/ais/ - Artificial Intelligence Tools

"In the Future, Entertainment will be Randomly Generated" - some Christian Zucchini

Index Catalog Archive Bottom Refresh
Name
Options
Subject
Message

Max message length: 12000

files

Max file size: 32.00 MB

Total max file size: 50.00 MB

Max files: 5

Supported file types: GIF, JPG, PNG, WebM, OGG, and more

E-mail
Password

(used to delete files and posts)

Misc

Remember to follow the Rules

The backup domains are located at 8chan.se and 8chan.cc. TOR access can be found here, or you can access the TOR portal from the clearnet at Redchannit 3.0.

US Election Thread

8chan.moe is a hobby project with no affiliation whatsoever to the administration of any other "8chan" site, past or present.

Use this board to discuss anything about the current and future state of AI and Neural Network based tools, and to creatively express yourself with them. For more technical questions, also consider visiting our sister board about Technology

AI Art / Stable Diffusion Thread #6 Anonymous 10/07/2023 (Sat) 22:18:55 Id: 243d38 No. 2422
Copied from previous OP (OP is now fairly outdated, needs to be refreshed at some point) Here's some guides to installing Stable Diffusion (if you have a NVIDIA GPU with 2GB+ VRAM). >Installing AUTOMATIC1111, the most feature-rich and commonly-used UI: https://rentry.org/voldy >Different UI with 1-click installation: https://github.com/cmdr2/stable-diffusion-ui >Installing various UIs in Docker: https://github.com/AbdBarho/stable-diffusion-webui-docker >AMD isn't well-supported. If you're running Linux there's a way to install AUTOMATIC1111 but otherwise features are very limited: Native: https://rentry.org/sd-nativeisekaitoo Docker: https://rentry.org/sdamd Onnx: https://rentry.org/ayymd-stable-diffustion-v1_4-guide To try it out without installing anything you can use a web service, though they are lacking in features. >Stable Horde is a free network of people donating their GPU time, how long it takes depends on how many people are using it. It supports negative prompts, seperate the positive and negative prompts with "###": https://aqualxx.github.io/stable-ui/ >Dreamstudio requires making an account and gives you around 200 free images before asking you to pay: https://beta.dreamstudio.ai/ Img2img: https://huggingface.co/spaces/huggingface/diffuse-the-rest | https://dezgo.com/image2image Ippainting: https://huggingface.co/spaces/fffiloni/stable-diffusion-inpainting | https://inpainter.vercel.app/paint >The most notable non-Stable Diffusion generator is Midjourney, which tends to be nicer-looking and doesn't require as much fiddling with prompts but can have a samey style. However the only way to use it is through a Discord bot. https://www.midjourney.com/home/ >If you're okay with more setup but don't have the hardware you can use a cloud-hosted install: Paperspace: https://rentry.org/865dy Colab: https://colab.research.google.com/drive/1kw3egmSn-KgWsikYvOMjJkVDsPLjEMzl >Various other guides: NovelAi: https://rentry.org/sdg_FAQ Dreambooth: https://github.com/ShivamShrirao/diffusers/tree/main/examples/dreambooth Inpainting/Outpainting: https://rentry.org/drfar Upscaling images: https://rentry.org/sdupscale Textual inversion: https://rentry.org/textard >Resources Search images and get ideas for prompts, can search by image to see similar images: https://lexica.art/ Index of various resources: https://pharmapsychotic.com/tools.html Artist styles: https://rentry.org/artists_sd-v1-4 | https://www.urania.ai/top-sd-artists | https://rentry.org/anime_and_titties | https://sgreens.notion.site/sgreens/4ca6f4e229e24da6845b6d49e6b08ae7 | https://proximacentaurib.notion.site/e28a4f8d97724f14a784a538b8589e7d Compiled list of various models, regularly updated: https://rentry.org/sdmodels (Check here before asking "where do I find x model!") Going past the OP copypasta, there's now a new AI generator on the scene: Bing. Yes, that Bing. And it produces some of the best results so far. https://www.bing.com/images/create The tech is moving very fast and I'm not sure how comfortable I am with it.
>>2943 >Venturebeat hitpiece It's interesting to note that boomers in government and business think that the news industry has the public's best interest at heart. Whether it's down to not understanding the belly of the beast or not thinking too deeply because of regularly working, I'm unsure. >>2956 Those may be funny, but hotdogs and noodles produce funnier results.
>>2957 Anon your challenge in bing bonging has been accepted. Now with blondes. Also if you want to get some of these nice rendering results try using "in the style of virgil finley". If you're a pulp aficionado you'll know what this means. I also add the word "nightmare" for darker tones.
>>2957 Some more noodley goodness.
>>2956 >>2957 >>2959 >bubblegum fetish What the fuck?
>>2955 Thanks. Glad you like them. The first 5 runway model images are pretty much the same prompt, just with a slightly different parameter here and there, different model, or with a style lora applied. The 5th one was a pretty big surprise since it uses a Bloodborne lora, so I was expecting a very different result. >Seeing ideas just come to life like this is amazing. This stuff is pretty fun and extremely addicting. Seeing some of my favorite characters doing silly things or drawn in the style of some famous artist is really cool. It's just a shame when you get an almost perfect image that just needs upscaling or a minor fix and then run out of VRAM, but unfortunately 4090s aren't cheap. I'm done with Kat for a while, but it's almost Christmas, so Kat and Raven are in a festive mood. And don't forget to purchase the Gravity Rush® Christmas Kat™ Premium Limited Edition Action Figure for only 9,999 gems + shipping + tax + tip!
>>2960 >What the fuck? Yes.
>>2960 And some more
So I'll ask here since I'm sure that the people dumping their creations are at least checking the thread out every once in a wihile. I am trying to recreate Rayman in Dalle-3, but the AI keeps throwing random bird-like characters instead of what I want. What are your tricks to getting it to work properly?
>>2962 >>2963 >>2964 Please take your censored shit elsewhere. >>2965 It's not worth the trouble, doing all shit to dance around the filter only to get a bootleg of a character you want, just use stable Diffusion
>>2966 >Please take your censored shit elsewhere. Nah. I can't because most stable diffusion joints currently don't and don't have readily available interpreters so far for these kinda bubble gum effects. For now. And it's not like you're putting in anything in this thread so stop being an absolute cunt about it and post your work. Or give me a lora that has this flexibility for my fetish
Something I just found and may be useful to you guys https://huggingface.co/spaces/hysts/DeepDanbooru Upload a picture and get a danbooru-like description in the form of tags to use for other AI generators
(510.03 KB 568x1024 undefined_image (2).png)

