Sociopath Community

Trying to get AI to refine these images (rather than upscale) so that it looks more natural... but Mike isn't popular enough to have a strong reflection when parsing. It keeps not looking anything like him by the time it's done: 

For future context when I ask myself wtf inspired trying to make this image work: 

16:49 MichaelLongstocking: hey look it's
16:50 MichaelLongstocking: Where's the beef? 🤷‍♀️
16:50 MichaelLongstocking: Quality is our recipe.
16:50 MichaelLongstocking: 🍔
16:50 MichaelLongstocking: It's better here.
16:51 MichaelLongstocking: Do what tastes right.
16:51 MichaelLongstocking: That's right
16:51 MichaelLongstocking: It's way better than fast food. It's Michael Longstocking's.
16:52 MichaelLongstocking: You know when it's real.
16:52 MichaelLongstocking: Now that's better.
16:52 MichaelLongstocking: We got you.
16:52 MichaelLongstocking: Choose wisely. Choose Michael Longstocking's.
16:53 MichaelLongstocking: Gotta Be Michael Longstocking's.
16:54 MichaelLongstocking: **vomits the force of 1000 cheeseburgers**🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔🍔 It is finished, my vrillords. It is finished
16:55 MichaelLongstocking: Archonnaught believe my I's
16:56 MichaelLongstocking: but it has finally happened. Now please, with powerful force and quickened speed, impale my anus with thy tentacles at once
16:57 MichaelLongstocking: Make haste, four my hole is deeply yearning for thy rod and thy staff
16:58 MichaelLongstocking: they comfort me. And soo do cheeseburgers. 🤭🤭🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣😂😂😂😂😂😂😂😂😂😂😂😂😭😭
16:58 MichaelLongstocking: soo yeahh
16:58 MichaelLongstocking: k byee
16:59 MichaelLongstocking: just joking about everything that I have ever communicated to you all in all forms, and so it is thank you, and so it is thank you
16:59 MichaelLongstocking: bye
17:47 MichaelLongstocking: https://youtu.be/ISrv9312Y0I?si=H2w_p11QgPjIKU_O
17:47 MichaelLongstocking: "I'm a long guy" 🤭🤭🤣🤣
17:49 MichaelLongstocking: ooh my god the first time that I saw this I thought that he was serious but he was probably just joking around like he really just said that he is a "long guy" when he was asked how he f**ks essentially
17:49 MichaelLongstocking: ooh. my god. I am soo happy that I found this again 🤭🤭😘😘
17:50 MichaelLongstocking: I'm a long guy. 🤭🤭😘😘
17:50 MichaelLongstocking: soo yeahh
17:50 MichaelLongstocking: k byee
17:50 MichaelLongstocking: just joking about everything that I have ever communicated to you all in all forms, and so it is thank you, and so it is thank you
17:50 MichaelLongstocking: bye

I give up, it refuses to capture his face shape. Here's the attempts: 

AI induced stupidity (and co-stalking)

Lower the strength for better results, or else it's not going to look like Mike. 

Your method is shit though, use another Ai. What you do is swap the face on Pipi longstocking herself, and if you want the body crop it and add it to Mike's body. Though I don't think you can crop an image well from the looks of it. 

Golden_Eagle said: 

Lower the strength for better results, or else it's not going to look like Mike. 

The program starts at 80 strength, and I even tried refining it to look for his face from his segment 'The Night Shift'. The more literal render styles were even further off, the closest to his shape came from more cartoony render styles. Lowered to 50 was the sweet spot for getting smoother results between the two zones, but they were still wrong even though they got closer and it began to show more artifacts at the edges from using a transparent image. Adjusting to 40 and below kept making it resemble the original source picture too much, which included the sharpness distortions between him versus his costume.

Doing it with a background made for more bleed between the subject and the environment, which with how it was already struggling to render his face correctly made that become even more challenging. Now that I know this about using transparent images with low strength AI rendering, I'll opt for flood filling the background layer next time instead so that it can be removed in post rather than prior. 

Even with the distortions made from Dezgo when using an image with a transparent background, if it got at least one right I could then sloppily paste it onto a background and have AI blend the two. I didn't want to bother using a tablet to digitally paint it, the point was to have it done as something low effort with some quick Google Image pics. 

Your method is shit though, use another Ai. What you do is swap the face on Pipi longstocking herself, and if you want the body crop it and add it to Mike's body. 

The problem with that is that it'll look too much like a fusion of the two, rather than looking like Mike wearing a dress and wig. I got plenty of fusion renders:

 



Pippi Longstockings has a lot more parsing strength than the nobody that is Mike Majlak. A lot of what had the above picture results not work out was over it trying to grab Pippi's face from the prompt, I had to add more negative prompts and add some positive ones about facial hair to start to refine the results further. 

