Now stable diffusion with Vladomatic can cut out objects from backgrounds.
00013.png (275.2 KiB) Viewed 19385 times
wc003.png (1.47 MiB) Viewed 19385 times
Re: Playing with a kindly AI
Posted: Mon May 01, 2023 1:34 pm
by CIA
When will get to the point where I can just tell the AGI app to create a 20 minute porn video of Lynda Carter as Wonder Woman in the famous Toys episode getting violated in all her holes by the evil robot?
When will get to the point where I can just tell the AGI app to create a 20 minute porn video of Lynda Carter as Wonder Woman in the famous Toys episode getting violated in all her holes by the evil robot?
Its going to be more an issue of how accurate you can describe what you exactly want, that's going to be the skill going forward.
Like authors, someone can just right a sex scene that is flat and sterile but a good author adds those elements, pacing etc that the reader likes but can't put their finger on.
Remember, all of us have notepad or word but few people can write a coherent story.
Re: Playing with a kindly AI
Posted: Tue May 02, 2023 3:40 pm
by CIA
Sure but because the results are instantaneously it won't be seconds to tweak the script.
Are there any programs out there that can do it now? You seem to be very well informed.
Sure but because the results are instantaneously it won't be seconds to tweak the script.
Are there any programs out there that can do it now? You seem to be very well informed.
Not video right now. That takes a lot of work. That pizza add for Pizza Hugs was all AI but it took a bit of work. A lot has to do with handing the AI images of what you kind of want to see so it can use those as templates. But I have no doubt in a year or two you'll see simply video on command.
All AI created, written, spoken.
Re: Playing with a kindly AI
Posted: Sat May 06, 2023 6:52 am
by Bladebur
Midjourney is getting really good with photo contents. It generates pretty convincing stills of faux 60s TV shows with almost no tweaks. It's a shame that it's the most prude and censored model ever, but for SFW images it's really easy to use.
Midjourney is getting really good with photo contents. It generates pretty convincing stills of faux 60s TV shows with almost no tweaks. It's a shame that it's the most prude and censored model ever, but for SFW images it's really easy to use.
I would tune in to watch this show
The costume from the top one and the mask and belt from the lower one.
Midjourney is getting really good with photo contents. It generates pretty convincing stills of faux 60s TV shows with almost no tweaks. It's a shame that it's the most prude and censored model ever, but for SFW images it's really easy to use.
I would tune in to watch this show
The costume from the top one and the mask and belt from the lower one.
I would love to see how the other characters would look populating this universe, Batgirl, Robin, Catwoman and the rogues gallery of villains.
Midjourney is getting really good with photo contents. It generates pretty convincing stills of faux 60s TV shows with almost no tweaks. It's a shame that it's the most prude and censored model ever, but for SFW images it's really easy to use.
That would be the shortest tutorial ever. All you have to do is go to the Midjourney discord and ask the bot "1966 TV show still, Batgirl wearing spandex". That's all. In fact, you don't need to go that far. One of the best images I got out from Midjourney was just asking for "Batgirl 1966". I didn't even write one celebrity name in any prompt, and still some images seem to resemble one or another. You can just ask for "1966 TV show still of Joker" (or Batman, or Catwoman) and you get great results.
(to be 100% honest, I had to ask for a 16:9 aspect ratio because it produces square images, but still).
It doesn't always work. It has no idea of who Robin or Enigma are, for example. Midjourney produces incredibly good results for some things, and complete disasters for others. It also lacks almost any kind of tool, so sometimes you get something that is *close* to what you want, but not quite, and there is no way in heaven you can make it right.
Stable Diffusion is a completely different beast which requires a tutorial even to make it run (there are plenty of those available though). But at least you can correct mistakes and guide the AI to put the characters in specific poses and a ton of other things. You can even make a 3D model of your image in Blender or similar and use it to guide it, although I am still learning that. Those are still early days and the UIs and tools are too immature and finicky.
Here is one Supergirl image that took quite a number of attempts and edits (and it's still not quite finished):
Re: Playing with a kindly AI
Posted: Wed May 10, 2023 5:27 am
by lionbadger
I did 2 in stable diffusion (Deliberate and Protogen models) of "Batgirl 1966" and one of "Batgirl 1969" (which gave me one with 4 arms, which I actually kinda like)
batgirl 1966 3.png (2.25 MiB) Viewed 18888 times
batgirl 1966 2.png (2.14 MiB) Viewed 18888 times
batgirl 1966.png (2.12 MiB) Viewed 18888 times
batgirl 1969.png (2.37 MiB) Viewed 18888 times
Re: Playing with a kindly AI
Posted: Wed May 10, 2023 8:22 am
by batgirl1969
LOVE THESE!!! I WISH I had 4 arms!!
