c/ai-art-mashups

That tip about adding tiny details to the prompt was dead wrong for mashups

My buddy Rick who does digital art kept telling me to stack every prompt with super specific stuff like lighting and texture words. He said it makes AI pics look more real. So I tried it on a mashup of a lighthouse made out of waffles and syrup. I typed like 40 different detail words and the thing came out looking like a blurry mess with weird faces in the clouds. Then I stripped it down to just "lighthouse, waffle texture, syrup waves" and got the funniest result I've made all month. Turns out for mashups, the silly concept carries the image and extra detail words just confuse the model. Has anyone else found that shorter prompts work better when you're mixing weird things together? I'm curious if it's just me or if the AI gets too focused on making everything perfect.

The weird shift from 'funny accidents' to 'hyper-real product shots' in these mashups

I noticed it about 3 months ago. People stopped posting the glitchy stuff, the melting faces and extra limbs. Now it's all clean 4K renders of like a croissant shaped like a toaster, perfect lighting. Guess the free tools got too good, or everyone just migrated to the paid ones. Kinda miss the chaos honestly, anyone else feel that way?

My prompt turned a church into a giant taco truck and I cant stop laughing

So yesterday I was messing around with prompts for a wedding venue mashup and typed in baroque cathedral + food truck. The AI gave me this massive stone building with a taco menu board carved into the pillars and a huge sombrero on the bell tower. It looked so serious and dumb at the same time. My roommate saw it over my shoulder and now we've been feeding it more combos for 3 hours. Has anyone else gotten a blend that was way more comedic than the two things actually are?

A judge's 3 word note killed my silly prompt habit

I entered a weekly AI mashup contest with a 'seal riding a unicycle' but the judge wrote 'too literal' and it clicked. Now I use random word pairs like 'library + volcano' and let the model do the weird work. Has a one liner from a contest ever made you redo your whole style?

My buddy said add more slices to the prompt, I doubted him hard

He told me to just copy paste 'extra pizza slices' 14 times into the image prompt for our cat vs volcano mashup. I thought it was dumb but the results were hilarious, the cat ended up juggling like 30 pizzas while the volcano shot garlic bread. Anyone else get surprisingly good results from weird prompt repetition tricks?

Had to pick between a llama with a toaster body or toast with llama legs, picked the llama-toaster and it came out haunted

I stared at the prompt box for 20 minutes last Tuesday and chose the llama with a toaster body over the toast with llama legs, but the AI gave it glowing red eyes and butter pats for hooves, has anyone else gotten a mashup that just looks like it wants to hurt you?

The burger-cow mashup at State Fair made me dislike overdone prompts

I was walking the Minnesota State Fair last September and every other AI print had the same melting creature with a bun for a torso. It was funny the first 40 times, but after seeing that same prompt style at three different booths I got bored. Anyone else getting tired of the identical 'food animal' mashups and wanting more unexpected pairings like a lighthouse shaped like a stapler?

Prompt stacking vs prompt simplicity, which side actually makes better mashups?

I keep seeing folks swear by 12 word prompt stacks for fused images, but my last 30 tries showed the short 3 word prompts gave way funnier results like a toaster shaped like a seagull, so where do you land on clutter vs clean input?

Hit 500 AI mashups saved and my 87th one is still my favorite

I went through my folder the other night and realized I've got 500 AI mashup images saved from this group. Most are silly one-time jokes, but number 87, the flamingo shaped like a traffic cone, still cracks me up every time. It's funny how a random generation from 3 years ago beats all the fancy new stuff I've made since. Has anyone else kept a favorite old one that newer AI just can't top?

TIL adding a 'weight' prompt fixes the melting look in AI mashups

I kept getting these weird blob hands and squished faces when I'd mash two things together, like a raccoon chef in a frying pan. Then I tried adding 'heavy, dense, solid' to the prompt and it fixed 8 out of 10 renders. Anyone else found other one word prompts that stop the AI from doing that gooey morph thing?

That AI image of a mailbox with legs made me see my commute totally different

Last Tuesday I was scrolling through this sub during my lunch break and someone posted a render of a blue USPS mailbox walking on chicken legs through a Subway parking lot. It was dumb but I laughed for five minutes. Then I got in my car to drive home and hit the same traffic light on Main Street that always takes three cycles to turn green. For some reason I pictured that mailbox just strolling past the line of cars and suddenly the whole wait felt less annoying. Now I play a little game where I imagine every boring object on my route, like stop signs or fire hydrants, with some ridiculous animal body. It sounds childish but it actually made my 40 minute drive feel shorter. Has anyone else had a random AI mashup change how they see something normal in real life? I'm curious what images stuck with you.

Spent 3 hours chasing a prompt typo before I learned to zoom in

I was making a mashup of a lighthouse and a croissant and kept getting weird melted buildings. Realized after way too long that I'd typed 'lighhouse' and the AI just ran with it. Zooming in on the generated text was the only way I caught it, cost me an entire afternoon. Now I check every prompt letter by letter before hitting generate. Anyone else get burned by a simple misspelling turning the whole image sideways?

Anyone remember when mashups were just bad photoshops?

