this post was submitted on 15 Feb 2024
216 points (87.2% liked)

AI Art & Image Generation

2039 readers
1 users here now

A place to share images and art generated by artificial intelligence and similar tools.

Rules:

Recommended Communities:

founded 1 year ago
MODERATORS
 

As requested here.

Never got a result that actually matched the prompt but this one was too funny and odd not to share.

If this is not the right place to post AI mishaps please LMK and I'll take it somewhere else.

top 15 comments
sorted by: hot top controversial new old
[–] PinkOwls@feddit.de 44 points 9 months ago (2 children)
[–] fsxylo@sh.itjust.works 10 points 9 months ago

God damnit.

[–] MacNCheezus 4 points 9 months ago
[–] silas@programming.dev 12 points 9 months ago (1 children)
[–] Gork@lemm.ee 3 points 9 months ago

More like a cobbler swift, but that doesn't flow very well.

[–] pageflight@lemmy.world 6 points 9 months ago

The way the two shoes are weirdly merged together is a nice touch.

[–] Steamymoomilk@sh.itjust.works 5 points 9 months ago

Skibbity swift?

[–] Sludgehammer@lemmy.world 2 points 9 months ago (1 children)

Which image generator is that?

[–] MacNCheezus 4 points 9 months ago (1 children)

This was done using Stable Cascade, the new experimental model from Stable Diffusion.

[–] Usernameblankface@lemmy.world 1 points 9 months ago (1 children)

I checked the comments wondering how you got away with Taylor Swift in a generated image. Ah, Stable Diffusion makes sense for that

[–] MacNCheezus 1 points 9 months ago (1 children)

Yeah I tried it on Bing initially but it seems they have blocked all prompts containing Taylor Swift.

Not even sure why, they already have a nudity filter so you shouldn’t be able to create any salacious images of her but apparently that wasn’t enough.

[–] Usernameblankface@lemmy.world 1 points 9 months ago

Yeah, I was able to prompt for other celebrities with "race car livery sponsored by ______." But not her

[–] Mog_fanatic@lemmy.world 1 points 9 months ago (2 children)

Haha well that's better than my results. I was recently trying to get images of people with school busses instead of arms and I could not for the life of me get it. All I got was boring two armed people standing in front of busses... And they say AI will take over the world.

[–] s12@sopuli.xyz 1 points 8 months ago

I feel like for every thing someone gets that’s not what they wanted, someone else gets what that person wanted, when they wanted what that person got.

[–] MacNCheezus 1 points 9 months ago

Yeah it seems that for complex or unusual ideas, plain text prompting on its own isn’t enough yet.

Two solutions to this that I’ve had a bit of success with are

  1. Prompt for a less complex base image, then prompt for the additional components you need individually on a neutral background, and assemble the entire scene in photoshop

  2. Prompt for a base image and then use inpainting to add the additional elements