Re: Playing with a kindly AI
Posted: Wed May 10, 2023 8:34 am
by sneakly
The four arms would make maintaining a secret identity a lot hard, but speed up shelving book a lot faster.
Some of the hands are pretty creepy. 4 to 6 digits. And Joker’s hands look like he stole them from Penguin
Re: Playing with a kindly AI
Posted: Wed May 10, 2023 9:20 am
by Bladebur
A-Zovya Photoreal model (with a LoRa and a lot of inpainting using ReV animated too).
This is a pretty cool render, and she is actually getting into some real danger....
Just add a spinning blade to human key duplicator...
Thanks! We're getting there, but this still takes a lot of work and is full of the typical AI bullshit (things which look cool at a distance, but are pure nonsense up close, like leather straps which go nowhere or just random chaos in the details).
This next one came up a bit better IMHO:
This is from that "ReV Animated" model. Obviously not photorealistic, but tends to give pretty artistic results.
Re: Playing with a kindly AI
Posted: Wed May 10, 2023 3:00 pm
by sneakly
Eventually, the AI will be more poseable. right know it knows how things look from studying thousands of pictures. As I understand it, It is making an image from what it has learned, not what it knows. It has seen how a hand is supposed to look, but doesn't have a good understanding of the anatomy of the hand. If you at the heroines in bondage towards the top of the page, their hands are just twisted pretzels, same with the Joker. The program doesn't know what to do. The Batgirl in light green has her hands on her hips, it is much easier because every picture of hands on hips will look similar.
Eventually, the AI will be more poseable. right know it knows how things look from studying thousands of pictures. As I understand it, It is making an image from what it has learned, not what it knows. It has seen how a hand is supposed to look, but doesn't have a good understanding of the anatomy of the hand. If you at the heroines in bondage towards the top of the page, their hands are just twisted pretzels, same with the Joker. The program doesn't know what to do. The Batgirl in light green has her hands on her hips, it is much easier because every picture of hands on hips will look similar.
Well, that's already there, it is already fully poseable, thanks to the ControlNet stuff. The way it works is you pose a stick skeleton using a specific tool (you can even draw the stick figure in an image if you want) and then an 'upper level' AI (which identifies human body parts) directs the normal AI to pose the characters in the way they fit that skeleton. It works surprisingly well, I used it for my last two images.
Eventually, the AI will be more poseable. right know it knows how things look from studying thousands of pictures. As I understand it, It is making an image from what it has learned, not what it knows. It has seen how a hand is supposed to look, but doesn't have a good understanding of the anatomy of the hand. If you at the heroines in bondage towards the top of the page, their hands are just twisted pretzels, same with the Joker. The program doesn't know what to do. The Batgirl in light green has her hands on her hips, it is much easier because every picture of hands on hips will look similar.
Well, that's already there, it is already fully poseable, thanks to the ControlNet stuff. The way it works is you pose a stick skeleton using a specific tool (you can even draw the stick figure in an image if you want) and then an 'upper level' AI (which identifies human body parts) directs the normal AI to pose the characters in the way they fit that skeleton. It works surprisingly well, I used it for my last two images.
Plus you can use depth maps to precisely control the pose. Its just you have to keep redrawing with sub pieces to get the correct face, mask etc.
Re: Playing with a kindly AI
Posted: Wed May 10, 2023 11:35 pm
by Supergirllegend
Take no notice of the labelling of these photos.
Sir_Frogdick_dressed_as_Supergirl_5071992d-871d-460e-ac8d-dfbfe97e07cc.png (4.61 MiB) Viewed 18703 times
Sir_Frogdick_coming_out_of_the_ocean_a999dc3f-7c60-4f63-ac2d-85d8a04aac62.png (6.21 MiB) Viewed 18703 times
Sir_Frogdick_sinking_in_mud_5e8f8ebb-a27e-47a8-9b6d-6b0a09392b52.png (6.28 MiB) Viewed 18703 times
Sir_Frogdick_dressed_in_wedding_dress_ce8fb295-0899-44c2-8d56-9c164bd29a59.png (4.66 MiB) Viewed 18703 times
Re: Playing with a kindly AI
Posted: Thu May 11, 2023 3:42 am
by Bladebur
Scaling AIs have also improved quite a bit. Here is a higher resolution version of the previous image with a few edits.