I was digging through my old hard drive and found some pics from 2012 where people would paste a celebrity head on a cat body or something. Now the AI does it all in 10 seconds, but honestly the old ones had more charm because someone actually sat there and messed with the lasso tool for 3 hours. My students today see a weird mashup and just shrug, they don't know the struggle of saving a .psd wrong and losing everything. Is it just me, or do the flaws in the old stuff make it funnier than the clean AI versions?

My cat looked like a burnt bagel in every mashup until I changed one prompt word

Ngl, I used to just type 'cat + food' and get these weird blob monsters. Last week I finally wrote 'orange tabby cat, photorealistic, wearing a sesame bagel costume' and got the funniest thing I've seen in months. Has anyone else found that adding one specific descriptor totally fixes the mashup?

That time an AI mashed up a fire hydrant and a french bulldog and I almost flooded my kitchen

Last week I was messing around with prompts and got a picture of a french bulldog that looked exactly like a fire hydrant, complete with a little spigot on its back. Showed it to my crew on the job site and one guy asked if I could hook it up to the main line. Has anyone else had an AI mashup that was a little too realistic?

My best AI mashup week got ruined by a prompt typo

Last Tuesday I was on fire making these crazy bulldog/pizza slice hybrids that looked hilarious. Got like 15 good ones in a row using the same seed number. Then on Thursday I accidentally typed "bulldpg" instead of bulldog and the AI gave me this weird melted dog thing that honestly looked like a crime scene. Took me 30 minutes to figure out why everything went wrong. Has anyone else had a whole batch of images ruined by one stupid typo?

GPT-4 makes way better cat pizza hybrids than Stable Diffusion does

I tried the same prompt on both models last night. I asked for "a golden retriever driving a tiny fire engine." GPT-4 gave me a dog with actual paws on a steering wheel and a little helmet. Stable Diffusion gave me a blob with four legs and a red smudge. I ran it 3 times each and GPT-4 won every time. Has anyone else noticed a big quality gap between these two for weird mashup stuff?

Just hit 100 AI mashups in my folder and it finally clicked why I keep doing this

I was scrolling through 87 images of a giraffe in a tuxedo eating spaghetti and realized the real magic is how AI accidentally captures that weird feeling of something being both wrong and perfect at the same time, has anyone else noticed their mashup library has a hidden theme?

Tried a beagle crossed with a hot air balloon in Midjourney

Prompt was 'beagle shaped hot air balloon'. Got a dog floating in the sky with a giant round stomach. No basket. Just a dog drifting. The face was pure confusion. Learned to specify 'basket attached to dog' or you get a flying beagle. Anyone else get weird anatomy from mashup prompts?

Found out Midjourney puts random text in images and now I can't unsee it

Spent 3 hours fighting with Midjourney for a giraffe in a tuxedo and finally found the secret

I kept getting weird blobs with long necks until I added 'formal event' to the prompt instead of just 'tuxedo'. Suddenly it started spitting out actual giraffes in suits. Has anyone else found a random phrase that just unlocks the AI's brain? lol

Hot take: Midjourney smokes DALL-E 3 for animal mashups

I spent all last weekend trying to get a good "pug dressed as a taco" image from both platforms. Midjourney nailed the fur texture and taco shell crunch look in under 2 minutes per try. DALL-E 3 kept giving me a weird blob that looked like a melted dog. I ran 15 versions on each and Midjourney won 12 out of 15 times for realism and funny details. Has anyone else tested these two side by side for specific mashup ideas like fruit animals or food vehicles?

Spent 4 hours trying to get AI to merge a toaster with a goldfish

I wanted one of those ridiculous mashup pics where a goldfish pops out of a toaster slot but the AI kept making the fish look burnt or putting the toaster inside a fishbowl. Took me way too long to figure out I had to describe it as "goldfish swimming out of a toaster, surrealist style" before it finally clicked. Anyone else have a mashup idea that took way more prompts than you expected?

PSA: Found out the hard way that Midjourney v6 handles food+object blends way better than DALL-E 3

I spent like 3 hours last weekend trying to get DALL-E 3 to make me a picture of a waffle that looks like a vintage car with headlights and tires. Every result came out looking like a blob of syrup with wheels barely attached. Got frustrated and threw the same prompt into Midjourney v6 out of spite. First try came back with this crispy golden waffle that actually had grill slats made of butter pats and the wheels were perfectly formed. The detail on the car doors as slices of bacon was insane. I think DALL-E just struggles when you ask it to fuse objects with completely different textures like metal and batter. Has anyone else noticed one AI handles mashups way cleaner than the others for specific combos?

Debate: Should AI mashups be funny first or technically good first?

I've been messing around with DALL-E and Midjourney lately, making weird combos like a giraffe wearing a tuxedo or a toaster shaped like a octopus. Some people in this group seem to care more about the prompt engineering and getting the lighting perfect, while others just want the funniest, most random result. I spent 3 hours last night tweaking my prompt for a T-Rex playing a banjo on a surfboard, and the best version had 5 fingers on its tiny arms but made me laugh. Which side are you on - do you sacrifice realism for a good chuckle, or does it bug you when the AI messes up obvious details?