(508.42 KB 568x1024 undefined_image (3).png)

(2.04 MB 1136x2048 undefined_image (1).png)

(3.34 MB 1536x2048 undefined_image (5).png)

(3.83 MB 1536x2048 undefined_image.png)

Yo Anon I heard you got a fever and the only cure for it was more fat Melona!
>>2967 I'm not sexually aroused by any of these, but some of them do feel... oddly satisfying. I hope you can make nudes one day, anon.
(791.43 KB 752x1024 undefined_image (6).png)

(341.45 KB 424x768 undefined_image (7).png)

(3.23 MB 1536x2048 undefined_image (10).png)

(864.06 KB 768x1024 undefined_image (13).png)

(787.02 KB 768x1024 undefined_image (14).png)

(759.71 KB 768x1024 undefined_image (25).png)

(910.92 KB 768x1024 undefined_image (26).png)

(1.11 MB 768x1024 undefined_image (27).png)

(1.00 MB 768x1024 undefined_image (29).png)

(654.20 KB 768x1024 undefined_image (8).png)

And now some odd Xmas'y Poison gens.
(951.81 KB 768x1024 undefined_image (20).png)

(507.62 KB 576x768 undefined_image (16).png)

(465.63 KB 576x768 undefined_image (17).png)

(781.79 KB 768x1024 undefined_image (9).png)

(447.39 KB 576x768 undefined_image (18).png)

Merry Christmas fellas and happy prompting.
(98.75 KB 454x454 aifail01_potter.jpg)

(83.59 KB 454x454 aifail02_clown.jpg)

(401.75 KB 862x3115 aifail03_coyote.jpg)

(376.02 KB 920x2721 aifail04_superfail.jpg)

(102.24 KB 454x454 aifail05_failonmars.jpg)