With AI's Image-To-Image method you can do really rushed work as it's foundation image, like super rushed bullshit just to establish the scene normally, if the subject matter is popular enough with enough photos on the web to make a decent collage of them. The problem here, as I saw with trying to prompt just his name, is that the AI isn't trained enough to use him in the model. By comparison you can put in a name like Donald Trump and it will give you a very strongly trained render. 

I'd be more likely to get this to work if I found at least three celebrities that resemble features of him, then render it like 100 times until it gets the correct shape on it's own. 

Though I don't think you can crop an image well from the looks of it. 

The two source pics are their original crops and I was mostly slapping it together as a low effort joke originally; it didn't take more than a few minutes each. I used the base tolerance magic wand tool rather than going out of my way to lasso out the flaws or fine tune the selection with Selection Modifying, the intent at first was a shit post over that absurd hair being on a bald dude. 

Only once I went to Dezgo did it become about trying to make it look good, because it converts the process more into a guessing game of sorts with figuring out what prompt will make it the closest to the intended image. After so many guesses for trying to fix the face shape it stopped being fun, and I didn't feel like cleaning the artifacts out of the failures. 

Edit: Having felt socially pressured out of being too lazy I decided to put his face onto her body as a proof of concept as per your earlier pointer to try to demonstrate how unnatural it'd look, but as work went further into it over a period of around 10 minutes it continued to feel less and less worth it. That being said, it actually looks closer to one of the failed renders than I thought it did: 

 

Maybe I gave up too soon, with some artifact cleaning up, some eye correction with Liquify in Photoshop, and maybe eve-

 

Ugh it's closer but still shit. Back to giving up on this until goaded back into it. 

If you want to do this too to show how it's done, here's what I've been using purely through the browser: 

https://dezgo.com/
https://photopea.com/

I tend to have better luck when trying to make a fictional person, or when I work with actual celebrities or comic book characters in the prompts as resources (ex: Lex Luthor, Sarah Michelle Geller, Ghost Rider). Trying to make it work with what the AI sees as a nobody kept having it lose elements of the intended figure, which in this case was the only real goal that I had. 

For real, shit looks a lot better when going for fiction if not aiming for collages of popular faces. Here's some randos from my Download folder for d20 campaign imagery. As a concept fusion generator based on prompts as a guessing game it's pretty damn fun. 

Nah I kinda like Mike.

If you want to do it properly you'll want to deep fake it. 

First you install Pinokio which is a type of browser which allows us to run some Ai tools offline off of our own system.

The process is kinda finicky, as it needs to install some basic windows extensions. If it fails or freezes keep forcing it. I had to restart and I saw some components where checkmarked and had to continue installing the uncheckmarked until I was done.  

In Pinokio you'll find a list of goodies you can download and install. You want "Facefusion".

The images. Were dragged and dropped. Top is source, bottom is target. Here I took Wario's face and slapped it on Pipi. Notice how it captures the lighting, it knows the depth of that nose.

Works just like that. Your issue is lighting and quality integration. Facefusion will match the quality of the imagery, so if it's poor quality or grainy, the output will blend in. And of course the Ai has a freaky way of making the facial expressions work correctly. 

Under the preview image we see the Face Mask Types. I turned all 3 on which corrected the bangs, they were kinda getting cut off at first. That was the first time I ever had to do that, region and occlusion is usually not needed. There is more to this interface, if we scroll down. 

To the left side of the interface we see the Execution Providers. It's set to CPU which means it will use the CPU's processing power to perform the task. Since I had an AMD card installed it shows the dml option, which means it will use the GPU instead, which is faster and more powerful. If you have an Nvidia card, then Cuda will show up under Execution Providers.

For the Target Image.... We can totally put video there and it will swap any frame we scroll to, or we can batch out the video and it will swap the entire video. Under the target image we can see the output, for video. I also took the liberty of changing a value in one of facefusions files so I can do NSFW, cause why not.

It takes about 15 minutes to do 1 minute of video with my CPU. GPU is much faster I forget. If I wanted to rip out an entire movie I'd take advantage with the 27 GPU mining farm, which is excellent for machine learning and Ai tools in general. 

Only good if one were to say, do Ai stuff for money, like makinging daily content or swapping out entire movies where the customer gets to be the star. Though I imagine that might be illegal, but there can be a loophole if the customer owns the movie. 

.

At this point you can imagine how much easier it can be to crop out the head with the wig. But say you decided to do that, and you can't get the right lighting, and you probably won't. Bring it to Kling.

This is what Sora is supposed to be, and Sora isn't even out yet.

