Just playing around with AI some more

[Dr.Minimizer]

A few AI images from something called PicLumen, if anyone else wants to try it...















This stuff just keeps getting better all the time.

[AlecFair]

All look great. What prompts were you using for the sizes and model?

[Dr.Minimizer]

The handhelds were: An extreme close-up high angle view of a gigantic hand holding a tiny sexy female Caucasian cheerleader with short black hair wearing a white low-cut lace-up leotard, black leather belt, black leather collar, white leather gloves, fishnet stockings, and shiny white go-go boots. The cheerleader is cowering in the hand, looking up with a terrified expression.

The one on the scroll: An extreme close-up high angle view of a gigantic spell scroll (titled "SHRINK SPELL") upon which stands a tiny sexy female Caucasian witch with short black hair wearing a black extremely low-cut lace-up bodysuit, long silver cape, silver leather belt, black leather collar, silver leather gloves, fishnet stockings, and shiny silver over-the-knee lace-up boots. The witch has her hands on her hips and is looking up in dismay.

The other ones were this or slight variations: An extreme close-up low angle side view of a gigantic male sorcerer's face looking down at a gigantic medieval desk, upon which stands a tiny sexy female Caucasian witch with short black hair wearing a black extremely low-cut lace-up tonga-style bodysuit, silver leather belt, black leather collar, black leather opera gloves, dark sheer pantyhose, and shiny black over-the-knee lace-up boots. The tiny witch, viewed from behind, has her arms out and legs open as she gasps in dismay at the gigantic sorcerer leering down at her.

This particular AI doesn't have too many filters that block out terms like "extremely low-cut" or "fishnet stockings" nor does it mind when I specify the girl has her legs open or is cowering in the hand. However it will still block nudity and blur out NSFW images if you manage to get it to produce one. It's also extremely likely to produce malformed hands, extra fingers, etc. but you can just keep trying and it doesn't seem to have much of a limit on how many you can make. I did all these with a free account using my throwaway Dr. Minimizer gmail to log in (where I don't care how much spam I wind up getting).

Here are a few more that were made with similar prompts:

[Prof Sai]

Nice! Cute outfits, and just the "not too real" look that I like.

[Some Wanderer]

These looks good, and it wouldn't be a Minimizer piece if somebody wasn't wearing a leotard. I may mess around with that program later.

[Dr.Minimizer]

And boots. Don't forget those. Here's another of the more toy-like creations. I had better luck with realistic looking tinies when I defined the larger background items first. Describe the girl first, and she winds up looking more like a plastic doll.

[Dr.Minimizer]

Here's a few I made with Leonardo AI...I really like this first one, although I'm not really sure what's going on with that stapler.


The lighting is a bit too dark on this one, but you get the idea. She's going to find that antidote eventually...


Is the wizard trying to make a doll come to life, or turn a real person into a doll....?


One of the more realistic low angle rear views I've been able to make so far....


You really do need to watch out for things labeled "reducing formula."


I wasn't trying for this, but she almost looks like a little Classic Trek science officer in that outfit.


Where am I? How did everything get so big?

[Firewall]

I love these. I wonder what the A.I.M. trio would look like in AI format.

[Dr.Minimizer]

I love these. I wonder what the A.I.M. trio would look like in AI format.

Let me get back to you on that, but for now...today's theme is Beach Fun...or, what happens when you don't carefully read the instructions on your brand new bottle of "Reducing Tan Oil"...(all images made using PicLumen)

The tiny ladies are available in a variety of sizes and outfit colors for your enjoyment! (By the way, it's supposed to say "Reducing Tanning Oil, Lose Weight While You Tan" but apparently AI cannot simply transfer things in quotes directly onto your scene, it has to translate and corrupt the words and I don't feel like fixing them, sorry.)

[Sumguy14]

loving all of the beach images. Really well done.

[Dr.Minimizer]

I love these. I wonder what the A.I.M. trio would look like in AI format.

I took a crack at this but it's proving very difficult. Even when I describe the image very precisely, the image generators I've been using keep giving me stupid results. For example if I give it this prompt (from the part in the story where the ladies are about to go out exploring together for the first time):

• Extreme close up high angle photorealistic view of a gigantic office desktop with a gigantic phone, gigantic pen and gigantic notepad with the words "MINI CHEERLEADERS" written on it. Standing nearby are three minuscule beautiful photorealistic adult women, all approximately the same height. The first woman, in the center, is a Caucasian with short blonde hair wearing a skintight black catsuit and white over-the-knee boots, standing with hands on hips. The second woman, on the left, is a timid, fearful, fair skinned redhead with hair tied back in a ponytail, wearing red hot pants, a dark blue crop top, and black combat boots. The third woman, on the right, is a sexy Latina with long black hair who is standing with arms crossed and scowling angrily, wearing a gray long-sleeved bodysuit, blue knee-high boots and blue gloves. Render the women as photorealistically as possible.

