|
Dumb amateur question, but is there an AI art platform where you provide the seed images instead of it scrubbing the internet? E.G. if you were a professional photographer with lots of source material, or just wanted to upload your phone's camera roll. I'm new to all of this, but I've got several thousand image assets I've created over decades and thought it might be cool to gently caress around with them.
|
#
?
Dec 19, 2023 17:30
|
|
|
|
| # ? May 30, 2024 19:19 |
|
|
smoobles posted:Dumb amateur question, but is there an AI art platform where you provide the seed images instead of it scrubbing the internet? E.G. if you were a professional photographer with lots of source material, or just wanted to upload your phone's camera roll. Do you mean, like, instead of using SDXL or whatever *.safetensors file as your model, you would prefer to use a model that has been trained only on a catalogue of your own selection? Yes, that's a thing with Stable Diffusion. I haven't done it myself so I'm not the person to tell you how it's done, but I'm sure other posters in this thread can give you a nudge in the proper direction. I'd be interesting in training a model on the image folder I've been adding random poo poo to for 15 years just to see what happens myself.
|
|
#
?
Dec 19, 2023 17:36
|
|
|
        
|
|
#
?
Dec 20, 2023 03:31
|
|
|
smoobles posted:Dumb amateur question, but is there an AI art platform where you provide the seed images instead of it scrubbing the internet? E.G. if you were a professional photographer with lots of source material, or just wanted to upload your phone's camera roll. That's...not how this works. These are frozen-in-time models with the noise data inside them, there isn't actually art, images or otherwise, inside the models. You can not extract an image out of the file directly. Moreover, they don't learn in real time, new releases may contain new data from newer sources, but Stable Diffusion XL is one static model, MidJourney v5 is one static model, and whatever DallE is at the current time you use it is a static model. They might upgrade them, DallE stealthily, but they are functionally entirely new models. They must be actively trained, then actively posted. As of right now a fluid, real-time learning image model isn't a thing. You can train a model to your style though, as XYZAB says, but it's something you do yourself by feeding Stable Diffusion images using various techniques, either Dreambooth (the old way but powerful) or LORAs (new way, a bit more restrictive but far faster and smaller). Then when you use these things you've trained and ask it for your style it goes "ah, I know those pictures, like this" and will output something with the visual styles of what you trained it. It's a bit involved of a process but is essentially what https://civitai.com/ is completely comprised of. If you've ever had a friend really good at doodling, and you showed them a picture of like, anime or something and said "make one like this but a Hank Hill as a banana saying 'doughnut bobby' lmao" that's what these models are doing, which is why they look the way they do. You will never get a 1:1 image out of one of these models no matter how hard you train it to make that image, because it's inherent to how they work. They do not have a perfect memory, which is why they work at all.
|
|
#
?
Dec 20, 2023 03:48
|
|
|
I think I have a pretty good handle on what Automatic1111 covers, now I'm curious how I might run a batch of images through with the same settings. I was thinking passing some abstract Blender animations through SDXL could look cool. This is just a rotating array of wireframe cubes close to the camera:  Plus some other things I made. I'm modeling primitives and rendering a depth map and then using that for a control net. It's very addictive and I can easily spend hours fiddling with it.    
moist banana bread fucked around with this message at 04:35 on Dec 20, 2023 |
|
#
?
Dec 20, 2023 03:57
|
|
|
smoobles posted:Dumb amateur question, but is there an AI art platform where you provide the seed images instead of it scrubbing the internet? E.G. if you were a professional photographer with lots of source material, or just wanted to upload your phone's camera roll. If you want something easy and online, getimg.ai lets you create quick and easy dreambooth models which you can train on your own art or photos, or you can create custom styles from collections of images, or you can train it on an object or person. The cost of using it goes beyond the free tier, though.
|
|
#
?
Dec 20, 2023 07:12
|
|
|
I wish Instagram videos could be inlined here. Not mine but I just found this use of Suno in the wild, with a crazy video to boot. Edit: Oh my god yes. XYZAB fucked around with this message at 08:07 on Dec 20, 2023 |
|
#
?
Dec 20, 2023 07:37
|
|
|
Seth Peckinsiff Worst Nightmare Wildest Dream 
|
|
#
?
Dec 20, 2023 08:39
|
|
|
I made a AI generated album, really leaning into the glitchiness of the AI generation. My enjoyment might be based on having listened to a couple of thousand of AI generated tracks, so it might actually be terrible. I recommend headphones: https://soundcloud.com/brown-moses/...=social_sharing
|
|
#
?