Yes it's overkill, but I noticed the community likes to feed it images, so you can prompt a new body, maybe even use another reference to combine it. Then if you wish make a video too.

Browse the community, check out the imagery, look at the prompts find the prompt guide.

https://klingai.com/

Youtube is a crystal ball, and will show you everything I just said better than I can write it. Facefusion.

As for Photoshop. I use CS6, which is now pretty old but it's the last standalone Photoshop, and it's still very good by today's standards. I'm not interested in posting a cropping tutorial though. 

Golden_Eagle said: 

Nah I kinda like Mike.

If you want to do it properly you'll want to deep fake it. 

First you install Pinokio which is a type of browser which allows us to run some Ai tools offline off of our own system.

The process is kinda finicky, as it needs to install some basic windows extensions. If it fails or freezes keep forcing it. I had to restart and I saw some components where checkmarked and had to continue installing the uncheckmarked until I was done.  

In Pinokio you'll find a list of goodies you can download and install. You want "Facefusion".

The images. Were dragged and dropped. Top is source, bottom is target. Here I took Wario's face and slapped it on Pipi. Notice how it captures the lighting, it knows the depth of that nose.

I appreciate your saying "I like Mike" didn't end up an excuse to do nothing, Facefusion looks really nice and now that I've seen it I plan to use this program. There's been many cases where I've created a face in Dezgo, but how it was made was from so many prompts and trials that luck became an element. With that 'Luck Face' captured a program like this could help mass produce said face into other scenes. 

That being said, it looks more like some kind of bizarre Jim-Spawn rather than looking like Jim in a wig and dress. Ideally, I'd like to be able to use AI Image Generation to have it look like "Mike in a Wig and Dress", rather than "Mike as a Girl", which requires preserving the blocky quality of Mike's body while the clothing is applied. 

I'd ideally need an Inpainting Tool that can add stuff on top convincingly, but Dezgo's Inpainting Tools are really shoddy and primitive to such a point that it looks worse than if I did it myself in a hurry. When Inpainting is fixed I'm about to go off with D20 pics. 

Works just like that. Your issue is lighting and quality integration. Facefusion will match the quality of the imagery, so if it's poor quality or grainy, the output will blend in.

That problem is only present at lower strengths, and only over my mistake of using a transparent image. Ideally... rather than blending the quality I'd rather it redraw the entire thing in one style so it doesn't look as blatantly collaged, and with current tools that approach tends to work better when done in a painterly style. 

I bet that'll change soon, really soon, and then these digital painter-like renders can be used as the foundation image when upgrading into a more realistic style. 

And of course the Ai has a freaky way of making the facial expressions work correctly. 

They still look a little Liquify-esque, but with the speed of it's output it's very impressive compared to a digital painter spending far more time on it. 

For the Target Image.... We can totally put video there and it will swap any frame we scroll to, or we can batch out the video and it will swap the entire video. Under the target image we can see the output, for video. I also took the liberty of changing a value in one of facefusions files so I can do NSFW, cause why not.

Using a video as the target image? I'd seen something like that for Deepfakes but not for producing a still image. 

That's awesome. 

It takes about 15 minutes to do 1 minute of video with my CPU. GPU is much faster I forget. If I wanted to rip out an entire movie I'd take advantage with the 27 GPU mining farm, which is excellent for machine learning and Ai tools in general. 

Only good if one were to say, do Ai stuff for money, like makinging daily content or swapping out entire movies where the customer gets to be the star. Though I imagine that might be illegal, but there can be a loophole if the customer owns the movie. 

.

At this point you can imagine how much easier it can be to crop out the head with the wig. But say you decided to do that, and you can't get the right lighting, and you probably won't. Bring it to Kling.

Lighting when painted on always looks wrong, either fuzzy or glassy instead of dithering proper fades. Past a point I've had to accept that no number of layer transparency and filters has been enough to make it look convincing by-hand. 

...this naturally has me enjoying AI, things that once took a good deal of time investment it can practically vomit out like it was a mistake. 

This is what Sora is supposed to be, and Sora isn't even out yet.

Yes it's overkill, but I noticed the community likes to feed it images, so you can prompt a new body, maybe even use another reference to combine it. Then if you wish make a video too.

Browse the community, check out the imagery, look at the prompts find the prompt guide.

https://klingai.com/

I'll check it. 

Youtube is a crystal ball, and will show you everything I just said better than I can write it. Facefusion.

As for Photoshop. I use CS6, which is now pretty old but it's the last standalone Photoshop, and it's still very good by today's standards. I'm not interested in posting a cropping tutorial though. 

I'm poor and use a free version that resembles CS4, which overtime I realized that Photopea was similar enough to the retroware that I moved onto using just that for quick projects.