The AI will, 75% of the time, give me four women, picking seemingly one of them at random and duplicating her on the image. About one time in five there are even more women, sometimes 5 or 6. Additionally, even though I specify that all three are the same height, about half the time Caitlin is depicted about twice as tall as the other two, despite there being nothing there in the prompt that suggests this. And of course the AI frequently changes around the outfits, sometimes depicting them all as cheerleaders (even though I am quite specific about their attire, it still focuses on that word). And of course they are almost always shown as plastic toys, because they are small, therefore I guess they are supposed to be dolls, even though I say "photorealistic" everywhere. AI IS SO FRUSTRATING TO WORK WITH.

Anyway, that being said, here are a few marginal successes using PicLumen and Leonardo. Don't even get me started with ChatGPT. It's utterly useless. We'll start with a PicLumen shot that shows you what I mean by making them look like plastic, and with Caitlin being too tall compared to the others.



PicLumen has the right idea most of the time but they still look too much like dolls.



PicLumen also has several different generation modes, so I tried one that produces the highest quality images. It totally ignored my size commands and just about everything I said about the outfits, but actually I thought the AIM girls looked pretty good nonetheless. Too bad this mode doesn't allow rescaling, apparently.



I tried one more time to get PicLumen to cooperate using its Flux Schnell renderer, but this is the best it could do. PicLumen doesn't seem to have much of a limit on how many free uses a day I can get out of it, while the others all have some kind of credits that run out after a while; that's why I like to use it when I can.



Then I went over to Leonardo, which normally can make only a few images a day, but right now they have a new mode called "Flux State" that you can try quite a number of times. Here's my first decent attempt. Not really what I said with regards to the outfits, but they got the hair and size right.



This one is a bit better, but Leonardo's standard mode makes one image at a time and this is about the middle of the third batch. The rest have missing pieces, too many characters or sizes wrong. Also on this one Tania doesn't exactly look like a Latina...



Still looking a bit plastic, but I like the determination in this one.



This one is probably the closest thing it made to what I asked for. The outfits are almost exactly what I said and the ladies are all posed correctly without any unused material. (I gave up asking for Mini Cheerleaders and switched to just having it say AIM on the notepad)



Sometimes the AI changes things up unexpectedly. This one dressed the ladies in more normal looking attire but made them smaller than any other image, so even though they're kind of blurry it's still a cool scene.



Now here's one where it REALLY made them look plastic, but it came up with this style all on its own.



Finally this is the most photo-realistic shot I saw in about 100 tries. Not exactly what I wanted, but close. I'll keep trying and get back to you.

[bobvan2]

There ya go!

[Dr.Minimizer]

Now let me share with you my experience with ChatGPT. This is why I say it's useless. I'll post a few screenshots of my conversation with it and you can see why working with it is the most frustrating thing you can imagine.

First off I described each of the three AIM girls to the AI and managed to get it this point, where it could show me the three of them in cheerleader costumes. I was trying to get it to depict their first moments on the desk when they wake up to find that they've shrunk. To do that I had to first define the ladies and this is what it came up with.



After getting this image I proceeded to attempt to put them on the desk, and you can see from here how the conversation goes. It starts out well enough, but things go rapidly downhill.

















And that's when I beat my head against the desk and gave up.

[Firewall]

I love these. I wonder what the A.I.M. trio would look like in AI format.

This one is a bit better, but Leonardo's standard mode makes one image at a time and this is about the middle of the third batch. The rest have missing pieces, too many characters or sizes wrong. Also on this one Tania doesn't exactly look like a Latina...




This one is probably the closest thing it made to what I asked for. The outfits are almost exactly what I said and the ladies are all posed correctly without any unused material. (I gave up asking for Mini Cheerleaders and switched to just having it say AIM on the notepad)




Now here's one where it REALLY made them look plastic, but it came up with this style all on its own.

These three are the ones I like the most. Glad to know I'm not the only one that wants to strangle ChatGPT at times. I had a story that I'm saving for next Xmas but am somewhat developing the spin-off(Which went off rails with me not putting much scenes of shrinking in it) where I wanted the main female lead's art from and what do I get? The art is either her chibified or looking like a doll.

[Dr.Minimizer]

I've been trying to use AI to make videos for a while, but haven't had much luck. Hailuo (Minimax) seems to be the one of choice over on DA, but I can only get it to give me the same basic thing (overhead view of tiny woman next to coffee cup or other ordinary object) without any interaction or anything. There are plenty of those available if you want to see them. I'm having more luck with Pixverse, although you only get a few free videos a day and buying credits is expensive. Here are a couple of examples of what it can do...