Dec 20, 2023 13:54
|
|
|
Ask ChatGPT for dinosaurs with buccal fat enhancement, this is what you get.
|
|
#
?
Dec 20, 2023 17:36
|
|
|
XYZAB posted:Ask ChatGPT for dinosaurs with buccal fat enhancement, this is what you get. look at the size of the balls of this one, I think they rival the pig's!
|
|
#
?
Dec 20, 2023 17:43
|
|
|
XYZAB posted:Ask ChatGPT for dinosaurs with buccal fat enhancement, this is what you get. I remember this guy from a Rick & Morty episode
|
|
#
?
Dec 20, 2023 17:45
|
|
|
it's a good time of year to make your own wrapping paper                      
|
|
#
?
Dec 20, 2023 18:25
|
|
|
you probably want to be doing that in Stable Diffusion with "tile" checked so the edges match and also up scaling it so it prints better.
|
|
#
?
Dec 20, 2023 18:33
|
|
|
Midjourney v6 is out with "text and color controls" plus other improvements. Gonna give er a whirl
|
|
#
?
Dec 21, 2023 18:56
|
|
|
AARD VARKMAN posted:Midjourney v6 is out with "text and color controls" plus other improvements. Gonna give er a whirl The only issue is that now that it's so good at following prompts, it's not as randomly "creative" as 5.2 and if you need that stuff you might have to drop down or get super specific with prompts. Like I had a prompt that made a character surrounded by weird animatronic monsters, and now it not only doesn't do that but makes her too young. So I'll have to downgrade or rework it. I also had a prompt that made cool dark sci-fi images and now everything is too colorful, so the same issues apply.
|
|
#
?
Dec 21, 2023 19:02
|
|
|
Idle thought but are there any AI art generators that will do gore? Haven't seen one of them yet.
|
|
#
?
Dec 21, 2023 20:43
|
|
|
Mj v6: Lmao  Still struggles with the blue red panda.
|
|
#
?
Dec 21, 2023 20:47
|
|
|
Stupid_Sexy_Flander posted:Idle thought but are there any AI art generators that will do gore? Haven't seen one of them yet. People in this thread have posted some rather gory (but funny) images of Seinfeld characters exploding from within. I think they got around it by using "spaghetti sauce" instead of blood.
|
|
#
?
Dec 21, 2023 20:49
|
|
|
Ah, gonna have to check out the earlier posts. I was mainly wondering because I have had no luck trying to do a battle damaged t800 or even a wolverine with some damage healing. I did manage a somewhat decent metallo though, but I'm guessing it's because I didn't try to do the fleshy bits.
|
|
#
?
Dec 21, 2023 20:55
|
|
|
Bing will give gore without prompting at random and sometimes it's just not filtering it for some reason. You can always just load stable diffusion as well that will make anything. e: sounds like you're only going for like this level? I think the prompting here was a demon clown stopping a speeding car and suplexing it causing an explosion. these are probably 6-9 months old.  
pixaal fucked around with this message at 20:58 on Dec 21, 2023 |
|
#
?
Dec 21, 2023 20:56
|
|
|
hydroceramics posted:
What if you ask for a blue lesser panda? Instead of using red panda which has another color in the name.   loving around photograph of a blue Lesser Panda also known as a red panda this individual is remarkable in that it has bright blue markings instead.  doesn't like the word red
|
|
#
?
Dec 21, 2023 21:53
|
|
|
Choose your flavour  
GaryFrenchkiss fucked around with this message at 03:16 on Dec 22, 2023 |
|
#
?
Dec 22, 2023 03:13
|
|
|
Stupid_Sexy_Flander posted:Idle thought but are there any AI art generators that will do gore? Haven't seen one of them yet. Stable Diffusion? It's completely uncensored. V6 MidJourney is pretty great, mostly 1:1 context-wise with Dall-E now. Observe: 1982 LP of "Wheygazer" a new wave 4 piece band similar to A Flock Of Seagulls. One of them wears a leather jacket, another one has a pink mohawk, the third one is holding a slice of pizza, and the last one is a small dog Midjourney v6:  Dall-E 3 back in October when it was first released, exact same prompt 
KakerMix fucked around with this message at 03:24 on Dec 22, 2023 |
|
#
?
Dec 22, 2023 03:14
|
|
|
GaryFrenchkiss posted:Choose your flavour lol I loved the centipede one so much I tried it myself and got this;  Taste the crunch lol.
|
|
#
?
Dec 22, 2023 03:39
|
|
|
Taste the crunch
|
|
#
?
