norman Rockwell.

It turned out, weird. Like it drove him mad when he saw it, which is very Cthulhu. Or he was high as balls. I would not say it's the best example of what AI art can do.

 

[emwolf]

It turned out, weird. Like it drove him mad when he saw it, which is very Cthulhu. Or he was high as balls. I would not say it's the best example of what AI art can do.

View attachment 1795580

Definitely odd, but also very interesting. I don't think it will make the cover of the Saturday Evening Post.

 

[T Bone]

Here the duck does the hunting with some kind of fishing rod, the boy has some sort of partially assembled hand gun, and both of them seem to "hunt" the Nintendo.

That looks like a harpoon maybe? And does that boy have 2 right hands?

 

[nlife]

Now I want to play duck hunt. It’s been probably 35yrs since I’ve played.

I was 9 when duck hunt came out. Damn that laughing dog!

 

[ZJ68]

Definitely odd, but also very interesting. I don't think it will make the cover of the Saturday Evening Post.

I think the Kool-Aid Man part throws it. The AI definitely knows who Cthulhu is, but almost all of the published Kool-Aid Man pics on the site are nightmare fuel.

 

[Eben Stone]

I use Nightcafe's Stable Diffusion for fun, and it seems to do reasonably well with the latest versions. It still sometimes has issues with hands, how human joints work, the number of limbs most people have, and doesn't appear to know the difference between "catgirl" and "cat lady". Some things it does quite well though. I'm just a hack who types in idiotic prompts like "Cthulhu fights the Kool-Aid Man by Norman Rockwell", but people who know what they're doing with prompts do some amazing stuff.

I couldn't use your exact prompt because it prohibits the use of the word "child", so I substituted "boy" and got this:

View attachment 1754912

I thought giving it more context might be helpful. So I tried "child playing nintendo duckhunt" and got this:

 

[Eben Stone]

I tried the same "child playing nintendo duckhunt" using Microsoft's Copilot and it generated four images. Each with their own level of stupidity, but I though this one was at least interesting:

 

[Eben Stone]

I have a few credits left and apparently I have nothing better to do on my lunch break...

Microsoft Copilot

Spoiler: Context

real boy playing nintendo duckhunt videogame with pistol controller