Re: Playing with a kindly AI
Posted: Thu May 11, 2023 8:44 am
by Bladebur
This is an experiment trying to do a few of the things the AI is terrible at. Due to the AI going nuts with upside down people and people inside things, the result has bad lighting and broken perspective.
Scaling AIs have also improved quite a bit. Here is a higher resolution version of the previous image with a few edits.
Wow! What were your commands to get you that amazing picture?
Re: Playing with a kindly AI
Posted: Thu May 11, 2023 1:55 pm
by Supergirllegend
IMG_3452.png (11.64 MiB) Viewed 18603 times
Re: Playing with a kindly AI
Posted: Thu May 11, 2023 2:21 pm
by sneakly
Apparently, it no longer takes much an enormous voice sampling to get AI to recreate a voice. Imagine having Yvonne Craig’s Batgirl doing an audio book version of Perils of Batgirl II. Or a face swap of Bondagecafe’s Perils of Batgirl with Yvonne’s face and voice and Julie Newmar. Or fully animated porn with no models at all.
I suspect things will get pretty freaky at some point.
Supergirl moody photo, Supergirl tied in a creepy torture chamber, big green fluorescent ropes, kneeling, hands up, (vibrator on crotch:1.1), cape, (big green fluorescent rock on neck:1.1), suffering
Negative prompt: multiple girls, wolf, animals, (low quality, worst quality:1.4),(bad_prompt_version2:0.8), (monochrome:1.1)
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2156558378, Size: 512x768, Model hash: 4199bcdd14, Model: revAnimated_v122, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 1, ControlNet: "preprocessor: depth_midas, model: control_v11f1p_sd15_depth [cfd03158], weight: 1, starting/ending: (0, 1), resize mode: Crop and Resize, pixel perfect: False, control mode: My prompt is more important, preprocessor params: (512, 0, 0)"
However, the result is still crappy and requires many (dozens!) of iterations with inpainting and image to image transformations, and even manual edits in Photoshop. It's a convoluted process. I'd say I've needed one hour or more per image or more. Seems like a lot of work, but still, I would need several days of work to get a similar image in 3D, and I'm not confident I could reach the same level of quality. Even more important, it's an addictive process because you're not just making an image, but discovering one.
Re: Playing with a kindly AI
Posted: Thu May 11, 2023 8:11 pm
by Mr. X
Yes that's the thing. The inpainting and constant rerunning to get the results means this is not push button. DanO at SHC has to render out depth map pics to get poses then run inpaint iterations over and over to get kind of close to what he wants. He literally has to model the character in poser, render for img2img then a depth map then run passes with inpaint.
Its fun but this tech is not there yet other than to make random pin ups that are just variations of the LORA you use.
Yes that's the thing. The inpainting and constant rerunning to get the results means this is not push button. DanO at SHC has to render out depth map pics to get poses then run inpaint iterations over and over to get kind of close to what he wants. He literally has to model the character in poser, render for img2img then a depth map then run passes with inpaint.
Its fun but this tech is not there yet other than to make random pin ups that are just variations of the LORA you use.
LORAs look promising, but I always get terrible results with them. It's like the AI learns the subject the LORA is about, but also a hundred other things you may not want. Like, you get the bondage, but also hotel rooms and plain lighting and manga fox girls with it.
Perhaps the answer is training a custom LORA with superheroine stuff. I don't think I have enough GPU for that, I think.
That workflow about rendering depth maps directly seems interesting.
LORAs look promising, but I always get terrible results with them. It's like the AI learns the subject the LORA is about, but also a hundred other things you may not want. Like, you get the bondage, but also hotel rooms and plain lighting and manga fox girls with it.
you're probably right about this, I also find that you have to be careful with the negative prompts as soon as you start adding Loras
that said, this "bimbo effect" one worked pretty well (no adjustment, hot from the oven)
sovereign girl 4.png (553.25 KiB) Viewed 18477 times
Re: Playing with a kindly AI
Posted: Fri May 12, 2023 10:00 am
by Bladebur
I'm not enjoying the photo models that much, compared to the artistic ones. They look impressive, but extra custom details tend to look painted or photoshopped.
Re: Playing with a kindly AI
Posted: Sat May 13, 2023 5:50 am
by Bladebur
This is addictive. Also, I figured out how to upload images to the forum (duh).