Dec 22, 2023 04:13
|
|
|
aint a teenager no more, my fathers dead and only skill i know is being a ninja      Merry Xmas!
|
|
#
?
Dec 22, 2023 04:19
|
|
|
Midjourney V6 is freaking me out. Feels like we are 95% of the way to AI photos being indistinguishable from real photos.
|
|
#
?
Dec 22, 2023 05:47
|
|
|
null_pointer posted:People in this thread have posted some rather gory (but funny) images of Seinfeld characters exploding from within. I think they got around it by using "spaghetti sauce" instead of blood.
|
|
#
?
Dec 22, 2023 10:30
|
|
|
smoobles posted:Midjourney V6 is freaking me out. Feels like we are 95% of the way to AI photos being indistinguishable from real photos. what are you talking about this is my real grandpa 
|
|
#
?
Dec 22, 2023 10:34
|
|
|
*gutteral voice* is this the best you can do, robin?
|
|
#
?
Dec 22, 2023 13:29
|
|
|
Maybe I've just been tossing the wrong things MJ6's way, but so far I'm not really impressed. It's left out major elements from my prompts, misinterprets elements of prompts that Bing has no issues with, seems to have filtered out 90% of celebrities, still can't handle nuances between different paint techniques, doesn't have inpainting yet, is far worse than Bing at dinosaus, and can't make a bazooka to save its life.
|
|
#
?
Dec 22, 2023 13:46
|
|
|
feedmyleg posted:Maybe I've just been tossing the wrong things MJ6's way, but so far I'm not really impressed. It's left out major elements from my prompts, misinterprets elements of prompts that Bing has no issues with, seems to have filtered out 90% of celebrities, still can't handle nuances between different paint techniques, doesn't have inpainting yet, is far worse than Bing at dinosaus, and can't make a bazooka to save its life. AFAIK the tech hasn't changed fundamentally since it was introduced, it's just been incremental improvements in training, efficiency, prompt logistics, etc. So they're all still interpreting your prompt as a bunch of vibes to be tossed in a blender. Some blenders do well with dinos and others do well with bazookas.
|
|
#
?
Dec 22, 2023 15:07
|
|
|
https://files.catbox.moe/vplh97.mp4 I made a 7 minute long progressive rock song, lots of editing together of audio clips in audacity to avoid the generation loss that happens from continuing songs past around 3 generations. It's not perfect, nothing from suno.ai ever is, but there's lots of nice melodies in this one. Lyrics are mine.
|
|
#
?
Dec 22, 2023 18:05
|
|
|
|
|
#
?
Dec 22, 2023 20:22
|
|
|
            
|
|
#
?
Dec 22, 2023 22:01
|
|
|
Has anyone been loving around with pika.art or ai.invideo.io?
|
|
#
?
Dec 23, 2023 01:36
|
|
|
Suno can now also be used from within Microsofts Co-Pilot via plugin. https://blogs.bing.com/search/december-2023/Turn-your-ideas-into-songs-with-Suno-on-Microsoft-Copilot Craze has died down a bit here, but I assume this will also give you some additional credits to burn through.
|
|
#
?
Dec 23, 2023 16:41
|
|
|
I hadn't played with Suno in a bit, hadn't realized it had a Full Song option now. Here's a full movie end credits song I just tossed together. This poo poo is bananas. https://cdn1.suno.ai/0ef81cc4-0259-45a3-877e-3a5d4b43a6c7.mp4 Roman fucked around with this message at 18:59 on Dec 23, 2023 |
|
#
?
Dec 23, 2023 18:54
|
|
|
|
| # ? May 30, 2024 19:19 |
|
|
feedmyleg posted:Maybe I've just been tossing the wrong things MJ6's way, but so far I'm not really impressed. It's left out major elements from my prompts, misinterprets elements of prompts that Bing has no issues with, seems to have filtered out 90% of celebrities, still can't handle nuances between different paint techniques, doesn't have inpainting yet, is far worse than Bing at dinosaus, and can't make a bazooka to save its life.  1976 ektachrome photo of an anthropomorphic cute mushroom dressed as a police officer writing a ticket for a speeding grasshopper  Child's drawing of Cherenkov radiation glowing in the core of a nuclear reactor pool  A photograph of a badly-made hot dog on a plate, with a speech bubble of it saying "Interesting if true"  A large rack of ribs being interviewed on The Tonight Show with Johnny Carson circa 1985 It's the ability to get context from it. It's a lot different than the previous versions of Midjourney where you were more at the mercy of the machine, now it wants a bit more simple guidance.
|
|
#
?
Dec 23, 2023 21:51
|
|