Gab AI had an update and is still glitchy, so it's only slightly less bad. It still doesn't give good results. Common problems: * It doesn't understand English instructions but takes elements from the prompt and installs them wherever it feels like. * It knows some characters but doesn't know others. The characters that it is aware of, it does not draw on model. It draws something close to them but something is always off. * When there are multiple characters in a scene, it bleeds elements from one into another. It also gets confused and duplicates characters. * It has a few default/common art styles. You tell it to do something in another art style and it tries to fit it into one of those art styles instead. I've tried to tell it to do Warner Brothers cartoon style, newspaper comic strip, Gary Larson / Far Side, My Little Pony, it doesn't seem to understand. It is vaguely aware of Peanuts and will make an attempt. * It does not seem possible to train the AI, so you can only hope that it understands you. It's still interesting to see how it fucks up and wonder about how that happens, so let's show off some AI art failures. AI Failure #1: I asked for Harry Potter in the style of Ralph Bakshi. It gave me a pretty good Harry Potter image that looks nothing like a Ralph Bakshi cartoon, so it does not know Bakshi's style (Fritz the Cat, the Lord of the Rings cartoon, Wizards). AI Failure #2: "Clownpiece from Touhou dancing in a grassy field". It does not know Clownpiece as a character, but can draw a generic Touhou in generic clown colors. And the most famous clown is The Joker. So it created The Joker as a 2hu. AI Failure #3: I wanted to see Wile E. Coyote holding a giant magnet with Iron Man stuck to it. The AI does not know what a giant cartoon magnet is supposed to look like, how someone would hold it, or how someone would be stuck to it. The AI also seems to have its own idea of what Wile E. Coyote should look like, repeating a similar character who I do not recognize as Wile E. Coyote. And for a "reroll" it gave me a really awesome image of a wolf for some reason. AI Failure #4: The concept is a simple sight gag, Superman mixes his reds and blues in the wash and they come out purple. The AI kept drawing Clark Kent wearing Superman's uniform, and made everything in the image purple except for Superman's uniform. That kept coming out in its perfect blue. The AI refused to recolor it, I'm guessing because it knows what Superman's uniform is supposed to look like and resists changing it. AI Failure #5: This is supposed to be Commander Keen fighting Marvin the Martian. The AI has no idea what Marvin is supposed to look like and only a vague idea of Commander Keen, and it gave Keen a pair of Martian antennae. It did draw a decent fantasy Martian background.
(103.20 KB 454x454 aifail06_oystory.jpg)

(80.38 KB 454x454 aifail07_tenshi.jpg)

(68.84 KB 454x454 aifail08_froggyfreeze.jpg)

(219.11 KB 1361x454 aifail09_shanghai.jpg)

(316.54 KB 925x1974 aifail010_dolls.jpg)

>>2979 AI Failure #6: "Alice Margatroid from Touhou plays with Woody from Toy Story and Buzz Lightyear from Toy Story". Instead we get Woody's head on Buzz's body, Buzz's chestplate on Woody's body, another Buzz, and a pair of mutant deformed Buzzes in the background. And Alice became a doll. If you play with dolls too much... AI Failure #7: "Tenshi Hinanawi eating a corndog on Route 66." The AI does not know who Tenshi is and cannot draw her, so it gave me some Japanese woman instead. And it seriously fucked by adding that giant corndog on a misaligned stick that she isn't even holding. However, the woman's design and posture give off the right impression of being a tourist enjoying a corndog. That's a pretty good result even if her hand is messed up. AI Failure #8: "Cirno freezing Pepe the Frog". The AI produced their bastard daughter instead. AI Failure #9: "Shanghai Doll from Touhou wielding a fencing rapier", "Vivian James as a Touhou character", and "Rydia from Final Fantasy IV as a Touhou character." The AI can generate good looking characters in the Touhou style even if they do not resemble the character that was requested. Rydia has the right color scheme but I don't recognize her as the character. AI Failure #10: My plan was to put a bunch of evil dolls together for a Toy Story parody. There is no way that Gab AI can produce this, so I gave up. The first mistake is my own, asking for Chatty Cathy from the Twilight Zone Episode The Living Doll. That doll's name is Talky Tina. Chatty Cathy is the doll that she was based on. The doll that the AI gave me is neither of them, but is a good generic doll character. Next I asked for her and Chucky in the same image. The AI knows Chucky, but also bled elements from Chucky into the other doll. And vice versa: the AI put Chucky in a skirt because the other doll was wearing one. That is one of the ways the AI consistently fails when drawing multiple characters or objects. It mixes their aspects together.
(373.67 KB 910x911 aifail011_vashandreimu.jpg)