Re: Playing with a kindly AI
Posted: Sat May 13, 2023 2:24 pm
by sugarcoater
Just want to thank all of you posting your AI works here. It really is some impressive stuff! (I would like to learn how to use AI for images to enhance my written stories, but it seems like a steep learning curve.)
Re: Playing with a kindly AI
Posted: Sun May 14, 2023 5:56 pm
by Bladebur
I've upload edited versions of my AI images so far to my Deviantart page, if anyone is interested.
(and here is a new one, btw)
Re: Playing with a kindly AI
Posted: Mon May 15, 2023 7:36 am
by Supergirllegend
Sir_Frogdick_Supergirl_the_movie_1984_made_at_Pinewood_starring_85c6e65d-a86e-44e9-96b6-7faef0aa26d0.png (4.93 MiB) Viewed 18190 times
Sir_Frogdick_Supergirl_in_comic_book_form_photorealistic_Superg_17beff49-53b8-4d3a-a1a0-2fac4e0975b1.png (1.02 MiB) Viewed 18190 times
Sir_Frogdick_Supergirl_trapped_in_chains_Helen_Slater_as_Superg_fd30f677-4e95-44e8-b6f9-07a551a4ea5c.png (4.18 MiB) Viewed 18190 times
Sir_Frogdick_Supergirl_the_Movie_1984_Directed_by_Zack_Snyder_d_c542fc09-8e41-4b8f-bd5a-d96c60aa4315.png (1004.75 KiB) Viewed 18190 times
Re: Playing with a kindly AI
Posted: Mon May 15, 2023 7:38 am
by Supergirllegend
Sir_Frogdick_Supergirl_1984_Helen_Slater_in_the_1980s_dressed_i_b971d5f7-765d-4618-b3ce-741cf8cf897b.png (1017.43 KiB) Viewed 18189 times
Sir_Frogdick_Supergirl_1984_Helen_Slater_in_the_1980s_dressed_i_03b92e06-609c-48e1-aa5d-e58ff1603d7c.png (3.91 MiB) Viewed 18189 times
Sir_Frogdick_Supergirl_1984_Helen_Slater_in_the_1980s_dressed_i_4778fca5-4648-46bc-84a5-297990879db4.png (901.64 KiB) Viewed 18189 times
Re: Playing with a kindly AI
Posted: Wed May 17, 2023 9:36 am
by Bladebur
A couple of AI images I did with a ton of inpainting.
A couple of AI images I did with a ton of inpainting.
Yes the inpainting. I can hand you a pencil paper and you now have all the tools to draw a comic book but most people can't draw a comic book. Same with this. To get the pics that are more than just a pin up you still have to have the skills to start with a good base and inpaint and use depth maps to get poses, masks, gags etc.
A couple of AI images I did with a ton of inpainting.
Yes the inpainting. I can hand you a pencil paper and you now have all the tools to draw a comic book but most people can't draw a comic book. Same with this. To get the pics that are more than just a pin up you still have to have the skills to start with a good base and inpaint and use depth maps to get poses, masks, gags etc.
I can imagine Batgirl's wrists are cuffed and hooked to a stainless steel ass hook also, making her struggle futile but also thrilling and filling!!!!
The initial prompt for the Batgirl image is "masterpiece, high quality photo, best quality, 1girl, solo, basement, lying, hogtie, bound, bondage, bound arms, bound legs, arms behind back, looking at viewer, close-up, angry, glaring, open mouth, POV, bdsm, short hair, black hair, bodysuit, medium breasts, <lora:hogtie_v1:1>". The model was amIReal_V2 (but I used others for inpainting and relighting later). It's nothing special. I'm having better luck looking at standard prompts from CivitAI (mixing the words the specific models are well trained at), and then painting and inpainting other details on top.
Just to illustrate what I'm talking about, I've attached here the original image for the Batgirl one, and a number of the intermediate steps I got in the process (the prompts used for inpainting are in the names). It takes many, many attempts and iterations. I've generated at least 1000 images just for that Batgirl.
Yes the inpainting. I can hand you a pencil paper and you now have all the tools to draw a comic book but most people can't draw a comic book. Same with this. To get the pics that are more than just a pin up you still have to have the skills to start with a good base and inpaint and use depth maps to get poses, masks, gags etc.
This, 100%.
Depending of what you want, you can get a decent image out of the AI... But that's only if you want the most mundane images the AI can do, which is mostly about character pin ups. Almost anything else requires babysitting the AI and takes a ton of work.