[Hand-Holder]

Oh I feel you bro, even with just stories, I can have endless discussions with GPT, sometimes even insult it...

[Nropyub]

[/quote]

Oh my god. This stuff makes me wonder what we will have in 20 years.

Can you do it again at 3-5 inches tall? I’d like to see that.

You know.

For science.

[Dr.Minimizer]

Sadly, the AI doesn't seem to listen to orders regarding exact sizes. If I say 3 inches or 1 inch or anything like that, it just ignores it completely. However, as I recently learned, there is a "transition" feature available. Give it as starting image and then a finishing image, and it tries to fill in the gaps, based on your prompt. Nearly every time I try to make a shrink happen, the AI adds some kind of weird transition (usually the girl jumps up and down and maybe there's a flash or burst of smoke, even though I didn't ask for anything like that, and even if I specifically tell it not to do it). Here are a couple of examples...

The first one didn't include any jumping or transitioning so it's probably the cleanest of the three.



Here she starts off with arms crossed, and ends just standing there, but the AI added this weird hop right in the middle of the scene for no apparent reason.



In this one I was going for a defiant heroine by starting with a raised fist, but she instead goes straight into looking happy and excited, and again does the shrink-hop. Maybe it's a new dance craze...?

[Dr.Minimizer]

Here's another one...Pixverse did great here, it even added the reflections in the jars on its own (and the sloshing liquid too) without being prompted to do so.

[Dr.Minimizer]

One more then I'll move on to something else, probably...

[Dr.Minimizer]

Okay enough videos, here's some more images of the AIM girls as rendered by AI...this time looking more like spies. They have been sent to steal a secret formula but someone spotted them...!

All of these were done with Leonardo on a free account. Example prompt:

SpoilerShow

Brightly lit overhead photorealistic tabletop scene. A huge jar labeled "SECRET FORMULA" dominates the background. There are exactly three characters in this scene, no more and no less. In the foreground, a minuscule adult female Caucasian spy with short blonde hair, dressed in a black skintight catsuit, black thigh boots and black leather gloves, looks up in dismay as a giant hand menaces her. Standing nearby is another minuscule adult female Latina spy, slightly taller than the first, with long black hair, dressed in a black leotard, black knee-high boots and black leather gloves, looking up at the hand angrily. Slightly behind her stands another minuscule adult female Caucasian spy, slightly shorter than the first, with red hair tied back in a ponytail, dressed in a black skintight bodysuit, white thigh boots and white leather gloves, looking up at the hand fearfully.

Note: When I used this prompt, Leonardo gave me the wrong number of women (usually 4 but sometimes 2, 5, or even 6 about 60-65% of the time. Someone please explain to me how a prompt that so specifically names precisely three women, can give other results. Oh and if you take out the "exactly three characters" line, the chance of getting 4 or more goes up to about 80%. sigh.

This first one is about as close to perfect as I think the AI will ever get. I think it took about 20+ renders of 4+ images at a time before I got just this one. All the characters look correct and even the words are spelled right on the jar.



I like everything about this one except Randi's different size boots.



Proving once again that the AI can't do text correctly. Also their faces are a bit unfinished in this one, but I do like Randi's boots this time.



This one's more cartoon-like, but at least they have intense expressions...Tania is scowling like she always does, and Randi is doing a "I've got eyes on you" pose which is pretty cool. Sometimes you get neat stuff you didn't ask for and can never repeat.



I told it to make an overhead shot this time. I don't know why it gave Caitlin a mask, but it looks pretty cool, actually (and so does the view).



Finally, in this one the AI completely ignored my instructions on the outfits, but who cares, I think they look great!

[QTips]

I see in the last one that the evil Dr Sixfinger is behind all this!

[Dr.Minimizer]

We're in real trouble if the AI ever figures out how to make hands look realistic.

[Elana]

[Dr.Minimizer]

LOL good one.

I subscribed to Leonardo for a month ($12, guess I'll go to one less movie) and tried out their image conversions and enhancement tools for a bit. Here are a few examples. I'm sure there will be more to come later.

These first three are just basic images based on various scenes I've create in the past with collaging. As you can see there is still a sort of plastic-like appearance to the tiny figure, but it's getting better.







Here's a rehash of that last AIM pic I posted yesterday. I uploaded that image and told it to make it look more realistic. Damn, the ladies look incredibly hot in those outfits...although getting around is probably super difficult in those heels.



Finally here's about the best looking one I've seen so far. The generator actually obeyed my instructions and made each of the girls look exactly like they were described. Plus it did a fantastic job on their costumes. I upscaled it too for higher detail.