(87.54 KB 454x454 aifail012_remilia.jpg)

(67.04 KB 454x454 aifail013_nazrin.jpg)

(66.19 KB 454x454 aifail014_farsidecow.jpg)

(92.65 KB 454x454 aifail015_haddock.jpg)

AI Failure #11: I asked for Trigun in the styles of Peanuts and Tintin, and Reimu in the styles of Peanuts and "Marvel comic book". The AI recognizes "Trigun" as Vash the Stampede and draws that character, but the art styles are not what I had requested. The AI does a better job style-wise with Reimu. It does not know what Reimu is supposed to be holding so it gave her a mushroom / tiny parasol thing. Other than that they both look pretty good. AI Failure #12: "Remilia Scarlet shows a road map to a confused Simon Belmont". It's interesting to see what the AI did with Remilia and Simon, but the image lacks the elements that would have made the concept funny. AI Failure #13: "Nazrin sitting on the moon eating cheese". The AI figured out that Nazrin is a mouse and made her a furry. I am impressed. The AI goofed by putting another moon on the moon and a moon in the background. Is that enough moon for you? AI Failure #14: "A cow from The Far Side in a My Little Pony cartoon". The AI made a cute cow in a 3-D artstyle. AI Failure #15: Captain Haddock from Tintin throwing Monkey Luffy from One Piece off of his boat. Except, I misspelled Monkey Luffy's name. The AI drew a decent Captain Haddock but it kept drawing Monkey "Luffuy" as a literal monkey dressed like The Man in the Yellow Hat from Curious George. And it kept drawing the boats and the characters floating in the air over the sea.
(90.69 KB 454x454 vegeta.jpg)

(79.14 KB 454x454 hellokitty.jpg)

(67.16 KB 454x454 tonbeery.jpg)

(68.40 KB 454x454 chocobo.jpg)

Some flawed but half decent results: Vegeta eating a fajita, Hello Kitty in Hell, "a tonberry from final fantasy holding a can of beer", "A chocolate in the shape of a chocobo from Final Fantasy, in a partially opened foil wrapper", and The Centennial Festival for Magical Girls.
(107.44 KB 454x454 centennial.jpg)

(75.33 KB 454x454 iscream.jpg)

>>2982 Centennial Festival was stripped on upload. Are webms not allowed? Did a copyright filter eat the song? Here's the jpg. And enjoy some I Scream.
About time to move the thread to >>>/ais/?
>>2984 Agreed. If you wish to, make a new AI thread too, but this time include both visuals and audio in the same OP, since the audio world is kind of very slow moving compared to the leaps and bounds in picture generation.
(381.37 KB 454x454 space_battleship_tomato.png)

Guess the anime
(2.42 MB 1920x1080 00035.png)

(2.21 MB 1920x1080 00037.png)

(1.61 MB 1280x960 00060-2069474834.png)

New Year, time to hit the gas pedal with diffusion.
>>2988 Prompts? Just asking for a friend haha
(83.08 KB 721x891 CHAOS.png)

>>2908 It has, and it's glorious.
(529.80 KB 576x1024 undefined_image (3).png)

(741.04 KB 576x1024 undefined_image (14).png)

(630.50 KB 576x1024 undefined_image (12).png)

(960.62 KB 768x1024 undefined_image (13).png)

(828.05 KB 768x1024 undefined_image (11).png)

Also futzing around with some Poison Laras with a few muscle and inflations ones to see what comes out of the experiments.
(2.77 MB 1360x2048 undefined_image (2).png)

(598.29 KB 576x1024 undefined_image (4).png)

(931.63 KB 768x1024 undefined_image (8).png)

(567.64 KB 768x1024 undefined_image (9).png)

(1.05 MB 768x1024 undefined_image.png)

>>2991 Thiccer, Plumper, and gassier than ever before?


Forms
Delete
Report
Quick Reply